Note: Descriptions are shown in the official language in which they were submitted.
CA 02689035 2009-12-23
1
SYSTEMS AND METHODS FOR PREDICTIVE BUILDING ENERGY
MONITORING
1. TECHNICAL FIELD
The present invention relates generally to systems and methods for building
energy
management. More specifically, the present inventive systems and methods
relate to
building energy monitoring using predicted building energy use values.
2. BACKGROUND OF THE INVENTION
As the desire to improve energy efficiency has become more widespread, a need
for
effective methods and infrastructure to monitor and control the consumption of
energy by
buildings of all sizes has become of increasing importance, and for large
buildings in
particular. As a result, building management systems have become increasingly
common.
Building management systems typically provide for the measurement and control
of
building energy systems, including the measurement of consumption of various
energy-
related resources by building systems including for example electrical energy,
natural gas,
thermal energy, steam energy, or water usage, through the use of electronic
and/or
computerized building energy meters. Such building energy meters then provide
building
energy consumption data to an automated or computer controlled building
management
system.
In addition to the adoption of automated building management systems to
measure
and control building systems and building energy consumption, in order to
improve
building energy efficiency, it has also become of interest to provide
predicted or expected
typical values for building energy loads for a particular building, in order
to enable more
accurate and efficient planning, management, and fault detection of energy
consumption.
To that end, multiple approaches have been made towards the determination of
predicted or
typical building energy load values for a building, which have been adapted to
attempt to
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
2
take into account the effect of various parameters that may influence building
energy
consumption, and therefore the building energy load values that are predicted.
Certain of the known approaches to determining predicted or typical values for
building energy load have been based on the use of detailed physical building
models to
calculate the predicted or typical building energy load values for given
environmental or
predictor variables such as temperature, humidity, wind, occupancy, etc. Such
detailed
physical building models may typically model heating and cooling behavior and
associated
energy consumption for a building by constructing energy consumption models
for the
building including consideration of many building-specific design parameters,
such as
dimensions, materials, orientation, window coverage, insulation, HVAC systems,
and
lighting, for example, in addition to consideration of the effect of external
environmental
variables such as temperature, humidity and wind, for example. However, such
physical
building models may be very complex, and as a result typically require
extensive building-
specific evaluation and analysis of design factors by skilled professionals to
produce
accurate models. Therefore, the use of such physical building model methods to
generate
predicted or typical building energy load values may be impractical for use on
multiple
dissimilar buildings, due to the computational and economic cost of developing
such
models for each building. This approach may also be unsuitable for use in
connection with
buildings for which detailed design data is not available or impractical to
obtain.
Other known approaches to determining predicted or typical values for building
energy loads have been based on numerical modelling tools that attempt to
relate
previously measured building energy load values and associated variations in
external
environmental variables such as temperature, humidity and wind, to provide
predicted or
typical building energy load values corresponding to observed or predicted
environmental
variables. One such approach makes use of linear or higher order polynomial
functions that
attempt to fit measured historical building energy load values to measured
values of
environmental variables, after which the resulting polynomial model may be
applied to
environmental variable values for the predicted conditions, to obtain a
predicted or typical
value for building energy loads. However, models based on such polynomial
functions
may not provide a good fit to building energy load data, in particular low
order polynomials
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
3
may be unable to follow step changes and higher order polynomials may overfit
to noise.
Also, the use of polynomial functions may result in poor estimated or
predicted values for
variables outside the range of training datasets.
Other known numerical modelling approaches are based on the use of neural
networks to fit historical building energy load and predictor variable data
for a building, to
enable prediction of building energy load values for given predictor variable
value inputs.
However, the use of neural network techniques may require complex computation
of many
parameters in order to fit or learn from a historical building energy load
dataset. This
approach may further be prone to unstable or erroneous prediction behavior
when used to
produce predicted building energy load values corresponding to predictor
variable values
which are outside the range of the historical dataset used to fit or train the
neural network
system. Also, the use of neural network techniques may typically involve the
use of a large
number of parameters relative to their expressive power, which may contribute
to
estimation or prediction errors related to overfitting in some applications,
depending upon
the size and/or range of the historical building energy load dataset available
to use in
training or initializing of a neural network model.
Accordingly, a need exists for improved methods and systems for building
energy
monitoring such as may be desirably adaptable for use with multiple buildings
and without
the requirement of extensive computational resources.
3. SUMMARY OF THE INVENTION
It is an object of the present invention to provide a system and method for
predictive
building energy monitoring.
According to another object of the present invention, a system and method for
generating an alert based on predictive building energy monitoring is
provided.
According to yet a further object of the invention, a computer readable medium
and
computer implemented method for predicting building energy load values is
provided.
A computer readable medium according to an embodiment of the present invention
is provided, comprising executable instructions to receive historical building
energy load
DM_VAN/253729-15331 /7496905 1
CA 02689035 2009-12-23
4
values and corresponding historical predictor variable data comprising values
for one or
more predictor variables; define a scaling factor for each said predictor
variable by cross-
validation error minimization for said historical building energy load values
using said
corresponding historical predictor variable values scaled by said scaling
factor values;
receive measured building energy load values measured by at least one building
energy
meter; receive measured predictor variable values corresponding to the
measured building
energy load values from a predictor variable datasource; determine a predicted
building
energy load value by kernel smoothing of one or more nearest neighbor
historical building
energy load values using said defined scaling factors for scaling each said
predictor
variable; compare said predicted building energy load value with said measured
building
energy load value to determine if an alert threshold level is exceeded; and to
transmit an
alert signal to a user when said alert threshold level is exceeded.
A computer implemented method of building energy monitoring is provided
according to another embodiment of the invention, the method comprising
receiving
historical building energy load values and corresponding historical predictor
variable data
comprising values for one or more predictor variables; defining a scaling
factor for each
predictor variable by cross-validation error minimization for the historical
building energy
load values using corresponding historical predictor variable values scaled by
said scaling
factors; receiving measured building energy load values measured by at least
one building
energy meter; receiving measured predictor variable values corresponding to
the measured
building energy load values from a predictor variable datasource; determining
a predicted
building energy load value by kernel smoothing using one or more nearest
neighbor
historical building energy load values and using the defined scaling factors
for scaling each
predictor variable; comparing the predicted building energy load value with
the measured
building energy load value to determine if an alert threshold level is
exceeded; and
transmitting an alert signal to a user when the alert threshold level is
exceeded.
According to a further embodiment of the invention, a computer implemented
method of building energy monitoring is provided comprising: receiving
historical building
energy load values and corresponding historical predictor variable data
comprising values
for one or more predictor variables; receiving scaling factor values for each
said predictor
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
variable; receiving measured building energy load values measured by at least
one building
energy meter; receiving measured predictor variable values corresponding to
the measured
building energy load values from a predictor variable datasource; determining
a predicted
building energy load value by kernel smoothing using one or more nearest
neighbor
5 historical building energy load values and using said scaling factor
values for scaling each
said predictor variable; comparing said predicted building energy load value
with said
measured building energy load value to determine if an alert threshold level
is exceeded;
and transmitting an alert signal to a user when said alert threshold level is
exceeded.
Further embodiments of the invention will become apparent when considering the
drawings in conjunction with the detailed description.
