CN105335800A - Method for forecasting electricity consumption of power consumers based on joint learning - Google Patents
Method for forecasting electricity consumption of power consumers based on joint learning Download PDFInfo
- Publication number
- CN105335800A CN105335800A CN201510801660.4A CN201510801660A CN105335800A CN 105335800 A CN105335800 A CN 105335800A CN 201510801660 A CN201510801660 A CN 201510801660A CN 105335800 A CN105335800 A CN 105335800A
- Authority
- CN
- China
- Prior art keywords
- user
- power supply
- supply area
- matrix
- power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for forecasting electricity consumption of power consumers based on joint learning. The method for forecasting the electricity consumption of the power consumers based on the joint learning comprises the following steps: constructing a consumer electricity consumption-consumer geographical location information matrix; partitioning power supply regions, and constructing an overall consumption behaviour similarity matrix for consumers between regions; and constructing electricity consumption forecasting models, and carrying out joint optimization solution. The method for forecasting the electricity consumption of the power consumers based on the joint learning has the advantages that the consumers are partitioned into different power supply regions while geographical locations and consumption behaviours of the consumers are considered, the consumers in each region are adjacent in the geographical location and have similar electricity demands; and a forecasting model is constructed for different power supply regions respectively, and each forecasting model not only considers unique electricity demands of the consumers in each region but also considers relation between power supply regions and public factors influencing the electricity demands of the consumers, so that electricity consumption forecasting accuracy is improved.
Description
Technical field
The invention belongs to Computer Applied Technology, data mining, electric power data analysis technical field, particularly relate to a kind of power consumer electricity demand forecasting method based on combination learning.
Background technology
Along with the development of intelligent grid, communication network technology and sensor technology, make in network system, to have accumulated a large amount of data, bring new challenge also to the analysis of electric power data simultaneously.Based on the power consumption of user in these data prediction future time sections, it is one of most important task during electric power data is analyzed.Such as, Hu Jie etc. compare the several method that electric load is conventional from aspects such as applicable elements, data mode, computation complexity and the scope of applications and analyze and sum up.Zhang Suxiang etc. propose parallel local weighted linear regression algorithm, study and solve mass data Short-Term Load Forecasting problem.Also some researcher adopts linear regression model (LRM) to predict the power consumption of user, in the electricity demand forecasting task in certain province, obtain good performance.King DS literary composition etc. proposes the parallel load Forecasting Methodology based on random forests algorithm.Huang Yuansheng etc. then adopt One-variable Linear Regression and trend ratio error transfer factor method, and based on 5 years in the past China's seasons monthly power consumption data, to future, the power consumption of China has in a short time made the prediction of science.Cai Jianbiao proposes the basic skills of the intelligent grid load prediction platform construction under cloud computing, and utilizes intelligentized multilevel coordination technology, improves the load prediction precision of electrical network at different levels.
But traditional electricity demand forecasting work is all that all users are integrally carried out modeling, and an overall prediction is done to power system power supply load, and the difference that have ignored between the feature of user self, user and incidence relation.
Summary of the invention
In order to solve the problem, the object of the present invention is to provide a kind of power consumer electricity demand forecasting method based on combination learning.
In order to achieve the above object, the power consumer electricity demand forecasting method based on combination learning provided by the invention comprises the following step carried out in order:
Step 1) input user power utilization record data, then utilize above-mentioned user power utilization record data construct user power utilization record matrix, utilize the electricity consumption total amount of user power utilization record matrix computations user every day afterwards, and be expressed as vector; What finally export all users uses electrographic recording matrix; Input user geographical location information, then represents above-mentioned user's geographical location information level, weighs afterwards to user's geographical location information similarity, finally export geographical location information and the geographical location information similarity matrix of all users;
Step 2) according to step 1) geographical location information of all users that obtains divides power supply area, builds user's overall electricity consumption behavior similarity matrix between region;
Step 3) according to step 1) in obtain by electrographic recording matrix and step 2) in user's overall electricity consumption behavior similarity matrix between the region that obtains, build the linear relationship model of user power utilization amount in date dimension, and linear prediction model is solved; Build the multiple linear prediction models in different power supply area, to go forward side by side between line linearity forecast model Knowledge delivery with mutually learn, the multiple linear prediction model of last combined optimization.
