Method for identifying pattern of load cycle
A method for identifying a pattern of a load cycle includes: performing statistics on a daily load of a system based on smart meter data; generating a curve of the daily load of the system according to the statistics on the daily load of the system; acquiring a result of clustering curves of loads of typical days by applying shape-based time sequence clustering analysis using the curve of the daily load of the system; and identifying a pattern of a load cycle according to the result of clustering the curves of the loads of the typical days.
1. A method for identifying a pattern of a load cycle, comprising:
performing statistics on a daily load of a system based on smart meter data;
generating a curve of the daily load of the system according to the statistics on the daily load of the system;
acquiring a result of clustering curves of loads of typical days by applying shape-based time sequence clustering analysis using the curve of the daily load of the system; and
identifying a pattern of a load cycle according to the result of clustering the curves of the loads of the typical days;
performing load prediction according to the result of clustering the curves of the loads of the typical days,
wherein the generating a curve of the daily load of the system comprises:
acquiring a curve of a load for 24 hours of the system by accumulating a curve of a load for 24 hours consumed by each smart meter user in an area or the system,
wherein the smart meter data comprise active power, reactive power, a voltage, a current, and a power factor, wherein the load is an active power reading,
wherein the curve of the daily load of the system describes variation of the load over time within a day, wherein the curve of the daily load varies depending on a workday, a weekend, or a holiday of a season in a region, wherein a curve of a load of a typical day in a typical season, comprising a curve of a load of a typical day and a typical curve of a continued daily load, is used,
wherein the performing load prediction according to the result of clustering the curves of the loads of the typical days comprises: searching for a similar day in history according to a factor, grouping or clustering by the shape-based time sequence clustering analysis, and a curve of the daily load in a historical year, and estimating a curve of a load of the system for a day to be predicted according to a curve of the load for the similar day in history, a curve of the load for recent days, and weather forecast data,
wherein the factor comprises at least one of a type of a date, a period of time for central heating, a temperature, or a rainfall.
2. The method according to claim 1 , wherein time sequence clustering analysis depends on measurement of a distance between a data point and a prototype, wherein curves of similar shapes are clustered together by shape-based clustering, to reduce impact of a difference in an amplitude and a difference in a phase on time sequence clustering,
wherein a similarity between shapes of two time sequences is measured via shape-based time sequence clustering analysis by computing cross-correlation of the two time sequences, by
comparing the similarity between a time sequence =(x 1 , . . . , x m ) and a time sequence =(y 1 , . . . , y m ), by first keeping the time sequence invariant and computing a distance by which the time sequence is to be translated as:
x
⇀
=
{
(
0
,
…
,
0
︷
s
,
x
1
,
x
2
,
…
,
x
m
-
s
)
,
s
≥
0
(
x
1
-
s
,
…
,
x
m
-
1
,
x
m
,
0
,
…
,
0
︸
s
)
,
s
<
0
,
wherein s∈[−m, m], CC ω ( , )=(c 1 , . . . , c ω ) the m is a number of time sequences, the ω represents a ωth time sequence, ωϵ{1, 2, . . . , 2m−1}, CC ω is a cross-correlation sequence, x 1 , . . . , x m are elements of the time sequence , y 1 , . . . , y m are elements of the time sequence , and the s is an offset of the ,
wherein a cross-correlation sequence of a length of 2m−1 is defined as:
CC
ω
(
x
→
,
y
→
)
=
R
k
(
x
→
,
y
→
)
,
ω
∈
{
1
,
2
,
…
2
m
-
1
}
,
wherein
R
k
(
x
→
,
y
→
)
=
{
∑
l
=
1
m
-
k
x
l
+
k
y
l
,
k
≥
0
R
-
k
(
y
→
,
x
→
)
,
k
<
0
,
wherein a value ω that maximizes the CC ω ( , ) is to be computed, an optimal value by which the is to be translated based on the is to be computed based on the value ω, s=k∘R k ( , ), the l is an integer starting at 1, the k is an integer with an absolute value less than the m, the R k represents a cross-correlation sequence, the x l+k is an element of the time sequence , the R −k represents a cross-correlation sequence, and the y l is an element of the time sequence .
3. The method according to claim 1 , wherein each cluster in the result of clustering the curves of the loads of the typical days acquired by the shape-based time sequence clustering analysis corresponds to a typical type of the daily load in actual data, the typical type of the daily load comprising one of a workday in winter, a holiday in spring, and a weekend in summer, wherein the pattern of the load cycle is identified by data-driven clustering of the curve of the daily load of the system acquired according to the smart meter data of all users in a region, wherein a result of the data-driven clustering describes and identifies the pattern of the load cycle of the system.