4. BRIEF DESCRIPTION OF THE DRAWINGS
The system and method of the present invention will now be described with
reference to the accompanying drawing figures, in which:
FIG. 1 illustrates an exemplary networked operating environment for
implementing
an embodiment of the present invention.
FIG. 2 illustrates an exemplary user computer architecture configured
according to
an embodiment of the invention.
FIG. 3 illustrates an exemplary monitoring server computer architecture
configured
according to an embodiment of the invention.
FIG. 4 illustrates an exemplary building datasource server computer
architecture
configured according to an embodiment of the invention.
FIG. 5 illustrates a series of processing operations associated with an
embodiment
of the invention.
FIG. 6 illustrates an exemplary graphical time series of building energy load
and
predictor variable values associated with an embodiment of the invention.
FIG. 7 illustrates an exemplary graphical representation of a scaling factor
learning
process in association with an embodiment of the present invention.
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
6
Like reference numerals refer to corresponding parts throughout the several
views
of the drawings. The sizes and relative positions of elements in the drawings
are not
necessarily drawn to scale. Further, the particular shapes of the elements as
drawn are not
intended to convey any information regarding the actual shape or form of the
elements, and
have been solely selected for ease of recognition in the drawings.
5. DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 illustrates an exemplary networked operating environment in which an
embodiment of the present invention may be implemented. The networked
environment
includes a user computer 10 connected to a communication network 50, which may
include
one or more of: a local area network, intranet, wide area network, world wide
web
(WWW), or the global Internet, for example, such that user computer 10 may
communicate
with other computers similarly connected to network 50. Other computers
connected to
network 50 may include a building monitoring server 20, building datasource
server 30
connected to building energy meter 32, and a predictor variable datasource
server 40, which
may each communicate with any other computer connected to the network 50.
Optionally,
building datasource server 30 may connect directly to building monitoring
server 20, such
as by connection 35, for example. Additionally, in a further optional
embodiment, building
energy meter 32 may connect directly to building monitoring server such as by
connection
35, or may connect to building datasource server 30 and thereby to either
computer network
50 or to building monitoring server 20 by connection 35, for example. User
computer 10
includes standard computing components for transmitting and receiving data to
and from
other computers connected to the user computer 10 through network 50,
including
receiving and transmitting webpage and/or web application data by commonly
known
methods and techniques such as hypertext markup language ("HTML"), extensible
hypertext markup language ("XHTML"), extensible markup language ("XML"),
wireless
markup language ("WML") or other available data formats which may incorporate
structural definitions into a data file, for example.
DMyAN/253729-15331/7496905.1
CA 02689035 2009-12-23
7
Building monitoring server 20 includes standard computing components for
transmitting and/or receiving data, including building monitoring data,
webpages, web
application data such as HTML, XHTML, XML, and/or WML and the like, to and
from
other computers connected over the computer network 50 or other connection 35,
including
user computer 10, building datasource 30 and predictor variable datasource 40,
and
optionally also building energy meter 32. In a particular embodiment, building
monitoring
server 20 includes or is connected to at least one monitoring data storage
repository 22 for
storing building monitoring data which may be transmitted or received to or
from other
computers or devices which may be connected to building monitoring server 20.
Storage
repository 22 may comprise a conventional data storage device such as a hard
disk or solid-
state memory located with and connected directly to building monitoring server
20, or may
alternatively comprise a remote data storage repository connected to building
monitoring
server 20, such as a network storage appliance, for example. Monitoring data
storage
repository 22 may comprise stored building energy load data 24, and stored
predictor
variable data 26, for example. In one embodiment, stored building energy load
data 24 may
comprise one or more of: historical measured building energy load values,
current
measured building energy load values, and predicted building energy load
values, for
example. Historical and current measured building energy load values may be
received by
monitoring server 20 from one or more of: a building datasource server 30, a
building
energy meter 32, and/or another suitable source of measured building energy
load values,
for example, and may be received by building monitoring server 20 such as
through
computer network 50 and/or connection 35. Predicted building energy load
values
comprised in building energy load data 24 may be calculated such as by
monitoring server
20, for example. In another embodiment, stored predictor variable data 26 may
comprise
one or more of: historical predictor variable values, and current predictor
variable values.
Such predictor variable values may be received by monitoring server from
predictor
variable datasource server 40 and/or another suitable source of predictor
variable data, and
may be received by building monitoring server 20 such as through computer
network 50 or
connection 35, for example.
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
8
Building datasource computer 30 includes standard computing components for
sending and receiving building energy data to and from other computers
connected to
network 50. In one embodiment, building datasource 30 is connected to at least
one
building energy meter 32 which provides measured building energy load data for
one or
more buildings to building datasource 30. Building datasource 30 may also
comprise a
building energy data storage repository (not shown) for storing measured
building energy
load data, which may comprise a conventional data storage device such as a
hard disk or
solid-state memory located with and connected directly to building datasource
30, or may
comprise a remote data storage facility connected building datasource 30. Such
stored
building energy load data may comprise at least one of past historical
building energy load
values, and current measured building energy load values. In one embodiment,
building
datasource 30 may comprise a building management or building automation
system, for
example, and may be connected to one or more building energy meters 32
corresponding to
one or more buildings. In another embodiment, building datasource 30 may
comprise a
computer or server connected to and/or integrated with a smart building energy
meter 32.
Alternatively, building datasource 30 may comprise a utility company computer
or system,
which may be connected to and/or communicate with one or more (typically many)
smart
building energy meters 32, to provide measured building energy load data
corresponding to
one or more buildings, for example.
Predictor variable datasource computer 40 includes standard computing
components
for sending and receiving predictor variable data comprising values for one or
more
predictor variables to and from other computers connected to network 50. In
one
embodiment, predictor variable datasource 40 is connected to at least one
predictor variable
data storage repository (not shown) for storing predictor variable value data.
Such
predictor variable data storage repository may comprise any suitable data
storage device
such as a hard disk or solid-state memory located with and connected directly
to predictor
variable datasource 40, or may comprise a remote data storage facility
connected to
predictor variable datasource 40. Predictor variable datasource 40 may store
predictor
variable data comprising past historical predictor variable values, as well as
current
measured predictor variable values, for at least one geographical location
corresponding to
DM_VAN/253729-15331 /7496905 I
CA 02689035 2009-12-23
9
at least one building, such as for a city, town, or region in which one or
more buildings of
interest are located, for example. In one embodiment, predictor variable
datasource 40 may
comprise a weather data server for a public weather service, such as a
government and/or
private weather service which may make historical and current predictor
variable values for
multiple geographical locations available to the public, such as over network
50, for
example. In another embodiment, predictor variable datasource 40 may comprise
a
computer server for a private or building-specific weather station, and may be
located on
and/or adjacent to a building of interest, for example, to measure and store
predictor
variable values from such weather station. In another embodiment, predictor
variable
datasource 40 may also comprise a computer server for a building specific
occupant
monitoring system, and may be located on and/or adjacent to a building of
interest, for
example, to measure and store occupancy variable values from such building. In
another
embodiment, predictor variable datasource 40 may comprise of any suitable
combination of
data sources providing predictor variable data.