In step 1) in, described input user power utilization record data, then utilize above-mentioned user power utilization record data construct user power utilization record matrix, utilize the electricity consumption total amount of user power utilization record matrix computations user every day afterwards, and be expressed as vector; The method with electrographic recording matrix finally exporting all users is:
All users in user power utilization record data are represented as set:
U={u
1,u
2,…,u
N}
Wherein N represents the user's number comprised in data, u
irepresent i-th user;
Electrographic recording matrix is used in being built as by electrographic recording data of i-th user:
Wherein D represents with the number of days that electrographic recording comprises in data, and T represents the number of the uniform sampling point comprised with electrographic recording of each user every day,
represent the nonnegative real number matrix of the capable D row of T; Meanwhile, use
with
representing matrix U respectively
it capable and d row, i.e. user u
iall electrographic recording on every day t time point and d days with electrographic recording, and to use
representing matrix U
ithe capable d row of t on element; According to U
i, the electricity consumption total amount of every day that can calculate user in D days, and be expressed as vector x
i; x
iin the computing formula of each element as follows:
What finally export all users uses electrographic recording matrix:
In step 1) in, described input user geographical location information, then above-mentioned user's geographical location information level is represented, weigh user's geographical location information similarity afterwards, the method for the geographical location information and geographical location information similarity matrix that finally export all users is:
The geographical location information of i-th user is represented as structure:
Wherein
for the string representation of certain ingredient in place of abode,
by administrative unit, i.e. province, city, district, small towns, street, community order arrangement from big to small;
Any two neighbouring relations of geographic position in kJi administrative unit are expressed as:
Wherein
be 0/1 value, represent that whether two geographic position are adjacent, δ (,) be logical function, the value 1 when two character strings are identical, otherwise be 0, by Query Database, η () judges whether the administrative unit of two same levels is adjacent on geographic position;
Finally, export all users geographical location information matrix and geographical location information similarity matrix:
g
1,g
2,…,g
N,
In step 2) in, described divides power supply area, and the method building user's overall electricity consumption behavior similarity matrix between region is:
Step 2.1) S2.1 stage of all minimum administrative units in statistics service area;
Extract all user's geographical location information matrix g
1, g
2..., g
nin the string representation of lowermost level administrative unit
duplicate removal, obtains all minimum administrative unit Ω={ ω in service area
1, ω
2..., ω
m;
Step 2.2) the division of the power supply area scheme initialized S2.2 stage;
By step 2.1) in all minimum administrative unit Ω={ ω of obtaining
1, ω
2..., ω
mas the initial division of power supply area;
Step 2.3) calculate S2.3 stage of power supply area adjacency matrix;
According to power supply area ω
pin all geographic position of comprising, calculate power supply area ω
pstructured representation, formula is as follows:
If
Otherwise
Then, any two power supply area ω are calculated
pand ω
qsyntople, and build adjacency matrix A, formula is as follows:
a
pq=η(ω
p,ω
q);
Step 2.4) calculate S2.4 stage of the overall electricity consumption similarity of user in power supply area;
Adopt the average of the power consumption vector of all users in electricity consumption region to represent the power consumption of user's entirety in power supply area, formula is as follows:
Then, adopt the cosine similarity of vector to calculate the similarity of the overall electricity consumption behavior of user between zones of different, build similarity relation matrix S, formula is as follows:
s
pq=Cos(x(ω
p),x(ω
q));
Step 2.5) merge geographically adjacent and S2.5 stage of the power supply area that the overall need for electricity of user is similar;
According to adjacency matrix A, to any two adjacent power supply area ω
pand ω
qif, user's overall electricity consumption behavior similarity s
pqbe greater than threshold value φ, then merge two power supply areas;
Step 2.6) judge the S2.6 stage whether division of the power supply area restrains;
If the division of power supply area restrains, then carry out step 2.7), otherwise return step 2.3);
Step 2.7) S2.7 stage of Output rusults;
Export division of the power supply area result, this flow process so far terminates.
In step 3) in, described according to step 1) in obtain by electrographic recording matrix and step 2) in user's overall electricity consumption behavior similarity matrix between the region that obtains, build the linear relationship model of user power utilization amount in date dimension, and to the method that linear prediction model solves be;
Building the model of user power utilization amount in date dimension on single power supply area is the basis of electricity demand forecasting; For power supply area ω, user u
ipower consumption y in following one day
iwith power consumption vector x
ibetween the model of relation be:
Namely the power consumption of user in continuous D+1 days is linear correlation, therefore power consumption y
ipower consumption vector x can be passed through
ithe linear combination of middle different element is predicted; Wherein w is linear combination parameter, b
ifor error; To all users in power supply area ω, linear combination parameter w is shared, and error b
ichange with user; The then power consumption vector x of all users
ipower consumption vector matrix X can be combined into, power consumption y
ithe power consumption linear prediction model being combined into power consumption vector y, power supply area ω can be expressed as:
y=X
Tw+b;
Estimate linear combination parameter w and error b according to the power consumption vector matrix X of users all in power supply area ω and the power consumption vector y that marked, adopt least-squares algorithm to solve; Suppose that the variance of error b is limited, i.e. E [x
ib
i]=0, then linear combination parameter w and error b have close as follows shape separate:
w=(XX
T)
-1Xy,
b=y-X
T(XX
T)
-1Xy。
In step 3) in, the multiple linear prediction models in the different power supply area of described structure, to go forward side by side between line linearity forecast model Knowledge delivery with mutually learn, the method for the multiple linear prediction model of last combined optimization is:
