CN105809286B - Incremental SVR load prediction method based on representative data reconstruction - Google Patents

Abstract

The invention discloses an incremental SVR load prediction method based on representative data reconstruction, which comprises the following steps: acquiring power load data; obtaining multi-input single-output mode data by utilizing a phase space reconstruction principle; establishing a support vector regression model by using the obtained mode data and a particle swarm algorithm; acquiring newly-added power load prediction data in real time; updating the optimal representative data subset by using an incremental learning algorithm; updating model parameters by using a nested particle swarm method; establishing a support vector regression model by using the updated model parameters and the optimal representative data subset; and determining an incremental load forecast and outputting the incremental load forecast value. The support vector of the support vector regression is applied to the knowledge understanding research of the mass data, the provided method can realize the reconstruction of the representative data caused by the newly added data, effectively solves the problems of high complexity and difficulty in extracting knowledge of the mass data, realizes the updating of the model parameters in a nested manner, and provides a reference basis for the planning and the operation of the power system.

Description

Incremental SVR load prediction method based on representative data reconstruction

Technical Field

The invention relates to the field of computer data rapid analysis, in particular to an incremental SVR load prediction method based on representative data reconstruction.

Background

Since electric energy is an energy source which is difficult to store in large quantities, the production, transmission, distribution and consumption of electric energy must be carried out at the same moment, which determines the premise that the result of the electric load prediction is safe, stable and economical operation of the electric power system. At present, typical load prediction methods mainly include statistical methods, neural network methods, gray methods and the like based on parameter assumptions, and these methods usually can only train a model under given data, but cannot extract representative data from a large amount of data, because only a small amount of representative data in a large amount of training data is determined, the knowledge which is artificially understood can be generated.

The support vector regression method is generated by aiming at sparse extraction of training data, has excellent prediction performance, and can extract a small amount of representative data (called support vectors). However, with the rapid development of the smart grid, the power system may continuously obtain new data in batches, which requires updating not only the representative data, but also the existing prediction method to realize incremental load prediction. However, the current support vector regression method needs to perform model selection and model training again, which may increase complexity of model training and storage and further affect learning accuracy of the model. There is a great need for those skilled in the art to solve the corresponding technical problems.

Disclosure of Invention

The invention aims to at least solve the technical problems in the prior art, and particularly creatively provides an incremental SVR load prediction method based on representative data reconstruction.

In order to achieve the above object, the present invention provides an incremental SVR load prediction method based on representative data reconstruction, including:

s1, analyzing historical power load data by using phase space reconstruction to obtain the embedding dimension and time delay of the power load data and obtain multi-input-single-output mode data;

s2, modeling historical power load data by using support vector regression by using an optimal training subset method and a particle swarm method to obtain power load representative data and support vector regression model parameters, and acquiring a power load prediction result at the moment according to the representative data and the support vector regression model parameters;

s3, acquiring new power load data, and updating the representative data by adopting a representative data reconstruction method according to the multi-input single-output mode data; and re-executes S3.

In the incremental SVR load prediction method based on representative data reconstruction, preferably, the S2 includes:

the initial power load representative data set is selected from the following formula:

wherein RDS is the initial representative data set, A is the initial acquisitionTaken power load data, y^*Is x^*Corresponding power load data output value, RDS ═ RDS { (x { } { [ RDS { } { [ U ] } { [ X ] } { [ MEANS ] of electrical load data^*,y^*) Until the elements in the RDS reach a set size k, i is the subscript of the elements in the set A, j is the subscript of the elements in the set RDS, i is 1., | A |, j is 1., | RDS |, wherein | A | is the number of the elements in the set A, and | RDS | is the number of the elements in the set RDS;

the optimal size k of the initial power load representative data set is determined by the following approximate convex frame:

wherein N is the number of elements in A, MAPE (k, A) is the mean absolute percentage error of support vector regression under RDS based on the size k, and lambda is the balance coefficient between the model complexity and the prediction precision, wherein N is⁺Representing a positive integer.