FIG. 2 illustrates an exemplary computer architecture for a user computer 10
configured in accordance with an embodiment of the invention. User computer 10
includes
standard components, including a central processing unit 102 and input/output
devices 104,
which are linked by a bus 108. Input/output devices 104 may comprise a
keyboard, mouse,
touch screen, monitor, printer, and the like, for example. Network interface
106 is also
connected to the bus 108. Network interface 106 provides connectivity to a
network 50,
such as the exemplary computer network 50 described above, thereby allowing
the
computer 10 to operate in a networked environment. Also connected to the bus
108 is a
computer-readable memory 110. The memory 110 stores executable instructions to
implement functions of the present invention. The computer-readable memory 110
may
comprise any available computer-readable media or device that can be accessed
by
computer 10.
In an embodiment of the present invention, one or more of the following
program
modules and data files may be stored in the memory 110 of the computer 10: an
operating
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
system module 112, a web browser module 114 and a monitoring server access
module
116.
Operating system module 112 may be suitable for controlling the operation of a
networked user or analysis computer, and in particular may comprise
instructions for
5 handling various system services, such as file services or for performing
hardware
dependant tasks. Operating system module 112 may also comprise instructions
for
standard computer operation, including receiving input from input devices such
as a
keyboard or mouse, and for displaying output in a graphical format on a
monitor, for
example.
10 Operating system module 112 may comprise any known executable operating
system instructions, such as may be suitable for controlling the general
operations of
networked user computer 10, and in particular may comprise instructions for
handling
various system services, such as file services or for performing hardware
dependent tasks.
Operating system module 112 may also comprise instructions for standard
computer
operation, including receiving input from input devices such as a keyboard or
mouse, and
for displaying output in a graphical format on a monitor, for example. In one
embodiment,
operating system module 112 may comprise one or more known proprietary or open-
source
computer operating systems, such as for example, Windows (TM), MacOS (TM),
UNIX
(TM) or Linux (TM) operating systems.
Web browser module 114 comprises instructions for browsing webpages and/or
viewing web applications provided by a web server or other source, such as
instructions for
requesting and receiving a webpage and/or web application from a monitoring
server and
displaying the webpage and/or web application on a display device such as a
monitor. Web
browser module 114 also comprises instructions for receiving input from a
user's
interaction with a webpage and/or web application, such as from input devices
like a
keyboard and mouse for example, and for transmitting such user input to a web
server.
Web browser module 114 may also comprise instructions for executing processing
commands comprised in webpages and/or web applications.
Monitoring server access module 116 comprises instructions for authenticating
and/or verifying a user's identity to building monitoring server 20, and to
receive and
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
11
display one or more web application pages transmitted by building monitoring
server 20 as
part of a building energy monitoring web application, for example. Monitoring
server
access module 116 may also comprise instructions to transmit input from a user
operating
user computer 10, and/or building energy management data input on user
computer 10 back
to monitoring server 20, for example. In one embodiment, monitoring server
access
module 116 may further comprise instructions for displaying multiple building
energy
monitoring interface displays received from monitoring server 20, such as may
be desirable
to display different aspects of building energy data, for example. In a
particular
embodiment, monitoring server access module 116 may also comprise instructions
for
receiving one or more alert messages from monitoring server 20, and for
displaying and/or
otherwise transmitting such alert messages to a user of user computer 10 to,
for example,
notify a user regarding a building energy monitoring alert condition.
The above-described program modules incorporate instructions to implement
processing operations associated with the present invention. Various
embodiments of the
processing operations of the above-described program modules are described
below with
reference to FIG. 5. The modules stored in memory 110 are exemplary, and
additional
modules can be included. It should be appreciated that the functions of the
presented
modules may be combined. In addition, a function of a module need not be
performed on a
single machine. Instead, the function may be distributed across a network such
as network
50, to one or more other computers or devices if desired. It is the functions
of the invention
that are significant, not where they are performed or the specific manner in
which they are
performed.
FIG. 3 illustrates an exemplary computer architecture for a building
monitoring
server computer, such as monitoring server 20 as illustrated in the computer
system of FIG.
1, configured in accordance with an embodiment of the present invention.
Monitoring
server computer 20 includes standard components, including a central
processing unit 202
and input/output devices 204, which are linked by a bus 208. Input/output
devices 204 may
comprise a keyboard, mouse, touch screen, monitor, printer, and the like, for
example.
Network interface 206 is also connected to the bus 208. Network interface 206
provides
connectivity to a network 50, such as the exemplary computer network 50
described above,
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
12
thereby allowing the monitoring server computer 20 to operate in a networked
environment. Monitoring server 20 also comprises a monitoring data storage
repository 22
capable of storing one or more of: building energy load data and/or predictor
variable data,
for example. As described above, monitoring data storage repository 22 may
comprise a
conventional data storage device such as a hard disk or solid-state memory
located with and
connected directly to monitoring server 20 such as by bus 208 as shown in FIG.
3, or may
alternately comprise a remote data storage facility accessibly connected to
monitoring
server 20. Connected to the bus 208 is a computer-readable memory 210. The
memory 210
stores executable instructions to implement functions of the present
invention. Computer-
readable memory 210 may comprise any available computer-readable media or
device that
can be accessed by the monitoring server computer 20.
In an embodiment of the invention, one or more of the following program
modules
and data files may be stored in memory 210 of monitoring server computer 20:
an operating
system module 212, a monitoring server module 214, a scaling factor learning
module 216,
a prediction module 218 and an alert module 220.
Similar to module 112 described above, operating system module 212 may
comprise any known executable operating system instructions, such as may be
suitable for
controlling the general operations of networked monitoring server 20, and in
particular may
comprise instructions for handling various system services, such as file
services or for
performing hardware dependent tasks. Operating system module 212 may also
comprise
instructions for standard computer operation, including receiving input from
input devices
such as a keyboard or mouse, and for displaying output in a graphical format
on a monitor,
for example. In one embodiment, operating system module 212 may comprise one
or more
known proprietary or open-source computer operating systems, such as Windows
(TM),
MacOS (TM), UNIX (TM) or Linux (TM) operating systems. In a particular
embodiment,
operating system module 212 may comprise a Linux (TM) operating system, for
example.
Monitoring server module 214 comprises instructions for transmitting and
receiving
building energy data to and from other computers connected to monitoring
server 20 such
as by computer network 50 and/or other data connections such as connection 35,
for
example. In particular, monitoring server module 214 comprises instructions
for receiving
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
13
historical and current measured building energy load values corresponding to
one or more
buildings, such as from one or more building datasources 30 and/or one or more
building
energy meters 32, for example. Monitoring server module 214 also comprises
instructions
for receiving historical and current predictor variable values corresponding
to one or more
geographical locations related to one or more buildings, such as from one or
more predictor
variable datasource 40, for example.
Monitoring server module 214 also comprises instructions for transmitting and
receiving building energy to and from user computer 10 over computer network
50. In
particular, monitoring server module 214 may comprise instructions to provide
webpages
and/or web application data to user computer 10 including at least one of
measured building
energy load values and predicted building energy load values, for example. In
one
embodiment, monitoring server module 214 may comprise instructions for
providing web
server functionality such as to interactively transmit building energy
management data to
user computer 10 in the form of one or more web application interface displays
to compare
measured and predicted building energy load values for one or more buildings.
Optionally
monitoring server 214 may also receive building energy data from user computer
10 such
as may be input by a user of computer 10, for example. In a particular
embodiment,
monitoring server module 214 may also comprise instructions to transmit one or
more
building energy alert messages to user computer 10 such as alert messages in
response to a
comparison of measured building energy load values with predicted building
energy load
values, for example. In one embodiment, monitoring server module 214 may
additionally
optionally comprise instructions to transmit such alert messages to other
computers and/or
devices over a communication network such as exemplary network 50, in addition
to or in
place of user computer 10, for example.