4. A device for identifying a pattern of a load cycle, comprising:
a processor; and
memory storing instructions executable by the processor,
wherein the processor is arranged for implementing a method for identifying a pattern of a load cycle, the method comprising:
performing statistics on a daily load of a system based on smart meter data;
generating a curve of the daily load of the system according to the statistics on the daily load of the system;
acquiring a result of clustering curves of loads of typical days by applying shape-based time sequence clustering analysis using the curve of the daily load of the system, and
identifying a pattern of a load cycle according to the result of clustering the curves of the loads of the typical days;
performing load prediction according to the result of clustering the curves of the loads of the typical days,
wherein the generating a curve of the daily load of the system comprises:
acquiring a curve of a load for 24 hours of the system by accumulating a curve of a load for 24 hours consumed by each smart meter user in an area or the system,
wherein the smart meter data comprise active power, reactive power, a voltage, a current, and a power factor, wherein the load is an active power reading,
wherein the curve of the daily load of the system describes variation of the load over time within a day, wherein the curve of the daily load varies depending on a workday, a weekend, or a holiday of a season in a region, wherein a curve of a load of a typical day in a typical season, comprising a curve of a load of a typical day and a typical curve of a continued daily load, is used,
wherein the performing load prediction according to the result of clustering the curves of the loads of the typical days comprises: searching for a similar day in history according to a factor, grouping or clustering by the shape-based time sequence clustering analysis, and a curve of the daily load in a historical year, and estimating a curve of a load of the system for a day to be predicted according to a curve of the load for the similar day in history, a curve of the load for recent days, and weather forecast data,
wherein the factor comprises at least one of a type of a date, a period of time for central heating, a temperature, or a rainfall.
5. The device according to claim 4 , wherein time sequence clustering analysis depends on measurement of a distance between a data point and a prototype, wherein curves of similar shapes are clustered together by shape-based clustering, to reduce impact of a difference in an amplitude and a difference in a phase on time sequence clustering,
wherein a similarity between shapes of two time sequences is measured via shape-based time sequence clustering analysis by computing cross-correlation of the two time sequences, by
comparing the similarity between a time sequence =(x 1 , . . . , x m ) and a time sequence =y 1 , . . . , y m ), by first keeping the time sequence invariant and computing a distance by which the time sequence is to be translated as:
x
⇀
=
{
(
0
,
…
,
0
︷
s
,
x
1
,
x
2
,
…
,
x
m
-
s
)
,
s
≥
0
(
x
1
-
s
,
…
,
x
m
-
1
,
x
m
,
0
,
…
,
0
︸
s
)
,
s
<
0
,
wherein s∈[−m, m], CC ω ( , )=(c 1 , . . . , c ω ), the m is a number of time sequences, the ω represents a ωth time sequence, ωϵ{1, 2, . . . , 2m−1}, CC ω is a cross-correlation sequence, x 1 , . . . , x m are elements of the time sequence , y 1 , . . . , y m are elements of the time sequence , and the s is an offset of the ,
wherein a cross-correlation sequence of a length of 2m−1 is defined as:
CC
ω
(
x
→
,
y
→
)
=
R
k
(
x
→
,
y
→
)
,
ω
∈
{
1
,
2
,
…
2
m
-
1
}
,
wherein
R
k
(
x
→
,
y
→
)
=
{
∑
l
=
1
m
-
k
x
l
+
k
y
l
,
k
≥
0
R
-
k
(
y
→
,
x
→
)
,
k
<
0
,
wherein a value ω that maximizes the CC ω ( , ) is to be computed, an optimal value by which the is to be translated based on the is to be computed based on the value ω, s=k∘R k ({right arrow over (x)}, {right arrow over (y)}) the l is an integer starting at 1, the k is an integer with an absolute value less than the m, the R k represents a cross-correlation sequence, the x l+k is an element of the time sequence , the R −k represents a cross-correlation sequence, and the y l is an element of the time sequence .
6. The device according to claim 4 , wherein each cluster in the result of clustering the curves of the loads of the typical days acquired by the shape-based time sequence clustering analysis corresponds to a typical type of the daily load in actual data, the typical type of the daily load comprising one of a workday in winter, a holiday in spring, and a weekend in summer, wherein the pattern of the load cycle is identified by data-driven clustering of the curve of the daily load of the system acquired according to the smart meter data of all users in a region, wherein a result of the data-driven clustering describes and identifies the pattern of the load cycle of the system.