Step 3.1) on each power supply area, build S3.1 stage of user power utilization amount linear prediction model respectively:
At power supply area ω
pon, according to above-mentioned formula, utilize power consumption matrix X
pthe power consumption vector y marked
pbuild user power utilization amount linear prediction model, then adopt least-squares algorithm to solve linear combination parameter w
pwith error b
p, as power supply area ω
pin benchmark model;
Step 3.2) according to overall electricity consumption behavior similarity matrix S to power supply area ω
pcarry out the S3.2 stage of data fusion:
To every other power supply area ω
q, according to other power supply areas ω
qwith power supply area ω
poverall need for electricity similarity s
pq, with probability s
pqrandomly draw other power supply areas ω
pin user data, and with power supply area ω
pin user data merge mutually and obtain model parameter X
p ∪ qwith power consumption vector y
p ∪ q;
Step 3.3) adopt fused data to upgrade the S3.3 stage of benchmark model:
Adopt least-squares algorithm, according to X
p ∪ qand y
p ∪ qsolve w
p ∪ qand b
p ∪ q, select all average forecasting errors to be less than the research on optimizing information fusion F of benchmark model
p={ w
p ∪ q| E [b
p ∪ q] <E [b
p], and therefrom select predicated error E [b
p ∪ q] minimum model parameter w
p ∪ qas power supply area ω
pin new benchmark model;
Step 3.4) judge whether the model on all regions upgrades the complete S3.4 stage:
If the model modification on all regions is complete, then carry out step 3.5), otherwise return step 3.2);
Step 3.5) evaluation algorithm S3.5 stage of whether restraining:
If algorithm convergence, then carry out step 3.6), otherwise return step 3.2);
Step 3.6) S3.6 stage of Output rusults:
Export power consumption vector y
p ∪ q, this flow process so far terminates.
The effect of the power consumer electricity demand forecasting method based on combination learning provided by the invention:
The present invention considers the geographic position of user and the electricity consumption behavior of user simultaneously, user is divided in different power supply areas, and the user in each region is not only adjacent on geographic position, and has similar need for electricity; For different power supply areas builds forecast model respectively to predict the power consumption of user, model had both considered unique need for electricity of user in each region, have also contemplated that contact between power supply area and affect the common factor of user power utilization demand, thus improve the accuracy of electricity demand forecasting.
Accompanying drawing explanation
Fig. 1 is the overall system structure schematic diagram of the power consumer electricity demand forecasting method based on combination learning provided by the invention;
Fig. 2 is the power consumer electricity demand forecasting method flow diagram based on combination learning provided by the invention.
Fig. 3 is the method flow diagram based on dividing and form user's overall electricity consumption behavior similarity matrix between region in the power consumer electricity demand forecasting method of combination learning to power supply area provided by the invention.
Fig. 4 is provided by the invention based on the electricity demand forecasting method flow diagram based on combination learning in the power consumer electricity demand forecasting method of combination learning.
Embodiment
Below in conjunction with the drawings and specific embodiments, the power consumer electricity demand forecasting method based on combination learning provided by the invention is described in detail.
As shown in Figure 1, the present invention mainly adopts data mining theories and method to analyze the user in electric power data, in order to ensure the normal operation of system, in concrete enforcement, require that the computer platform used is equipped with the internal memory being not less than 8G, core cpu number is not less than 4 and dominant frequency is not less than 64 bit manipulation systems of 2.6GHz, Windows7 and above version, and installs the Kinds of Essential Software environment such as oracle database, Java1.7 and above version, Matlab2011b and above version.
As shown in Figure 2, the power consumer electricity demand forecasting method based on combination learning provided by the invention comprises the following step performed in order:
Step 1) input user power utilization record data, then utilize above-mentioned user power utilization record data construct user power utilization record matrix, utilize the electricity consumption total amount of user power utilization record matrix computations user every day afterwards, and be expressed as vector; What finally export all users uses electrographic recording matrix; Input user geographical location information, then represents above-mentioned user's geographical location information level, weighs afterwards to user's geographical location information similarity, finally export geographical location information and the geographical location information similarity matrix of all users;
Step 2) according to step 1) geographical location information of all users that obtains divides power supply area, builds user's overall electricity consumption behavior similarity matrix between region;
According to step 1) in the geographical location information of all users that obtains divide power supply area demand fulfillment four pacing itemss: be 1) nonoverlapping between any two power supply areas, 2) all power supply areas can cover the place of abode of all users of electric power data, 3) user in identical power supply area has similar electricity consumption behavior, 4) in different power supply area, the overall electricity consumption behavior of user is different; Build user's overall electricity consumption behavior similarity matrix between region and mainly comprise two aspects: the 1) expression of the overall electricity consumption behavior of user in power supply area, 2) calculating of user's overall electricity consumption behavior similarity between region;
Step 3) according to step 1) in obtain by electrographic recording matrix and step 2) in user's overall electricity consumption behavior similarity matrix between the region that obtains, build the linear relationship model of user power utilization amount in date dimension, and linear prediction model is solved; Build the multiple linear prediction models in different power supply area, to go forward side by side between line linearity forecast model Knowledge delivery with mutually learn, the multiple linear prediction model of last combined optimization.