In another embodiment, monitoring server module 214 may additionally comprise
instructions for storing at least one of: historical building energy load
values, historical
predictor variable values, current measured building energy load values,
predicted building
energy load values and current predictor variable values. Monitoring server
module 214
may store such instructions in monitoring data storage repository 22. In one
embodiment,
monitoring server module 214 may comprise instructions to store historical
and/or current
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
14
building energy load values and/or predictor variable values in monitoring
data storage
repository 22 as such values are received by monitoring server 20, such as
from other
computers over network 50, for example. Monitoring server module 214 may also
comprise instructions to store predicted building energy load values in
monitoring data
storage repository 22 as they are calculated by monitoring server 20, or
alternatively as they
are transmitted to a user computer such as part of a web application page, for
example.
Scaling factor learning module 216 comprises instructions for learning or
determining a scaling factor value for each of multiple predictor variables,
to optimize
prediction of building energy load values by cross-validation error
minimization of
historical building energy load values using historical predictor variable
values received by
monitoring server 20. In particular, scaling factor learning module 216
comprises
instructions for determining a scaling factor value for each of multiple
predictor variables
by cross validation to iteratively minimize the difference between predicted
and measured
historical building energy load values, and for determining predicted values
for said
historical building energy loads calculated by kernel smoothing nearest
neighbor values of
historical building energy loads using corresponding historical predictor
variable values
scaled or weighted according to the learned scaling factor values. Exemplary
predictor
variables may comprise one or more of: time, day, date, temperature, humidity,
wind speed,
wind direction, solar radiation and occupancy. In a particular embodiment,
scaling factor
learning module 216 may comprise instructions to define a training data set of
historical
measured building energy load values and corresponding historical predictor
variable
values as well as a validation data set of historical measured building energy
load values
and corresponding historical predictor variable values. Scaling factor
learning module 216
may also comprise instructions to determine a scaling factor value for each of
multiple
predictor variables by calculating and employing a root mean squared (rms)
error function
defining the error between measured historical building energy loads in the
validation data
set and predicted building energy load values calculated by kernel smoothing
of historical
building energy load values within the training data set using historical
predictor variable
values that are weighted or scaled according to the scaling factor values. The
rms error
function may be minimized by using multiple successive iterations to vary the
scaling
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
factor values. For example, the rms error function for the difference between
historical
measured building energy load values and corresponding calculated predicted
building
energy load values may be expressed by:
( 1
5 e= ____
Ns
\ les
where S is the validation data set of historical building energy load values,
yi is the
historical measured building energy load value, and Sfi is the corresponding
predicted
building energy load value.
In such case, monitoring server module 214 may comprise instructions to
optimize
10 the learned scaling factor values by iteratively minimizing the rms
error function using the
Levenberg-Marquardt optimization algorithm for varying the learned scaling
factor values
for subsequent iterations of the scaling factor learning process.
Alternatively, other suitable
known optimization algorithms may be implemented by scaling factor learning
module 216
for iteratively optimizing learned scaling factor values.
15 In one embodiment, scaling factor learning module 216 also comprises
instructions
for calculating predicted historical building energy load values by kernel
smoothing of
historical measured building energy load values using multiple historical
predictor variable
values that are scaled by the scaling factor values. A unit Gaussian kernel
function may be
used in such case. The predicted historical building energy load values may
then be
calculated using a weighted average of historical measured building energy
load values.
For example, a predicted building energy load value may be calculated
according to the
equation:
k(x' x)
where Sr is the calculated predicted building energy load value for the
corresponding values
of predictor variables in vector x, y, are the measured historical building
energy load
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
16
values for the arbitrarily selected number of nearest neighbors with predictor
variable
vectors xi that have been scaled by the scaling factors, and k(x,x,) is the
unit kernel
function.
In such embodiment, the scaling factor learning module 216 may also comprise
instructions for using a suitable known k-dimensional (or k-d) tree such as a
"best-bin first"
k-d tree algorithm, where the k dimensions are those of the predictor
variables, to determine
an arbitrary number of nearest neighbor historical building energy load values
and the
distances to those values from the predictor variables associated with the
prediction in the
space of the scaled predictor variables.
In another embodiment, scaling factor learning module 216 comprises
instructions
for periodically redetermining a training data set and a validation data set
from historical
measured building energy load values and historical predictor variable values
received by
monitoring server 20 for one or more buildings. In such an embodiment,
following such
periodic redetermination of a training data set and validation data set,
scaling factor
learning module 216 also comprises instructions for relearning or determining
scaling
factor values for each predictor variable using the redetermined training data
set and
validation data set, to periodically update the scaling factor learning
process and thereby
provide updated scaling factor values for each predictor variable for each of
one or more
buildings, for example. In a particular embodiment, the scaling factor
learning module 216
may comprise instructions to receive a determination of the training data set
and validation
data set of historical measured building energy load values and predictor
variable values
from user computer 10, such as may be input by a user of computer 10, for
example.
Prediction module 218 comprises instructions for calculating a current
predicted
building energy load value (also known as a typical building energy load
value) which
corresponds to current measured predictor variable values received from
predictor variable
datasource 40, and using historical building energy load and predictor
variable values. In
particular, prediction module 218 comprises instructions to calculate current
predicted
building energy load values by kernel smoothing of nearest neighbor historical
building
energy load using historical predictor variable values scaled by the learned
scaling factor
values. Prediction module 218 may comprise instructions to calculate current
predicted (or
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
17
typical) building energy load values for one or more buildings for which
historical building
energy load and predictor variable values are stored by monitoring server 20.
Prediction
module 218 may also comprise instructions for periodically calculating new
current
predicted building energy load values for one or more buildings, such as to
provide
regularly updated predicted (or typical) building energy load values, such as
for ongoing
regular monitoring of building energy consumption of one or more buildings,
for example.
In one embodiment, similar to as described above in reference to scaling
factor
learning module 216, prediction module 218 comprises instructions for
calculating a
predicted or typical building energy load value according to the equation:
k(x,x,)y,
Y = ___________________
Lik(x,x,)
where Sr is the calculated predicted building energy load value for the
corresponding values
of predictor variables in vector x, y, are the measured historical building
energy load
values for the arbitrarily selected number of nearest neighbors with predictor
variable
vectors x, that have been scaled by the scaling factors, and k(x,x1) is the
unit kernel
function. In such an embodiment, the kernel function may comprise a unit
Gaussian kernel.
Also in such an embodiment, the prediction module 218 may further comprise
instructions
for using a suitable known k-dimensional (or k-d) tree such as a "best-bin
first" k-d tree
algorithm to determine an arbitrary number of nearest neighbor values of
historical energy
load values and nearest neighbour distances used in the kernel smoothing
calculation of the
current predicted or typical building energy load value which corresponds to
the current
measured predictor variable values for a building.