In the incremental SVR load prediction method based on representative data reconstruction, preferably, the S2 further includes:

s2-1: determining three different sizes k by using initial power load representative data set selection method with power load data at recent time point as initial point₁,k₂,k₃A representative data set of (a);

s2-2: training the SVR by using a particle swarm method aiming at the three power load representative data sets of S2-1 to obtain parameters of the three SVR methods, and performing nonlinear prediction;

s2-3: for a given approximate convex frame, update k by 0.618 method₁,k₂,k₃If max (k)₁,k₂,k₃)-min(k₁,k₂,k₃) If the SVR is less than or equal to 3, obtaining an optimal SVR and a power load representative data set, and terminating; if max (k)₁,k₂,k₃)-min(k₁,k₂,k₃) > 3, return to and execute S2-1.

Preferably, the incremental SVR load prediction method based on representative data reconstruction in S3 includes:

updating the representative data by using a representative data reconstruction method, wherein the updated power load representative data subset is

A_V∪B_V∪N_m

Wherein A is_VRepresenting a data set for the electrical load of the raw data A, B_VFor the power load representative dataset based on the existing SVR model and the new data B:

B_V＝{(x_i,y_i)|(x_i,y_i)∈B,|y_i-p_i|＞σ}

wherein p is_iFor existing SVR model pair input data x_iThe predicted value of (a) is the standard deviation of error of the newly added data set predicted by the existing SVR model, N_mIs A_V∪B_VThe union of the m nearest neighbor data of each point in the data set;

representing data subset A based on updated power loads_V∪B_V∪N_mUpdating model parameters by using a nested particle swarm method, and representing data subset A for the power load_V∪B_V∪N_mGenerated ith initial particle p₂(i) Comprises the following steps:

p₂(i)＝p₁(i)+λ_i×[p_{best_1}-p₁(i)]

wherein p is₁(i) For random particles generated in the last parameter space, p_{best_1}Is the previous data A_VIs set to [ p ] according to the optimum parameter_{best_1}-p₁(i)]Global contraction factor, λ, for the last optimum parameter_iRandom puncturing weights distributed to obey U (0, 1);

and establishing a support vector regression model to obtain updated power load representative data and model parameters of the SVR, and obtaining a power load prediction result at the moment.

The incremental SVR load prediction method based on representative data reconstruction preferably further includes: each time new power load data is obtained, S3 is repeatedly executed to update the power load representative data subset and the SVR parameters, and the power load data is continuously updated iteratively.

In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:

1. the phase space reconstruction technology adopted by the invention can obtain phase space load points with physical significance;

2. the method adopts a nested particle swarm method to realize the updating of the SVR model parameter setting;

3. according to the method, the representative data is updated and extracted by using the sparsity of SVR support vector regression and adopting a representative data subset reconstruction method;

4. the invention uses a representative data reconstruction method, and has the following significance: the modeling complexity of the SVR load prediction model is small at the initial stage of load data acquisition and when the data volume is small, the load data volume is accumulated continuously along with the continuous acquisition of the load data, and the learning difficulty is increased continuously; the invention can continuously update the representative data set and the SVR model parameters and realize incremental learning, so the load prediction has lower complexity and higher precision.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

FIG. 1 is a general flow diagram of the prediction method of the present invention.

Detailed Description

Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.

In the description of the present invention, it is to be understood that the terms "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on those shown in the drawings, and are used merely for convenience of description and for simplicity of description, and do not indicate or imply that the referenced devices or elements must have a particular orientation, be constructed in a particular orientation, and be operated, and thus, are not to be construed as limiting the present invention.

In the description of the present invention, unless otherwise specified and limited, it is to be noted that the terms "mounted," "connected," and "connected" are to be interpreted broadly, and may be, for example, a mechanical connection or an electrical connection, a communication between two elements, a direct connection, or an indirect connection via an intermediate medium, and specific meanings of the terms may be understood by those skilled in the art according to specific situations.

As shown in fig. 1, the present invention provides an incremental Support Vector Regression (SVR) load prediction method based on representative data reconstruction, which can extract an optimal representative data subset from historical data, and can continuously update the representative data, update SVR model selection, reduce the training and storage complexity of the prediction method, and obtain understandable knowledge through the reconstructed representative data to implement more accurate and understandable prediction, and the prediction method includes the following steps:

step 1: and analyzing historical power load data by using a phase space reconstruction technology to obtain the embedding dimension and time delay of the power load data and obtain multi-input-single-output mode data.