7. A non-transitory computer-readable storage medium having stored therein instructions that, when executed by a processor, cause the processor to perform a method for identifying a pattern of a load cycle, the method comprising:
performing statistics on a daily load of a system based on smart meter data;
generating a curve of the daily load of the system according to the statistics on the daily load of the system;
acquiring a result of clustering curves of loads of typical days by applying shape-based time sequence clustering analysis using the curve of the daily load of the system; and
identifying a pattern of a load cycle according to the result of clustering the curves of the loads of the typical days;
performing load prediction according to the result of clustering the curves of the loads of the typical days,
wherein the generating a curve of the daily load of the system comprises:
acquiring a curve of a load for 24 hours of the system by accumulating a curve of a load for 24 hours consumed by each smart meter user in an area or the system,
wherein the smart meter data comprise active power, reactive power, a voltage, a current, and a power factor, wherein the load is an active power reading,
wherein the curve of the daily load of the system describes variation of the load over time within a day, wherein the curve of the daily load varies depending on a workday, a weekend, or a holiday of a season in a region, wherein a curve of a load of a typical day in a typical season, comprising a curve of a load of a typical day and a typical curve of a continued daily load, is used,
wherein the performing load prediction according to the result of clustering the curves of the loads of the typical days comprises: searching for a similar day in history according to a factor, grouping or clustering by the shape-based time sequence clustering analysis, and a curve of the daily load in a historical year, and estimating a curve of a load of the system for a day to be predicted according to a curve of the load for the similar day in history, a curve of the load for recent days, and weather forecast data,
wherein the factor comprises at least one of a type of a date, a period of time for central heating, a temperature, or a rainfall.
8. The storage medium according to claim 7 , wherein time sequence clustering analysis depends on measurement of a distance between a data point and a prototype, wherein curves of similar shapes are clustered together by shape-based clustering, to reduce impact of a difference in an amplitude and a difference in a phase on time sequence clustering,
wherein a similarity between shapes of two time sequences is measured via shape-based time sequence clustering analysis by computing cross-correlation of the two time sequences, by
comparing the similarity between a time sequence =(x 1 , . . . , x m ) and a time sequence =y 1 , . . . , y m ), by first keeping the time sequence invariant and computing a distance by which the time sequence is to be translated as:
x
⇀
=
{
(
0
,
…
,
0
︷
s
,
x
1
,
x
2
,
…
,
x
m
-
s
)
,
s
≥
0
(
x
1
-
s
,
…
,
x
m
-
1
,
x
m
,
0
,
…
,
0
︸
s
)
,
s
<
0
,
wherein s∈[−m, m], CC ω ( , )=(c 1 , . . . , c ω ), the m is a number of time sequences, the ω represents a ωth time sequence, ωϵ{1, 2, . . . , 2m−1}, CC ω is a cross-correlation sequence, x 1 , . . . , x m are elements of the time sequence , y 1 , . . . , y m are elements of the time sequence , and the s is an offset of the ,
wherein a cross-correlation sequence of a length of 2m−1 is defined as:
CC
ω
(
x
→
,
y
→
)
=
R
k
(
x
→
,
y
→
)
,
ω
∈
{
1
,
2
,
…
2
m
-
1
}
,
wherein
R
k
(
x
→
,
y
→
)
=
{
∑
l
=
1
m
-
k
x
l
+
k
y
l
,
k
≥
0
R
-
k
(
y
→
,
x
→
)
,
k
<
0
,
wherein a value ω that maximizes the CC ω ( , ) is to be computed, an optimal value by which the is to be translated based on the is to be computed based on the value ω, s=k∘R k ({right arrow over (x)}, {right arrow over (y)}) the l is an integer starting at 1, the k is an integer with an absolute value less than the m, the R k represents a cross-correlation sequence, the x l+k is an element of the time sequence , the R −k represents a cross-correlation sequence, and the y l is an element of the time sequence .
9. The storage medium according to claim 7 , wherein each cluster in the result of clustering the curves of the loads of the typical days acquired by the shape-based time sequence clustering analysis corresponds to a typical type of the daily load in actual data, the typical type of the daily load comprising one of a workday in winter, a holiday in spring, and a weekend in summer, wherein the pattern of the load cycle is identified by data-driven clustering of the curve of the daily load of the system acquired according to the smart meter data of all users in a region, wherein a result of the data-driven clustering describes and identifies the pattern of the load cycle of the system.