Prediction module 218 may also comprise instructions to calculate a series of
predicted or typical building energy load values for one or more buildings,
spaced over a
period of time, and to relate the series of predicted values for each building
to form a
predicted building energy load curve, also referred to as a typical building
energy load
curve. The prediction module 218 may also comprise instructions to correlate
such typical
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
18
building energy load curve to a series of measured building energy load values
such as may
be received from building datasource 30, and may comprise measured energy load
values
for one or more buildings over a period of time. Accordingly, prediction
module 218 may
thereby include instructions to provide a typical building energy load curve
comprising
calculated predicted values using kernel smoothing of nearest neighbor
historical building
energy load values, and a corresponding actual building energy load curve
comprising
current measured building energy load values, such as for comparison of such
typical curve
and actual curve values. In one embodiment, prediction module 218 may comprise
instructions to provide such typical and actual building energy load curves to
a user
computer 10, such as by using monitoring server module 214 to provide the
typical and
actual curves to a user computer 10 in the form of a webpage and/or a web
application
interface, for example.
Alert module 220 comprises instructions to compare predicted or typical
building
energy load values against measured or actual building energy load values to
determine the
difference between predicted and measured building energy load or consumption
for one or
more buildings. Alert module 220 also comprises instructions to determine if
the difference
between predicted and measured building energy load values exceeds a
predetermined alert
threshold. If the alert threshold is exceeded, alert module 220 further
comprises
instructions to generate an alert signal or message to be transmitted to one
or more users to
alert them regarding the energy load status of one or more buildings. In one
embodiment,
alert module 220 may comprise instructions to transmit an alert signal and/or
message to a
user computer 10, such as by using monitoring server module 214 to provide the
alert
signal and/or message to user computer 10 in the form of a webpage and/or a
web
application interface, for example. In another embodiment, alert module 220
may comprise
instructions to transmit an alert signal and/or message to one or more users
using a
telecommunications network such as by transmitting a signal and/or message
through an
SMS message to a mobile device, by transmitting and email message, by paging a
pager
device, by placing a telephone call, or other suitable known means for
transmitting an alert
signal and/or message to a user. In one embodiment, the alert signal may
comprise one or
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
19
more of: of: an electrical impulse, an audible sound, a mechanical force, a
movement of an
object, and a visible display, for example.
FIG. 4 illustrates an exemplary computer architecture for a building
datasource
server computer 30, such as illustrated in the computer system of FIG. 1,
configured in
accordance with an embodiment of the present invention. Building datasource
computer 30
includes standard components, including a central processing unit 302 and
input/output
devices 304, which are linked by a bus 308. Input/output devices 304 may
comprise a
keyboard, mouse, touch screen, monitor, printer, and the like, for example. A
network
interface 306 is also connected to the bus 308. Network interface 306 provides
connectivity to a network 50, such as exemplary network 50 described above,
thereby
allowing building datasource computer 30 to operate in a networked
environment.
Building datasource 30 also comprises at least one energy meter 32 which is
operable to
measure energy load or consumption of at least one building energy input for a
building
and provide a digital signal readable by building datasource server 30. In a
particular
embodiment, energy meter 32 may comprise a digital electrical meter suitable
to digitally
measure building electrical load or consumption, for example. In other
embodiments,
energy meter 32 may be suitable to measure the load or consumption of one or
more
building energy inputs selected from the list of: electricity, gas, oil, coal,
steam, water and
the like.
In an optional embodiment, building datasource 30 may additionally comprise a
building data storage repository (not shown) capable of storing building
energy load or
consumption data for one or more buildings, for example. Such optional
building data
storage repository may comprise a conventional data storage device such as a
hard disk or
solid-state memory located with and connected directly to building datasource
30 such as
by bus 308 as shown in FIG. 4, or may alternately comprise a remote data
storage facility
accessibly connected to building datasource 30. Also connected to bus 308 is a
computer-
readable memory 310. The memory 310 stores executable instructions to
implement
functions of the present invention. The computer-readable memory 310 may
comprise any
available computer-readable media or storage device that can be accessed by
the building
datasource computer 30.
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
In an embodiment of the present invention, one or more of the following
program
modules and/or data files may be stored in the memory 310 of building
datasource
computer 30: an operating system module 312, a building datasource module 314,
and an
energy meter module 316.
5 Similar to module 112 described above the operating system module 312
may
comprise any known executable operating system instructions, such as may be
suitable for
controlling the general operations of a networked building datasource 30, and
in particular
may comprise instructions for handling various system services, such as file
services or for
performing hardware dependent tasks. Operating system module 312 may also
comprise
10 instructions for standard computer operation, including receiving input
from input devices
such as a keyboard or mouse, and for displaying output in a graphical format
on a monitor,
for example. In one embodiment, operating system module 312 may comprise one
or more
known proprietary or open-source computer operating systems, such as for
example,
WindowsTM, MacOSTM, UNIXTM or LinuxTm operating systems, for example.
15 Building datasource module 314 comprises instructions for transmitting
and/or
receiving building energy load data to and/or from monitoring server 20. In
particular,
building datasource module 314 may comprise instructions to communicate with
monitoring server 20 such as over a computer network 50 and/or another
suitable
connection such as connection 35, such as to transmit building energy load
data for at least
20 one building to monitoring server 20. Such building energy load data may
comprise
historical building energy load values as may be stored by building datasource
30 and/or
current measured building energy load values as may be measured by one or more
building
energy meters 32 corresponding to one or more buildings, for example. In one
embodiment, building datasource module 314 may comprise instructions to
encrypt and/or
compress building energy load data using suitable known encryption and/or
compression
techniques for communication with monitoring server 20, such as for the
transmission of
historical and/or current measured building energy load values to monitoring
server 20, for
example.
In an optional embodiment of the present invention, building datasource 30 may
comprise a building automation system and/or building management system for
one or
DM_VAN/253729-15331/7496905.I
CA 02689035 2009-12-23
21
more buildings, and in such case building datasource module 314 may also
comprise
instructions to access a database connected to such building automation and/or
management
system which stores building energy load data, and to transmit such stored
building energy
load data to monitoring server 20, for example. Such stored building energy
load data may
comprise historical building energy load values and/or current measured
building energy
load values, for example. In a further optional embodiment, building
datasource 30 may
comprise a utility company computer system storing building energy load data
for multiple
buildings, for example. In such further optional embodiment, building
datasource module
314 may then comprise instructions to access at least one database of such
utility company
system to retrieve historical and/or current measured building energy load
values for at
least one building, such as to transmit the building energy load data to
monitoring server
20.
Energy meter module 316 comprises instructions to communicate with at least
one
building energy meter 32, to receive measurements of building energy load
and/or
consumption values from one or more building energy meter 32 corresponding to
at least
one building. In particular, energy meter module 316 may comprise instructions
to
communicate with various common types of digital energy meters 32 which may be
used to
meter one or more building energy input (such as electricity, gas, oil, coal,
steam, and
water, for example), such as by use of common energy meter protocols including
one or
more of Modbus (TM), BACnet, Fieldbus and LonWorks (TM) communication
protocols,
for example. In one embodiment, energy meter module 316 comprises a building
energy
data "connector layer" and comprises instructions to communicate with at least
one
building energy meter 32 and/or stored building meter repository. Energy meter
module
316 may also measure and retrieve building energy load values from said
building energy
meter 32 and/or repository for transmission to monitoring server 20, for
example. In an
optional embodiment of the present invention, energy meter module 316 may also
comprise
instructions to convert building energy data retrieved from building energy
meter 32 and/or
a building meter repository to a data format adapted for transmission to
monitoring server
20, and/or adapted for use by monitoring server 20 for calculating predicted
or typical
building energy load values, for example.