In step 1) in, described input user power utilization record data, then utilize above-mentioned user power utilization record data construct user power utilization record matrix, utilize the electricity consumption total amount of user power utilization record matrix computations user every day afterwards, and be expressed as vector; The method with electrographic recording matrix finally exporting all users is:
All users in user power utilization record data are represented as set:
U={u
1,u
2,…,u
N}
Wherein N represents the user's number comprised in data, u
irepresent i-th user;
Electrographic recording matrix is used in being built as by electrographic recording data of i-th user:
Wherein D represents with the number of days that electrographic recording comprises in data, and T represents the number of the uniform sampling point comprised with electrographic recording of each user every day,
represent the nonnegative real number matrix of the capable D row of T; Meanwhile, use
with
representing matrix U respectively
it capable and d row, i.e. user u
iall electrographic recording on every day t time point and d days with electrographic recording, and to use
representing matrix U
ithe capable d row of t on element; According to U
i, the electricity consumption total amount of every day that can calculate user in D days, and be expressed as vector x
i; x
iin the computing formula of each element as follows:
What finally export all users uses electrographic recording matrix:
In step 1) in, described input user geographical location information, then above-mentioned user's geographical location information level is represented, weigh user's geographical location information similarity afterwards, the method for the geographical location information and geographical location information similarity matrix that finally export all users is:
The geographical location information of i-th user is represented as structure:
Wherein
for the string representation of certain ingredient in place of abode,
by administrative unit (province, city, district, small towns, street, community etc.) order arrangement from big to small;
Any two neighbouring relations of geographic position in kJi administrative unit are expressed as:
Wherein
be 0/1 value, represent that whether two geographic position are adjacent, δ (,) be logical function, the value 1 when two character strings are identical, otherwise be 0, by Query Database, η () judges whether the administrative unit of two same levels is adjacent on geographic position;
Finally, export all users geographical location information matrix and geographical location information similarity matrix:
g
1,g
2,…,g
N,
As shown in Figure 3, in step 2) in, described divides power supply area, and the method building user's overall electricity consumption behavior similarity matrix between region is:
Step 2.1) S2.1 stage of all minimum administrative units in statistics service area;
Extract all user's geographical location information matrix g
1, g
2..., g
nin the string representation of lowermost level administrative unit
duplicate removal, obtains all minimum administrative unit Ω={ ω in service area
1, ω
2..., ω
m;
Step 2.2) the division of the power supply area scheme initialized S2.2 stage;
By step 2.1) in all minimum administrative unit Ω={ ω of obtaining
1, ω
2..., ω
mas the initial division of power supply area;
Step 2.3) calculate S2.3 stage of power supply area adjacency matrix;
According to power supply area ω
pin all geographic position of comprising, calculate power supply area ω
pstructured representation, formula is as follows:
If
Otherwise
Then, any two power supply area ω are calculated
pand ω
qsyntople, and build adjacency matrix A, formula is as follows:
a
pq=η(ω
p,ω
q);
Step 2.4) calculate S2.4 stage of the overall electricity consumption similarity of user in power supply area;
Adopt the average of the power consumption vector of all users in electricity consumption region to represent the power consumption of user's entirety in power supply area, formula is as follows:
Then, adopt the cosine similarity of vector to calculate the similarity of the overall electricity consumption behavior of user between zones of different, build similarity relation matrix S, formula is as follows:
s
pq=Cos(x(ω
p),x(ω
q));
Step 2.5) merge geographically adjacent and S2.5 stage of the power supply area that the overall need for electricity of user is similar;
According to adjacency matrix A, to any two adjacent power supply area ω
pand ω
qif, user's overall electricity consumption behavior similarity s
pqbe greater than threshold value φ, then merge two power supply areas;
Step 2.6) judge the S2.6 stage whether division of the power supply area restrains;
If the division of power supply area restrains, then carry out step 2.7), otherwise return step 2.3);
Step 2.7) S2.7 stage of Output rusults;
Export division of the power supply area result, this flow process so far terminates.