Step 2: and modeling historical power load data by using a Support Vector Regression (SVR) by using an optimal training subset method and a particle swarm method to obtain representative data and model parameters of the SVR and obtain a final prediction result of the power load at the moment.

The optimal training subset method selects a subset of the training data by utilizing the sparsity of support vector regression, can minimize elements in the subset, and can cover approximately all information of the training data.

The particle swarm method can quickly screen out a proper parameter of the model by utilizing a heuristic search strategy.

The initial representative dataset is chosen by the idea of "orthogonal design":

wherein RDS is an initial representative data set, A is initially acquired power load data, y^*Is x^*Corresponding power load data output value, RDS ═ RDS { (x { } { [ RDS { } { [ U ] } { [ X ] } { [ MEANS ] of electrical load data^*,y^*) Until the elements in the RDS reach a set size k, i is the subscript of the elements in the set A, j is the subscript of the elements in the set RDS, i is 1., | A |, j is 1., | RDS |, wherein | A | is the number of the elements in the set A, and | RDS | is the number of the elements in the set RDS;

the optimal size k of the initial power load representative data set is determined by the following approximate convex frame:

wherein N is the number of elements in A, MAPE (k, A) is the average absolute percentage error of SVR under RDS based on the size k, lambda is the balance coefficient between the model complexity and the prediction precision, and N is⁺Representing a positive integer.

Substep 2-1: determining three different sizes k by using the initial representative data set selection method with the latest time point data as the initial point₁,k₂,k₃A representative data set of (a);

substep 2-2: and (3) aiming at the three subsets of the substep 2-1, training the SVR by using a particle swarm optimization method to obtain parameters of the three SVR methods, wherein the SVR is a multilayer feedforward neural network, can approach any nonlinear continuous function with any precision, has strong fault tolerance and fast processing speed, and is suitable for nonlinear prediction.

Substeps 2-3: for a given approximate convex frame, update k by 0.618 method₁,k₂,k₃If max (k)₁,k₂,k₃)-min(k₁,k₂,k₃) Obtaining an optimal SVR and a representative data set when the SVR is less than or equal to 3, and terminating; if max (k)₁,k₂,k₃)-min(k₁,k₂,k₃) > 3, go back to and perform substep 2-1.

And step 3: when new data enters the system, the data in the multi-input-single-output mode is obtained by using the step 1, the representative data is updated by using a representative data reconstruction method, and the updated representative data subset is

A_V∪B_V∪N_m

Wherein A is_VA representative data set of the original data A, B_VFor a representative dataset based on the existing SVR model and the new data B:

B_V＝{(x_i,y_i)|(x_i,y_i)∈B,|y_i-p_i|＞σ}

wherein p is_iFor existing SVR model pair input data x_iThe predicted value of (a) is the standard deviation of error of the newly added data set predicted by the existing SVR model, N_mIs A_V∪B_VThe union of the m nearest neighbor data for each point in the set.

Based on the updated representative data subset A_V∪B_V∪N_mUpdating model parameters using a nested particle swarm approach, which is applied to the representative data subset A_V∪B_V∪N_mGenerated ith initial particle p₂(i) Comprises the following steps:

p₂(i)＝p₁(i)+λ_i×[p_{best_1}-p₁(i)]

wherein p is₁(i) For random particles generated in the last parameter space, p_{best_1}Is the previous data A_VIs set to [ p ] according to the optimum parameter_{best_1}-p₁(i)]Global contraction factor, λ, for the last optimum parameter_iTo obey the random puncturing weights of the U (0,1) distribution.

And establishing a Support Vector Regression (SVR) model to obtain updated representative data and model parameters of the SVR and obtain a final prediction result of the power load at the moment.

And 4, step 4: when new data enters the system, step 3 is performed.