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
22
In another embodiment of the present invention, predictor variable datasource
server 40 may comprise an exemplary computer architecture substantially
similar to that of
building datasource server 30 as described above in reference to FIG. 4. In
particular,
predictor variable datasource server 40 may comprise substantially similar
hardware
components to those illustrated in FIG. 4 and described above. Such exemplary
predictor
variable datasource server 40 may also comprise a program module comprising
instructions
for transmitting and/or receiving predictor variable data to and/or from
monitoring server
20, such as historical and/or current measured predictor variable values
corresponding to
one or more buildings, for example.
The above described program modules incorporate instructions to implement
processing operations associated with aspects of the present invention.
Various
embodiments of the processing operations of the above-described program
modules are
described below with reference to FIG. 5. The modules stored in memory 310 are
exemplary, and additional modules can be included. It should be appreciated
that the
functions of the presented modules may be combined. In addition, a function of
a module
need not be performed on a single machine. Instead, a function of a module may
be
distributed across a network (such as network 50 or the like) to one or more
other
computers if desired. It is the functions of the present invention that are
significant, not
where they are performed or the specific manner in which they are performed.
FIG. 5 illustrates a series of processing operations or methods that may be
implemented by systems similar to those illustrated in FIG. 1, and the
exemplary computers
illustrated in FIGs 2-4, according to an embodiment of the invention. In the
first processing
operation 510 of FIG. 5, historical building energy load values and
corresponding historical
predictor variable data comprising values for one or more predictor variables
are received
by monitoring server 20. In one embodiment, monitoring server module 214 may
be used
to implement processing operation 510, such as by receiving historical
building energy load
values from building datasource 30, and by receiving historical predictor
variable data
comprising values for multiple predictor variables that corresponds to the
historical
building energy load values, such as from predictor variable datasource 40,
for example.
The historical building energy load values and corresponding historical
predictor variable
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
23
values may be received by the monitoring server 20, such as over computer
network 50, or
optionally over alternative connection 35, according to known methods. In an
alternative
embodiment, the historical building energy load and predictor variable values
may be
received by monitoring server 20 by means of a computer-readable memory
medium, such
as an optical disk or solid state memory device, for example.
In one embodiment, historical building energy load values for a building may
be
received by monitoring server 20 from building datasource 30, representing a
series of past
or historical building energy load values for at least one building energy
input (such as
electrical, gas, oil, coal, steam and heat inputs, for example) measured by
one or more
building energy meters 32 associated with the building. Similarly, historical
predictor
variable values corresponding to the building energy load values may be
received by
monitoring server from predictor variable datasource 40. Historical predictor
variable
values may include a series of past or historical values for multiple
predictor variables
(such as time, day, date, temperature, humidity, wind speed, wind direction,
solar radiation
and occupancy, for example) corresponding to a geographical location
associated with the
building, such as from a weather service, for example.
In a further embodiment, processing operation 510 may additionally include
storing
on monitoring server 20 at least one of: historical building energy load
values, historical
predictor variable values, current measured building energy load values,
predicted building
energy load values and current predictor variable values. This step may be
implemented by
monitoring server module 214 to store such building energy and predictor
variable values
in a monitoring data storage repository 22, for example. According to one
aspect of such
an embodiment, monitoring server module 214 may store historical and/or
current building
energy load values and/or predictor variable values in monitoring data storage
repository 22
as such values are received by monitoring server 20, such as from other
computers over
network 50, for example. According to another aspect of the present
embodiment,
monitoring server module 214 may also store predicted building energy load
values in
monitoring data storage repository 22 as they are calculated by monitoring
server 20, for
example.
DM_VAN/253729-15331/7496905.I
CA 02689035 2009-12-23
24
In the second processing operation 512 of FIG. 5, a scaling factor is defined
for
each predictor variable by minimizing cross-validation error between the
predicted and
measured historical building energy load values. Processing operation 512 may
be
implemented using scaling factor learning module 214 on monitoring server 20,
according
to one embodiment of the present invention. Scaling factor learning module 214
may
desirably be used to optimize calculation of predicted building energy load
values by a
kernel smoothing of historical building energy load values using historical
predictor
variable values received by monitoring server 20. In particular, scaling
factor values may
be determined for each of multiple predictor variables using scaling factor
learning module
216, by employing a cross-validation process to iteratively minimize the
difference between
measured and predicted historical building energy load values. The use of
arbitrary
uniform or randomized starting values for scaling factors for each predictor
variable may be
used at the outset of operation 512 to define the scaling factors through
minimizing cross-
validation error. Additionally, in some embodiments, a training data set and
validation data
set may be defined for use in defining scaling factors according to operation
512. Such a
training data set and validation data set may comprise at least a portion or
subset of the
historical building energy load values and corresponding historical predictor
variable values
received by monitoring server 20. Training data sets and validation data sets
may
optionally be defined automatically such as by monitoring server 20, or may be
defined by
selection of a user of user computer 10 such as may be received by monitoring
server 20 in
connection with the historical building energy load and predictor variable
data.
In order to optimize the learned scaling factor values defined, scaling factor
learning
operation 512 may comprise the use of scaling factor learning module 216 to
calculate an
rms error function defining the error between measured historical building
energy loads in
the validation data set and predicted building energy load values calculated
by kernel
smoothing of historical energy load values within the training data set using
their associated
predictor variable values, where each predictor variable is weighted or scaled
according to
the scaling factor values. For example, the rms error function for the
difference between
historical measured building energy load values and corresponding calculated
predicted
building energy load values may be expressed by:
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
(
1
eims )2
Ns
where S is the validation data set of historical building energy load values,
yi is the
historical measured building energy load value, and ST, is the corresponding
predicted
5 historical building energy load value.
In order to calculate the above-described rms error function, the scaling
factor
learning module 216 may also be used to calculate predicted values for
historical building
energy load values, by kernel smoothing, using a unit Gaussian kernel, of an
arbitrarily
selected number of nearest neighbor values of historical energy load values
using predictor
10 variable values scaled by the scaling factors. The euclidean distance in
the space of the
scaled predictor variables to each of the nearest neighbors is the input to
the kernel function
that determines the kernel weights for the kernel smoothing weighted average.
For
example, predicted historical building energy load values may be calculated
using scaling
factor learning module 216 according to the equation:
Eik(x,x,)y,
S)=
where ST is the calculated predicted building energy load value for the
corresponding values
of predictor variables in vector x, y, are the measured historical building
energy load
values for the arbitrarily selected number of nearest neighbors with predictor
variable
vectors x, that have been scaled by the scaling factors, and k(x,x,) is the
unit kernel
function.
In one embodiment, scaling factor learning processing operation 512 may
comprise
using scaling factor learning module 216 to optimize the learned scaling
factor values by
iteratively minimizing the above-defined rms error function using the
Levenberg-
Marquardt optimization algorithm for varying the learned scaling factor values
for
subsequent iterations of the scaling factor learning process. Such use of the
Levenberg-
DM_VAN/253729-15331 /7496905 1
CA 02689035 2009-12-23
26
Marquardt algorithm for optimization of linear systems may be implemented as
is known in
the art. In alternative embodiments, other suitable known optimization
algorithms or
techniques may be implemented by scaling factor learning module 216 for
iteratively
optimizing learned scaling factor values in processing operation 512.