In step 3) in, described according to step 1) in obtain by electrographic recording matrix and step 2) in user's overall electricity consumption behavior similarity matrix between the region that obtains, build the linear relationship model of user power utilization amount in date dimension, and to the method that linear prediction model solves be;
Building the model of user power utilization amount in date dimension on single power supply area is the basis of electricity demand forecasting; For power supply area ω, user u
ipower consumption y in following one day
iwith power consumption vector x
ibetween the model of relation be:
Namely the power consumption of user in continuous D+1 days is linear correlation, therefore power consumption y
ipower consumption vector x can be passed through
ithe linear combination of middle different element is predicted; Wherein w is linear combination parameter, b
ifor error; To all users in power supply area ω, linear combination parameter w is shared, and error b
ichange with user; The then power consumption vector x of all users
ipower consumption vector matrix X can be combined into, power consumption y
ithe power consumption linear prediction model being combined into power consumption vector y, power supply area ω can be expressed as:
y=X
Tw+b;
Estimate linear combination parameter w and error b according to the power consumption vector matrix X of users all in power supply area ω and the power consumption vector y that marked, least-squares algorithm can be adopted solve; The present invention supposes that the variance of error b is limited, i.e. E [x
ib
i]=0, then linear combination parameter w and error b have close as follows shape separate:
w=(XX
T)
-1Xy,
b=y-X
T(XX
T)
-1Xy。
In step 3) in, the multiple linear prediction models in the different power supply area of described structure, to go forward side by side between line linearity forecast model Knowledge delivery with mutually learn, the method for the multiple linear prediction model of last combined optimization is:
In the application of actual electrical force data, because the place of abode of user is often more concentrated, many power supply areas comprising less user are easy to over-fitting occurs in solving, and cause the accuracy rate of electricity demand forecasting lower; Therefore, as shown in Figure 4, in step 3) in, design is as follows based on the derivation algorithm of combination learning, by data sharing, merge the factor relevant to user place power supply area and the factor irrelevant with user place power supply area, optimize the linear prediction model on different power supply area simultaneously; Concrete steps are as follows:
Step 3.1) on each power supply area, build S3.1 stage of user power utilization amount linear prediction model respectively:
At power supply area ω
pon, according to above-mentioned formula, utilize power consumption matrix X
pthe power consumption vector y marked
pbuild user power utilization amount linear prediction model, then adopt least-squares algorithm to solve linear combination parameter w
pwith error b
p, as power supply area ω
pin benchmark model;
Step 3.2) according to overall electricity consumption behavior similarity matrix S to power supply area ω
pcarry out the S3.2 stage of data fusion:
To every other power supply area ω
q, according to other power supply areas ω
qwith power supply area ω
poverall need for electricity similarity s
pq, with probability s
pqrandomly draw other power supply areas ω
pin user data, and with power supply area ω
pin user data merge mutually and obtain model parameter X
p ∪ qwith power consumption vector y
p ∪ q;
Step 3.3) adopt fused data to upgrade the S3.3 stage of benchmark model:
Adopt least-squares algorithm, according to X
p ∪ qand y
p ∪ qsolve w
p ∪ qand b
p ∪ q, select all average forecasting errors to be less than the research on optimizing information fusion F of benchmark model
p={ w
p ∪ q| E [b
p ∪ q] <E [b
p], and therefrom select predicated error E [b
p ∪ q] minimum model parameter w
p ∪ qas power supply area ω
pin new benchmark model;
Step 3.4) judge whether the model on all regions upgrades the complete S3.4 stage:
If the model modification on all regions is complete, then carry out step 3.5), otherwise return step 3.2);
Step 3.5) evaluation algorithm S3.5 stage of whether restraining:
If algorithm convergence, then carry out step 3.6), otherwise return step 3.2);
Step 3.6) S3.6 stage of Output rusults:
Export power consumption vector y
p ∪ q, this flow process so far terminates.
The present invention adopts the dwelling places information of user power utilization record data in electric power data and user, according to the relation on the distribution of user on diverse geographic location and diverse geographic location between user power utilization amount, the bulk supply scope of electric system is divided into different power supply areas, makes the user in each power supply area have similar need for electricity.Then, in each power supply area, linear regression algorithm is adopted to build user power utilization amount linear prediction model respectively.Adopt multitask combination learning method, optimize the linear prediction model in different power supply area simultaneously, make model can consider to affect user power utilization behavior and the factor relevant to power supply area, also can consider the factor irrelevant with power supply area, improve the accuracy rate of electricity demand forecasting.Based on this, power department deeply can understand the overall power consumption of user in future time in different power supply area, thus organizes electrical production, reasonable arrangement power supply facilities targetedly.
It is emphasized that; embodiment of the present invention is illustrative; instead of it is determinate; therefore the present invention is not limited to the embodiment described in embodiment; every other embodiments drawn by those skilled in the art's technical scheme according to the present invention, belong to the scope of protection of the invention equally.
Claims (6)
1. based on a power consumer electricity demand forecasting method for combination learning, it is characterized in that: the described power consumer electricity demand forecasting method based on combination learning comprises the following step carried out in order:
Step 1) input user power utilization record data, then utilize above-mentioned user power utilization record data construct user power utilization record matrix, utilize the electricity consumption total amount of user power utilization record matrix computations user every day afterwards, and be expressed as vector; What finally export all users uses electrographic recording matrix; Input user geographical location information, then represents above-mentioned user's geographical location information level, weighs afterwards to user's geographical location information similarity, finally export geographical location information and the geographical location information similarity matrix of all users;
Step 2) according to step 1) geographical location information of all users that obtains divides power supply area, builds user's overall electricity consumption behavior similarity matrix between region;
Step 3) according to step 1) in obtain by electrographic recording matrix and step 2) in user's overall electricity consumption behavior similarity matrix between the region that obtains, build the linear relationship model of user power utilization amount in date dimension, and linear prediction model is solved; Build the multiple linear prediction models in different power supply area, to go forward side by side between line linearity forecast model Knowledge delivery with mutually learn, the multiple linear prediction model of last combined optimization.
