# US 12,224,586 B2 · Adaptive persistence forecasting for control of distributed energy resources

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,224,586 B2
- Document ID: US12224586B2
- Status: Granted
- Kind: B2
- Application number: 17/728,177
- Filed: 2022-04-25
- Published: 2023-10-26
- Granted: 2025-02-11
- Earliest claimed: 2022-04-25
- Assignees: Eaton Intelligent Power Limited
- Applicants: Eaton Intelligent Power Limited
- Inventors: Hossein Ghassempour Aghamolki; Arun Sukumaran Nair; Abhinandan Tripathi; David Wu Ganger
- Web page: https://patentgrep.com/patent/US12224586B2

## CPC

- H02J 3/003: {Load forecast, e.g. methods or systems for forecasting future load demand}
- G05B 13/048: {using a predictor}
- H02J 2103/30: Simulating, planning, modelling, reliability check or computer assisted design [CAD] of electric power networks
- H02J: ELECTRIC POWER NETWORKS; CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)

## Abstract

A method of adaptive persistence forecasting includes receiving historical load values for a site with at least one component that consumes energy; receiving historical temperature values corresponding to dates of the historical load values; evaluating the historical load values and the historical temperature values to determine a correlation coefficient; determining that there exists at least a threshold correlation between a load activity and temperature for the historical load values and the historical temperature values based on the correlation coefficient; in response to determining that there exists at least the threshold correlation, normalizing the historical load values based on a set temperature; and applying an adaptive seasonal persistence model to the normalized historical load values to output a forecast for use in controlling energy resources at the site.

## Claims

### Claim 1

A method of adaptive persistence forecasting for load control comprising:
receiving historical load values for a site comprising at least one component that consumes energy;
receiving historical temperature values corresponding to dates of the historical load values;
evaluating the historical load values and the historical temperature values to determine a correlation coefficient;
determining that there exists at least a threshold correlation between a load activity and temperature for the historical load values and the historical temperature values based on the correlation coefficient;
in response to determining that there exists at least the threshold correlation, normalizing the historical load values based on a set temperature;
applying an adaptive seasonal persistence model to the normalized historical load values to output a forecast;
receiving a temperature prediction;
generating a correction term using the temperature prediction and a correlation relationship between the historical load values and the historical temperature values identified when evaluating the historical load values and the historical temperature values;
applying the correction term to the forecast;
rescaling the forecast; and
controlling energy resources at the site using the rescaled forecast.

### Claim 2 (depends on 1)

The method of claim 1, wherein the correction term is given as:
correction
⁢
term
=
f
⁡
(
t
fi
)
-
f
⁡
(
t
ref
)
f
⁡
(
t
fi
)
where tref is the set temperature, tfi is the temperature prediction and f( ) represents a best fitting line equation from the correlation relationship between the historical load values and the historical temperature values identified when evaluating the historical load values and the historical temperature values.

### Claim 3 (depends on 1)

The method of claim 1, wherein the adaptive seasonal persistence model is given as:
P(t+Δt|t)=f(P(t−δit),
where P(t+Δt|t) is the forecast for a time period Δt made at time t, f represents a function used to summarize selected values from the historical load values, and δi are lag values for selecting a number of past measurements from the historical load values.

### Claim 4 (depends on 3)

The method of claim 3, further comprising training the adaptive seasonal persistence model on the received historical load values.

### Claim 5 (depends on 4)

The method of claim 4, wherein the adaptive seasonal persistence model is updated over time.

### Claim 6 (depends on 4)

The method of claim 4, wherein training the adaptive seasonal persistence model on the received historical load values comprises:
selecting a set of lag values;
selecting a function from a set of functions comprising mean, median, and mode;
generating a prediction for a particular date using the adaptive seasonal persistence model with selected set of lag values and selected function;
iterating the selecting of the set of lag values, selecting of the function, and generating the prediction for the particular date to generate a set of models; and
evaluating each model of the set of models using at least one error metric to identify an optimal seasonal persistence model, wherein the optimal seasonal persistence model is used as the adaptive seasonal persistence model when applying the adaptive seasonal persistence model to the normalized historical load values to output the forecast.

### Claim 7 (depends on 6)

The method of claim 6, wherein the at least one error metric is selected from the group consisting of:
mean absolute percentage error (MAPE),
root mean square percentage error (RMSPE), and
mean square percentage error (MSPE), where
MAPE
=
1
m
⁢
∑
i
=
1
m
⁢
(
1
n
⁢
∑
t
=
1
n
⁢
❘
"\[LeftBracketingBar]"
y
ˆ
t
-
y
t
❘
"\[RightBracketingBar]"
y
t
)
RMSPE
=
1
m
⁢
∑
i
=
1
m
⁢
(
1
n
⁢
∑
t
=
1
n
⁢
(
y
ˆ
t
-
y
t
y
t
)
2
)
MSPE
=
1
m
⁢
∑
i
=
1
m
⁢
(
1
n
⁢
∑
t
=
1
n
⁢
(
y
ˆ
t
-
y
t
y
t
)
2
)
where ŷt are the predictions of particular dates from the seasonal persistence model with selected lags and selected functions, yt are actual values of the historical load values from the particular dates, n is a number of data points in a forecast time period, and m is a number of iterations.