In a particular embodiment, scaling factor learning processing operation 512
may
also comprise using scaling factor learning module 216 to implement a suitable
known k-
dimensional (or k-d) tree technique, such as a "best-bin first" k-d tree
algorithm, where
each of the k dimensions corresponds to a predictor variable that has been
scaled by the
scaling factor, to determine the nearest neighbor historical energy load
values and nearest
neighbor distances for a given set of predictor variable values, to be used
for the kernel
smoothing calculations. Such use of a k-d tree to compute the nearest neighbor
historical
energy load values and nearest neighbor distances may desirably be implemented
to
improve the speed and efficiency of calculation of predicted building energy
load values.
Alternatively, other suitable methods may be implemented to facilitate
computation of
nearest neighbor values for use in calculation of predicted building energy
load values, for
example.
In the next processing operation 514 of FIG. 5, monitoring server 20 receives
measured building energy load values from a building energy meter 32. In one
embodiment, monitoring server module 214 on monitoring server 20 may be used
to
implement processing operation 514, such as by receiving measured energy load
values
measured by at least one building energy meter 32, and transmitted to
monitoring server 20
either through building datasource 30, or alternatively transmitted directly
from energy
meter 32 to monitoring server 20, for example. In another embodiment, current
measured
building energy load values may be received by monitoring server 20 for one or
more
buildings on an arbitrary regular periodic basis such as an hourly semi-hourly
"real-time"
basis, for example. In a further embodiment, processing operation 514 may also
comprise
storing current measured building energy values on monitoring server 20 as
they are
received, such as by storing such measured building energy load values on
monitoring data
storage repository 22, for example.
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
27
In the next processing operation 516 of FIG. 5, measured predictor variable
values
from predictor variable datasource 40, which correspond to the above-mentioned
measured
building energy load values in operation 514, are received by monitoring
server 20. In one
embodiment, monitoring server module 214 on monitoring server 20 may be used
to
implement processing operation 516, such as by receiving current measured
predictor
variable values for multiple predictor variables transmitted from a predictor
variable
datasource 40, and corresponding to the measured building energy load values
received in
processing operation 514 above, for example. In a particular embodiment, the
current
measured predictor variable values may correspond to measured building energy
load
values for one or more buildings, such as for one or more buildings located in
the same
geographical area, for example. Alternatively, measured predictor variable
values may be
received which correspond specifically to one individual building. Similar to
as described
above, current predictor variable values corresponding to one or more
buildings may also
be received by monitoring server 20 on an arbitrary regular periodic basis
such as an hourly
semi-hourly "real-time" basis as may be provided by a weather service, for
example. In a
further embodiment, processing operation 516 may also comprise storing current
measured
predictor variable values on monitoring server 20 as they are received, such
as by storing
such measured predictor variable values on monitoring data storage repository
22, for
example. Also, as described above in reference to processing operation 510,
measured
predictor variable values may comprise values for predictor variables
including one or more
of: time, day, date, temperature, humidity, wind speed, wind direction, solar
radiation and
occupancy, for example.
In the next processing operation 518 of FIG. 5, predicted building load values
are
determined by kernel smoothing of nearest neighbor historical building load
values using
predictor variables that have been scaled by the scaling factors. In one
embodiment,
prediction module 218 on monitoring server 20 may be used to implement
processing
operation 518. In particular, processing operation 518 may comprise using
prediction
module 218 to calculate a current predicted building energy load value (also
known as a
typical building energy load value) which corresponds to the current measured
predictor
variable values received from predictor variable datasource 40, by kernel
smoothing of
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
28
historical building energy loads using predictor variable values received in
processing
operation 510 that have been scaled by the scaling factors, for example.
Accordingly,
prediction module 218 may be used to calculate a current predicted building
energy load
value for a building at a particular time, by applying a kernel smoothing of
the nearest
neighbor historical building energy load values using predictor variable
values for that
building that have been scaled using the corresponding scaling factor values
for each of the
multiple predictor variables.
In one embodiment, similar to as described above in reference to scaling
factor
learning processing operation 512, processing operation 518 may comprise using
prediction
module 218 to calculate a current predicted (or typical) building energy load
value ST, for a
specific vector of predictor variable values, x, according to the equation:
k(x,x,)y,
k(x,x,)
where ST is the calculated predicted building energy load value for the
corresponding values
of predictor variables in vector x, y, are the measured historical building
energy load
values for the arbitrarily selected number of nearest neighbors with predictor
variable
vectors x, that have been scaled by the scaling factors, and k(x,x,) is the
unit kernel
function. In such an embodiment, the kernel function k(x,xi) may comprise a
unit
Gaussian kernel. Also in such an embodiment, processing operation 518 may also
comprise using prediction module 218 to implement a suitable k-d tree
technique, such as
the "best-bin first" k-d tree algorithm mentioned above, to determine an
arbitrary number of
nearest neighbor values of the historical energy load values and nearest
neighbour distances
using the predictor variable values scaled by the scaling factors to be used
for the weighted
average kernel smoothing calculations of current predicted building energy
load. In one
embodiment, an arbitrary number of nearest neighbor values of historical
energy load
values may be used for the weighted average kernel smoothing calculation, as
may be
determined automatically by the monitoring server, or as may be determined by
a user, such
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
29
as to control the duration or computing resources required to complete the
kernel
smoothing calculations, for example.
In a further embodiment, processing operation 518 may also comprise using
prediction module 218 to calculate current predicted (or typical) building
energy load
values for any of one or more buildings for which historical building energy
load and
predictor variable values are stored by monitoring server 20. In a similar
embodiment,
processing operation 518 may include using prediction module 218 to
periodically calculate
new current predicted building energy load values for one or more buildings at
arbitrary
regular time intervals, thereby providing regularly updated predicted (or
typical) building
energy load values, that may be used for ongoing regular monitoring of
building energy
consumption of one or more buildings, for example.
In the next processing operation 520 of FIG. 5, monitoring server 20 is used
to
compare the predicted building energy load with the measured building energy
load to
determine if their difference exceeds an alert threshold level. In one
embodiment, alert
module 220 of monitoring server 20 may be used to implement processing
operation 520.
In a particular embodiment, an alert threshold level may be arbitrarily
predetermined, either
automatically such as by monitoring server 20, or by a user of user computer
10, for
example. In a case where the alert threshold level is determined
automatically, such as by
monitoring server 20, the alert threshold level may comprise a relative
comparison such as
an alert level corresponding to a 10%, 25%, or 50% difference between the
predicted (or
typical) building energy load value and the measured building energy load
value, for
example, which may be automatically selected in order to reflect a likely
indicator that an
unusual building energy load level has been measured for a building and that
an alert is
therefore justified. In a case where the alert threshold level is determined
by a user of user
computer 10, the user may preset any desired alert threshold level such as a
relative or an
absolute difference between the predicted (or typical building energy load
value) and
measured building energy load values that is desired by the user to be used to
determine if
an alert is justified. Such a user-defined alert threshold level may be
transmitted from the
user computer 10 to the monitoring server 20, such as in conjunction with the
transmission
of historical building energy load values to the monitoring server 20, or in
connection with
DM_VAN/253729-15331/7496905 1
CA 02689035 2009-12-23
selection of a training data set or validation data set or other parameters
related to
calculation of predicted building energy load values, which may be transmitted
from user
computer 10 to monitoring server 20, for example.