The support vector of the support vector regression is applied to the knowledge understanding research of the mass data, the provided method can realize the reconstruction of the representative data caused by the newly added data, effectively solves the problems of high complexity and difficulty in extracting knowledge of the mass data, realizes the updating of the model parameters in a nested manner, and provides a reference basis for the planning and the operation of the power system.

In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims (2)

1. An incremental SVR load prediction method based on representative data reconstruction is characterized by comprising the following steps:

s1, analyzing historical power load data by using phase space reconstruction to obtain the embedding dimension and time delay of the power load data and obtain multi-input-single-output mode data;

s2, modeling historical power load data by using support vector regression by using an optimal training subset method and a particle swarm method to obtain power load representative data and support vector regression model parameters, and acquiring a power load prediction result at the moment according to the representative data and the support vector regression model parameters;

the S2 includes:

the initial power load representative data set is selected from the following formula:

wherein RDS is the initial power load representative data set, A is the initially acquired power load data, y^*Is x^*Corresponding power load data output value, RDS ═ RDS { (x { } { [ RDS { } { [ U ] } { [ X ] } { [ MEANS ] of electrical load data^*,y^*) Until the elements in the RDS reach a set size k, i is the subscript of the elements in the set A, j is the subscript of the elements in the set RDS, i is 1., | A |, j is 1., | RDS |, wherein | A | is the number of the elements in the set A, and | RDS | is the number of the elements in the set RDS;

the optimal size k of the initial power load representative data set is determined by the following approximate convex frame:

wherein N is the number of elements in A, MAPE (k, A) is the mean absolute percentage error of support vector regression under RDS based on the size k, and lambda is the balance coefficient between the model complexity and the prediction precision, wherein N is⁺Represents a positive integer;

the S2 further includes:

s2-1: determining three different sizes k by using initial power load representative data set selection method with power load data at recent time point as initial point₁,k₂,k₃Represents a data set;

s2-2: training the SVR by using a particle swarm method aiming at the three power load representative data sets of S2-1 to obtain parameters of the three SVR methods, and performing nonlinear prediction;

s2-3: for a given approximate convex frame, update k by 0.618 method₁,k₂,k₃If max (k)₁,k₂,k₃)-min(k₁,k₂,k₃) If the SVR is less than or equal to 3, obtaining an optimal SVR and a power load representative data set, and terminating; if max (k)₁,k₂,k₃)-min(k₁,k₂,k₃) > 3, go back to and execute S2-1;

s3, acquiring new power load data, and updating the representative data by adopting a representative data reconstruction method according to the multi-input single-output mode data; until the updating is finished;

the representative data reconstruction method in S3 includes:

updating the representative data by using a representative data reconstruction method, wherein the updated power load representative data subset is

A_V∪B_V∪N_m

Wherein A is_VRepresenting a data set for the electrical load of the raw data A, B_VFor the power load representative dataset based on the existing SVR model and the new data B:

B_V＝{(x_i,y_i)|(x_i,y_i)∈B,|y_i-p_i|＞σ}

wherein p is_iFor existing SVR model pair input data x_iThe predicted value of (a) is the standard deviation of error of the newly added data set predicted by the existing SVR model, N_mIs A_V∪B_VThe union of the m nearest neighbor data of each point in the data set;

representing data subset A based on updated power loads_V∪B_V∪N_mUpdating model parameters by using a nested particle swarm method, and representing data subset A for the power load_V∪B_V∪N_mGenerated ith initial particle p₂(i) Comprises the following steps:

p₂(i)＝p₁(i)+λ_i×[p_{best_1}-p₁(i)]

wherein p is₁(i) For random particles generated in the last parameter space, p_{best_1}Is the previous data A_VIs set to [ p ] according to the optimum parameter_{best_1}-p₁(i)]Global contraction factor, λ, for the last optimum parameter_iTo obey the random receiving of U (0,1) distributionReducing the weight;

and establishing a support vector regression model to obtain updated power load representative data and model parameters of the SVR, and obtaining a power load prediction result at the moment.

2. The incremental SVR load prediction method based on representative data reconstruction of claim 1, further comprising: each time new power load data is obtained, S3 is repeatedly executed to update the power load representative data subset and the SVR parameters, and the power load data is continuously updated iteratively.

Images