### Claim 8 (depends on 1)

The method of claim 1, wherein the historical load values are for a week of load values.

### Claim 9 (depends on 1)

The method of claim 1, wherein the forecast is provided to a display.

### Claim 10 (depends on 1)

The method of claim 1, wherein the forecast is provided to a load controller.

### Claim 11

A system comprising:
a processor,
a memory storage; and
instructions for adaptive persistence forecasting stored on the memory storage that when executed by the processor, direct the system to:
receive historical load values for a site comprising at least one component that consumes energy;
receive historical temperature values corresponding to dates of the historical load values;
evaluate the historical load values and the historical temperature values to determine a correlation coefficient;
determine that there exists at least a threshold correlation between a load activity and temperature for the historical load values and the historical temperature values based on the correlation coefficient;
normalize the historical load values based on a set temperature;
apply an adaptive seasonal persistence model to the normalized historical load values to output a forecast;
receive a temperature prediction;
generate a correction term using the temperature prediction and a correlation relationship between the historical load values and the historical temperature values identified when evaluating the historical load values and the historical temperature values;
apply the correction term to the forecast; and
provide the forecast for use in controlling energy resources at the site; and
control instructions stored on the memory storage that when executed by the processor, direct the system to:
receive the forecast; and
control the energy resources using the forecast.

### Claim 12 (depends on 11)

The system of claim 11, wherein the system is a single board computer.

### Claim 13 (depends on 11)

The system of claim 11, wherein the system is a programmable logic controller.

### Claim 14 (depends on 11)

The system of claim 11, wherein the system further comprises a network interface;
wherein the instructions to receive the temperature prediction direct the system to:
communicate, via the network interface, with a weather application programming interface to obtain the temperature prediction.

### Claim 15 (depends on 11)

The system of claim 11, wherein the correction term is given as:
correction
⁢
term
=
f
⁡
(
t
f
⁢
i
)
-
f
⁡
(
t
ref
)
f
⁡
(
t
f
⁢
i
)
where tref is the set temperature, tfi is the temperature prediction and f( ) represents a best fitting line equation from the correlation relationship between the historical load values and the historical temperature values identified when evaluating the historical load values and the historical temperature values.

### Claim 16 (depends on 11)

The system of claim 11, wherein the adaptive seasonal persistence model is given as:
P(t+Δt|t)=f(P(t−δit),
where P(t+Δt|t) is the forecast for a time period Δt made at time t, f represents a function used to summarize selected values from the historical load values, and δi are lag values for selecting past measurements from the historical load values.

### Claim 17 (depends on 11)

The system of claim 11, wherein the instructions further direct the system to:
train the adaptive seasonal persistence model on the received historical load values by:
selecting a set of lag values;
selecting a function from a set of functions comprising mean, median, and mode;
generating a prediction for a particular date using the adaptive seasonal persistence model with selected set of lag values and selected function;
iterating the selecting of the set of lag values, selecting of the function, and generating the prediction for the particular date to generate a set of models; and
evaluating each model of the set of models using at least one error metric to identify an optimal seasonal persistence model, wherein the optimal seasonal persistence model is used as the adaptive seasonal persistence model when applying the adaptive seasonal persistence model to the normalized historical load values to output the forecast.

### Claim 18 (depends on 17)

The system of claim 17, wherein the at least one error metric is selected from the group consisting of:
mean absolute percentage error (MAPE),
root mean square percentage error (RMSPE), and
mean square percentage error (MSPE), where
MAPE
=
1
m
⁢
∑
i
=
1
m
⁢
(
1
n
⁢
∑
t
=
1
n
⁢
❘
"\[LeftBracketingBar]"
y
ˆ
t
-
y
t
❘
"\[RightBracketingBar]"
y
t
)
RMSPE
=
1
m
⁢
∑
i
=
1
m
⁢
(
1
n
⁢
∑
t
=
1
n
⁢
(
y
ˆ
t
-
y
t
y
t
)
2
)
MSPE
=
1
m
⁢
∑
i
=
1
m
⁢
(
1
n
⁢
∑
t
=
1
n
⁢
(
y
ˆ
t
-
y
t
y
t
)
2
)
where ŷt are the predictions of particular dates from the seasonal persistence model with selected lags and selected functions, yt are actual values of the historical load values from the particular dates, n is a number of data points in a forecast time period, and m is a number of iterations.