2. the power consumer electricity demand forecasting method based on combination learning according to claim 1, it is characterized in that: in step 1) in, described input user power utilization record data, then above-mentioned user power utilization record data construct user power utilization record matrix is utilized, utilize the electricity consumption total amount of user power utilization record matrix computations user every day afterwards, and be expressed as vector; The method with electrographic recording matrix finally exporting all users is:
All users in user power utilization record data are represented as set:
U={u
1,u
2,…,u
N}
Wherein N represents the user's number comprised in data, u
irepresent i-th user;
Electrographic recording matrix is used in being built as by electrographic recording data of i-th user:
Wherein D represents with the number of days that electrographic recording comprises in data, and T represents the number of the uniform sampling point comprised with electrographic recording of each user every day,
represent the nonnegative real number matrix of the capable D row of T; Meanwhile, use
with
representing matrix U respectively
it capable and d row, i.e. user u
iall electrographic recording on every day t time point and d days with electrographic recording, and to use
representing matrix U
ithe capable d row of t on element; According to U
i, the electricity consumption total amount of every day that can calculate user in D days, and be expressed as vector x
i; x
iin the computing formula of each element as follows:
What finally export all users uses electrographic recording matrix:
3. the power consumer electricity demand forecasting method based on combination learning according to claim 1, it is characterized in that: in step 1) in, described input user geographical location information, then above-mentioned user's geographical location information level is represented, weigh user's geographical location information similarity afterwards, the method for the geographical location information and geographical location information similarity matrix that finally export all users is:
The geographical location information of i-th user is represented as structure:
Wherein
for the string representation of certain ingredient in place of abode,
by administrative unit, i.e. province, city, district, small towns, street, community order arrangement from big to small;
Any two neighbouring relations of geographic position in kJi administrative unit are expressed as:
Wherein
be 0/1 value, represent that whether two geographic position are adjacent, δ (,) be logical function, the value 1 when two character strings are identical, otherwise be 0, by Query Database, η () judges whether the administrative unit of two same levels is adjacent on geographic position;
Finally, export all users geographical location information matrix and geographical location information similarity matrix:
g
1,g
2,…,g
N,
4. the power consumer electricity demand forecasting method based on combination learning according to claim 1, is characterized in that: in step 2) in, described divides power supply area, and the method building user's overall electricity consumption behavior similarity matrix between region is:
Step 2.1) S2.1 stage of all minimum administrative units in statistics service area;
Extract all user's geographical location information matrix g
1, g
2..., g
nin the string representation of lowermost level administrative unit
duplicate removal, obtains all minimum administrative unit Ω={ ω in service area
1, ω
2..., ω
m;
Step 2.2) the division of the power supply area scheme initialized S2.2 stage;
By step 2.1) in all minimum administrative unit Ω={ ω of obtaining
1, ω
2..., ω
mas the initial division of power supply area;
Step 2.3) calculate S2.3 stage of power supply area adjacency matrix;
According to power supply area ω
pin all geographic position of comprising, calculate power supply area ω
pstructured representation, formula is as follows:
If
Otherwise
Then, any two power supply area ω are calculated
pand ω
qsyntople, and build adjacency matrix A, formula is as follows:
a
pq=η(ω
p,ω
q);
Step 2.4) calculate S2.4 stage of the overall electricity consumption similarity of user in power supply area;
Adopt the average of the power consumption vector of all users in electricity consumption region to represent the power consumption of user's entirety in power supply area, formula is as follows:
Then, adopt the cosine similarity of vector to calculate the similarity of the overall electricity consumption behavior of user between zones of different, build similarity relation matrix S, formula is as follows:
s
pq=Cos(x(ω
p),x(ω
q));
Step 2.5) merge geographically adjacent and S2.5 stage of the power supply area that the overall need for electricity of user is similar;
According to adjacency matrix A, to any two adjacent power supply area ω
pand ω
qif, user's overall electricity consumption behavior similarity s
pqbe greater than threshold value φ, then merge two power supply areas;
Step 2.6) judge the S2.6 stage whether division of the power supply area restrains;
If the division of power supply area restrains, then carry out step 2.7), otherwise return step 2.3);
Step 2.7) S2.7 stage of Output rusults;
Export division of the power supply area result, this flow process so far terminates.