In an embodiment of the present invention including multiple buildings and the
5 calculation of multiple current predicted building energy load values,
processing operation
520 may also comprise using alert module 220 to compare a series of predicted
or typical
building energy load values for each of one or more buildings, spaced over a
period of time
with a series of corresponding measured building energy load values, and to
represent the
series of predicted values for each building as a predicted building energy
load curve (also
10 referred to as a typical building energy load curve) and the series of
measured values for
each building as a measured or actual building energy load curve. Processing
operation
520 may further include comparison of such a predicted or typical building
energy curve
with a measured or actual building energy curve to determine if the difference
between
predicted and measured values exceeds the alert threshold level over a period
of time. The
15 period of time considered in such an embodiment may be predetermined
such as
automatically by monitoring server 20, or by user input from user computer 10,
for
example. In a further embodiment, processing operation 520 may additionally
comprise
using monitoring module 520 to provide such predicted (or typical) building
energy load
curves and measured (or actual) building energy load curves to a user computer
10, such as
20 by means of using monitoring server module 214 on monitoring server 20
to provide the
typical and actual curves to a user computer 10 in the form of a webpage
and/or a web
application interface such as over computer network 50, for example. In a
particular such
embodiment, such typical and actual building energy curves may be provided to
user
computer 10 in a graphical format, such as part of one or more graphical user
interfaces
25 which may be adapted for use by various users to monitor and understand
variations in
predicted and measured building energy use for one or more buildings under
monitoring.
In the last processing operation 522 of FIG. 5, monitoring server 20 is used
to
transmit an alert signal to a user if an alert threshold level is exceeded. In
one embodiment
of the present invention, alert module 220 of monitoring server 20 may be used
to
30 implement processing operation 522. In a particular embodiment, if the
difference between
DM_VAN/253729-15331/7496905.1
CA 02689035 2009-12-23
31
a predicted building energy load value and a corresponding building energy
load value for a
particular building exceeds the alert threshold level as described above in
processing
operation 520, alert module 220 may transmit an alert signal to a user, such
as a user of user
computer 10, for example. Such an alert signal may be transmitted to a user by
any suitable
means, such as over computer network 50 to user computer 10 as an email, VOIP
or instant
messaging alert signal for example, by a telecommunication link such as to a
mobile device
as a voice or text alert signal, or to a paging device or the like, for
example. In another
embodiment, the alert module 220 of monitoring server 20 may also transmit
details of at
least one of current predicted building energy load values and current
measured building
energy load values for a particular building corresponding to the alert signal
transmitted to
a user. For example, the alert module 220 may transmit an alert signal as well
as a series of
measured building energy load values such as an actual building energy curve,
and/or a
series of predicted building energy load values such as a predicted or typical
curve, as
described above, so that the user receiving the alert signal may also receive
and review the
status of predicted and actual building energy use for the building associated
with the alert
signal generation.
In an alternative embodiment of the present invention, the monitoring server
20 may
receive scaling factor values for predictor variables as well as historical
building energy
load values and corresponding historical predictor variable values, such as in
cases where
the scaling factor values may have been previously determined for a given
dataset of
historical building energy load and predictor variable values. In such an
alternative
embodiment, processing operation 510 may include receiving defined scaling
factor values
for each predictor variable, and processing operation 512 may be omitted from
the series of
processing operations shown in FIG. 5, for example.
In a further optional embodiment of the present invention, processing
operations
520 and 522 may be omitted from the series of processing operations shown in
FIG. 5, and
the predicted building energy load value determined according to processing
operation 518
may be transmitted to a user, such as by transmission to user computer 10 for
graphical
display to a user, for example. In such an embodiment, a user may receive the
predicted
building energy load value determined by monitoring server 20, such as in a
graphical
DM_VAN/253729-1533 I /7496905 I
CA 02689035 2009-12-23
32
representation of a time series of predicted building energy load values, such
as a typical
building energy load curve, or as a single predicted building energy load
value for a
particular time.
Referring to FIG. 6, a graphical representation of measured building energy
load
values, and predictor variable values over a period of time are shown,
according to an
embodiment of the present invention. In particular, FIG. 6 shows a graphical
presentation
of a time series of measured building energy load values 610 for a building,
and also shows
corresponding measured predictor variable values including temperature values
620 and
humidity values 630 for the same period of time. In one embodiment, such
building energy
load values 610 and temperature 620 and humidity 630 predictor variable values
may
represent historical measured building energy load values and corresponding
historical
predictor variable values, such as described above in reference to FIG. 5,
which may be
received by monitoring server 20, for example. Such historical building energy
load values
and corresponding predictor variable values may be used for determining
scaling factor
values such as by a scaling factor learning process implemented by scaling
factor learning
module 216 of monitoring server 20, for example. In another embodiment,
historical
measured building energy load values 610 and predictor variable values 620 and
630 may
also be used in a kernel smoothing operation to calculate predicted building
energy load
values, such as by a unit Gaussian kernel smoothing of nearest neighbor values
as described
above, which may be implemented by prediction module 218 of monitoring server
20, for
example.
Referring to FIG. 7, graphical representations of a scaling factor learning
process
similar to as is described above in reference to processing operation 512 is
shown.
According to one embodiment, a time series of measured building energy load
values 720
are shown along with predicted building energy load values 725 calculated
using a unit
Gaussian kernel smoothing of nearest neighbor values of the measured building
energy
loads 720 and measured predictor variable values 730 which correspond to the
measured
building energy load values. Also shown are values for scaling factor
parameters 740
which may be iteratively varied to optimize the prediction of building energy
load values
725, according to the scaling factor learning process described above in
reference to
DMyAN/253729-15331/7496905 1
CA 02689035 2016-08-10
,
33
processing operation 512, and which may be implemented using scaling factor
learning
module 216 according to an embodiment of the invention. As the scaling factor
values 740
are iteratively optimized during the scaling factor learning process, the rms
building energy
load error 710 of the differences between measured building energy load values
720 and
corresponding predicted building energy load values 725 is shown to decrease
with such
iteration, as shown in the downward trend of rms building energy load error
710 of FIG. 7,
for example. Therefore, as described above, the scaling factor learning
process may
provide for determining optimized values of scaling factors 740 for use in
calculating
predicted values of building energy load 725, according to embodiments of the
invention.
An embodiment of the present invention relates to a computer storage product
with
a computer-readable medium having computer code thereon for performing various
computer-implemented operations. The computer-readable media and computer code
may
be specially designed and constructed for the purposes of the present
invention, or they may
be of the kind well known and available to a person having skill in the
computer software
arts. Examples of computer-readable media include, but are not limited to:
magnetic media
such as hard disks, floppy disks, and magnetic tape; optical media such as CD-
ROMs and
holographic devices; magneto-optical media such as floptical disks; and
hardware devices
that are specially configured to store and execute program code, such as
application-
specific integrated circuits ("ASICs"), programmable logic devices ("PLDs"),
ROM and
RAM devices. Examples of computer code include machine code, such as produced
by a
compiler, and files containing higher-level code that are executed by a
computer using an
interpreter. For example, an embodiment of the invention may be implemented
using Java,
HTML/XML, JavaScript, C#, C++, or other scripting, markup and/or programming
languages and development tools. Another embodiment of the present invention
may be
implemented in hardwired circuitry in place of, or in combination with,
machine-executable
software instructions.