5. the power consumer electricity demand forecasting method based on combination learning according to claim 1, it is characterized in that: in step 3) in, described according to step 1) in obtain by electrographic recording matrix and step 2) in user's overall electricity consumption behavior similarity matrix between the region that obtains, build the linear relationship model of user power utilization amount in date dimension, and to the method that linear prediction model solves be;
Building the model of user power utilization amount in date dimension on single power supply area is the basis of electricity demand forecasting; For power supply area ω, user u
ipower consumption y in following one day
iwith power consumption vector x
ibetween the model of relation be:
Namely the power consumption of user in continuous D+1 days is linear correlation, therefore power consumption y
ipower consumption vector x can be passed through
ithe linear combination of middle different element is predicted; Wherein w is linear combination parameter, b
ifor error; To all users in power supply area ω, linear combination parameter w is shared, and error b
ichange with user; The then power consumption vector x of all users
ipower consumption vector matrix X can be combined into, power consumption y
ithe power consumption linear prediction model being combined into power consumption vector y, power supply area ω can be expressed as:
y=X
Tw+b;
Estimate linear combination parameter w and error b according to the power consumption vector matrix X of users all in power supply area ω and the power consumption vector y that marked, adopt least-squares algorithm to solve; Suppose that the variance of error b is limited, i.e. E [x
ib
i]=0, then linear combination parameter w and error b have close as follows shape separate:
w=(XX
T)
-1Xy,
b=y-X
T(XX
T)
-1Xy。
6. the power consumer electricity demand forecasting method based on combination learning according to claim 1, it is characterized in that: in step 3) in, multiple linear prediction models in the different power supply area of described structure, to go forward side by side between line linearity forecast model Knowledge delivery with mutually learn, the method for the multiple linear prediction model of last combined optimization is:
Step 3.1) on each power supply area, build S3.1 stage of user power utilization amount linear prediction model respectively:
At power supply area ω
pon, according to above-mentioned formula, utilize power consumption matrix X
pthe power consumption vector y marked
pbuild user power utilization amount linear prediction model, then adopt least-squares algorithm to solve linear combination parameter w
pwith error b
p, as power supply area ω
pin benchmark model;
Step 3.2) according to overall electricity consumption behavior similarity matrix S to power supply area ω
pcarry out the S3.2 stage of data fusion:
To every other power supply area ω
q, according to other power supply areas ω
qwith power supply area ω
poverall need for electricity similarity s
pq, with probability s
pqrandomly draw other power supply areas ω
pin user data, and with power supply area ω
pin user data merge mutually and obtain model parameter X
p ∪ qwith power consumption vector y
p ∪ q;
Step 3.3) adopt fused data to upgrade the S3.3 stage of benchmark model:
Adopt least-squares algorithm, according to X
p ∪ qand y
p ∪ qsolve w
p ∪ qand b
p ∪ q, select all average forecasting errors to be less than the research on optimizing information fusion F of benchmark model
p={ w
p ∪ q| E [b
p ∪ q] <E [b
p], and therefrom select predicated error E [b
p ∪ q] minimum model parameter w
p ∪ qas power supply area ω
pin new benchmark model;
Step 3.4) judge whether the model on all regions upgrades the complete S3.4 stage:
If the model modification on all regions is complete, then carry out step 3.5), otherwise return step 3.2);
Step 3.5) evaluation algorithm S3.5 stage of whether restraining:
If algorithm convergence, then carry out step 3.6), otherwise return step 3.2);
Step 3.6) S3.6 stage of Output rusults:
Export power consumption vector y
p ∪ q, this flow process so far terminates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510801660.4A CN105335800A (en) | 2015-11-19 | 2015-11-19 | Method for forecasting electricity consumption of power consumers based on joint learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510801660.4A CN105335800A (en) | 2015-11-19 | 2015-11-19 | Method for forecasting electricity consumption of power consumers based on joint learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105335800A true CN105335800A (en) | 2016-02-17 |
Family
ID=55286315
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510801660.4A Pending CN105335800A (en) | 2015-11-19 | 2015-11-19 | Method for forecasting electricity consumption of power consumers based on joint learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105335800A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108062720A (en) * | 2017-12-05 | 2018-05-22 | 贵州电网有限责任公司电力科学研究院 | A kind of load forecasting method based on similar day selection and random forests algorithm |
CN109740790A (en) * | 2018-11-28 | 2019-05-10 | 国网天津市电力公司 | A kind of user power consumption prediction technique extracted based on temporal aspect |
CN110443375A (en) * | 2019-08-16 | 2019-11-12 | 深圳前海微众银行股份有限公司 | A kind of federation's learning method and device |
CN111915338A (en) * | 2019-05-08 | 2020-11-10 | 腾讯科技(深圳)有限公司 | Target variable prediction method, device and storage medium |
CN113111578A (en) * | 2021-04-01 | 2021-07-13 | 上海晨翘智能科技有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
CN116109016A (en) * | 2023-04-11 | 2023-05-12 | 昆仑数智科技有限责任公司 | Energy consumption prediction model training and energy consumption prediction method and device |
CN117078112A (en) * | 2023-10-16 | 2023-11-17 | 云境商务智能研究院南京有限公司 | Energy consumption detection method and data analysis system applied to enterprise abnormal electricity management |
-
2015
- 2015-11-19 CN CN201510801660.4A patent/CN105335800A/en active Pending
Non-Patent Citations (2)
Title |
---|
刘春兰等: "多宿点多区域纠缠关联干扰下的电力调度自动化优化控制", 《科技通报》 * |
马坤隆: "基于大数据的分布式短期负荷预测方法", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108062720A (en) * | 2017-12-05 | 2018-05-22 | 贵州电网有限责任公司电力科学研究院 | A kind of load forecasting method based on similar day selection and random forests algorithm |
CN109740790A (en) * | 2018-11-28 | 2019-05-10 | 国网天津市电力公司 | A kind of user power consumption prediction technique extracted based on temporal aspect |
CN111915338A (en) * | 2019-05-08 | 2020-11-10 | 腾讯科技(深圳)有限公司 | Target variable prediction method, device and storage medium |
CN111915338B (en) * | 2019-05-08 | 2024-05-14 | 腾讯科技(深圳)有限公司 | Target variable prediction method, device and storage medium |
CN110443375A (en) * | 2019-08-16 | 2019-11-12 | 深圳前海微众银行股份有限公司 | A kind of federation's learning method and device |
CN113111578A (en) * | 2021-04-01 | 2021-07-13 | 上海晨翘智能科技有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
CN113111578B (en) * | 2021-04-01 | 2023-04-07 | 上海晨翘智能科技有限公司 | Power load prediction method, power load prediction device, computer equipment and storage medium |
CN116109016A (en) * | 2023-04-11 | 2023-05-12 | 昆仑数智科技有限责任公司 | Energy consumption prediction model training and energy consumption prediction method and device |
CN117078112A (en) * | 2023-10-16 | 2023-11-17 | 云境商务智能研究院南京有限公司 | Energy consumption detection method and data analysis system applied to enterprise abnormal electricity management |
CN117078112B (en) * | 2023-10-16 | 2024-01-02 | 云境商务智能研究院南京有限公司 | Energy consumption detection method and data analysis system applied to enterprise abnormal electricity management |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105335800A (en) | Method for forecasting electricity consumption of power consumers based on joint learning | |
Ye et al. | A data-driven bottom-up approach for spatial and temporal electric load forecasting | |
Xiao et al. | A combined model based on data pre-analysis and weight coefficients optimization for electrical load forecasting | |
Huang et al. | Gated spatial-temporal graph neural network based short-term load forecasting for wide-area multiple buses | |
CN109887282A (en) | A kind of road network traffic flow prediction technique based on level timing diagram convolutional network | |
CN108446794A (en) | One kind being based on multiple convolutional neural networks combination framework deep learning prediction techniques | |
Reis et al. | Water supply reservoir operation by combined genetic algorithm–linear programming (GA-LP) approach | |
CN106022614A (en) | Data mining method of neural network based on nearest neighbor clustering | |
CN104391919A (en) | System and method for realizing geographic visualization of smart power grid | |
Zheng et al. | Parallel computational building-chain model for rapid urban-scale energy simulation | |
CN115954933A (en) | Power grid digital twin system construction method and device, electronic equipment and storage medium | |
CN105809349A (en) | Scheduling method considering incoming water correlation cascade hydropower stations | |
Sangireddy et al. | Development of a surrogate model by extracting top characteristic feature vectors for building energy prediction | |
Tsoukalas et al. | A surrogate based optimization approach for the development of uncertainty-aware reservoir operational rules: the case of nestos hydrosystem | |
Yi et al. | Intelligent prediction of transmission line project cost based on least squares support vector machine optimized by particle swarm optimization | |
CN103839113B (en) | Microscopic simulation method based on house selecting models of house-renting selectors and house selecting models of house-purchasing selectors | |
Wang et al. | Predicting electric vehicle charging demand using a heterogeneous spatio-temporal graph convolutional network | |
CN113591368A (en) | Comprehensive energy system multi-energy load prediction method and system | |
Zhao et al. | Short-term microgrid load probability density forecasting method based on k-means-deep learning quantile regression | |
Ye et al. | A data driven multi-state model for distribution system flexible planning utilizing hierarchical parallel computing | |
Zhang et al. | A hybrid electric vehicle load classification and forecasting approach based on GBDT algorithm and temporal convolutional network | |
CN108205713A (en) | A kind of region wind power prediction error distribution determination method and device | |
Xiaoxun et al. | Research on wind speed behavior prediction method based on multi-feature and multi-scale integrated learning | |
Zhao et al. | LandSys II: Agent-based land use–forecast model with artificial neural networks and multiagent model | |
Li et al. | Optimal design and operation method of integrated energy system based on stochastic scenario iteration and energy storage attribute evaluation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160217 |
|
RJ01 | Rejection of invention patent application after publication |