CN105809286A - 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

A kind of based on the increment SVR load forecasting method representing data reconstruction

Technical field

The present invention relates to the quick analysis field of computer data, particularly relate to a kind of based on the increment representing data reconstruction SVR load forecasting method.

Background technology

Owing to electric energy is a kind of energy being difficult to mass storage, so the production of electric energy, carrying, distribute and consume must be With carrying out in a flash, this result just determining load forecast is power system security, stable, the premise of economical operation. Statistic law based on parametric assumption, neural net method, gray method etc. are mainly had at present than more typical load forecasting method, These methods often can only train model under data-oriented, and can not extract from mass data and represent data, because only Have to determine and a large amount of training data represents data on a small quantity, the knowledge of artificial understanding could be produced.

Support vector regression method is sparse that extract generation for training data just, and it has superior prediction performance, A small amount of representative data (referred to as supporting vector) can also be extracted.But, along with the fast development of intelligent grid, power system meeting Constantly obtaining batch new data, this does not require nothing more than to update and represents data, also requires to update existing Forecasting Methodology, it is achieved Increment load prediction.And current support vector regression method needs to re-start Model Selection and model training, this can make Model training and storage complexity constantly become big, can affect model learning precision further.This just needs those skilled in the art badly Solve corresponding technical problem.

Summary of the invention

It is contemplated that at least solve technical problem present in prior art, the most innovatively propose a kind of based on generation The increment SVR load forecasting method of table data reconstruction.

In order to realize the above-mentioned purpose of the present invention, the invention provides a kind of negative based on the increment SVR representing data reconstruction Lotus Forecasting Methodology, including:

S1, uses phase space reconfiguration analysis of history Power system load data, obtains the embedding peacekeeping time delay of Power system load data, Obtain multi input-mono-output mode data；

S2, utilizes optimum training subset method and particle swarm optimization, uses support vector regression to history electric load number According to modeling, obtain electric load and represent the model parameter of data and support vector regression, represent data according to this and support vector The model parameter returned obtains load forecast result now；

S3, obtains new Power system load data, according to this multi input-mono-output mode data, uses and represents data reconstruction Method updates this and represents data；And re-execute S3.

Described based on the increment SVR load forecasting method representing data reconstruction, it is preferred that described S2 includes:

Initial electrical load represents data set and is chosen by equation below:

$x^{*}=\underset{x_i\∈A}{\arg max}\left\{\frac{1}{\left|RDS\right|}\underset{x_j\∈RDS}{\Σ}\frac{\left(x_j-x_i\right){\left(x_j-x_i\right)}^T}{\left|x_i\right|\left|x_j\right|}\right\}$

Wherein, RDS is for initially to represent data set, and A is the Power system load data of initial acquisition, y^*For x^*Corresponding power load Lotus digital output value, RDS=RDS ∪ { (x^*,y^*) until the element in RDS reaches to be sized k, i under element in set A Mark, j is element subscript in set RDS, i=1 ..., | A |, j=1 ..., | RDS |, wherein | A | is element number in set A, | RDS | is element number in set RDS；

Initial electrical load represents the optimal size k of data set and is determined by with the convex framework of lower aprons:

$\begin{array}{cc}\underset{k\∈N^{+}}{\min }& F\left(k\right)=ln\left(n\right)\×\frac{k}{n}+\mathrm{\λ}\×MAPE(k,A)\end{array}$

Wherein, n is element number in A, and MAPE (k, A) is the average absolute of support vector regression under RDS based on size k Percentage error, λ is the coefficient of balance between model complexity and precision of prediction, wherein, N⁺Represent positive integer.

Described based on the increment SVR load forecasting method representing data reconstruction, it is preferred that described S2 also includes:

S2-1: with nearest time point Power system load data as initial point, represent the data set side of choosing with initial electrical load Method determines three different size k₁,k₂,k₃Representative data set；

S2-2: three electric loads for S2-1 represent data set, train SVR with particle swarm optimization, obtain three SVR The parameter of method, carries out nonlinear prediction；

S2-3: for the given convex framework of approximation, uses 0.618 method to update k₁,k₂,k₃If, max (k₁,k₂,k₃)-min (k₁,k₂,k₃)≤3, obtain optimum SVR and electric load represents data set, terminate；If max is (k₁,k₂,k₃)-min(k₁,k₂, k₃) ＞ 3, return to and perform S2-1.

Described based on the increment SVR load forecasting method representing data reconstruction, it is preferred that described S3 represents data weight Structure method includes:

Utilization represents data reconstruction method renewal and represents data, and the electric load after renewal represents data subset and is

A_V∪B_V∪N_m

Wherein, A_VElectric load for initial data A represents data set, B_VFor obtaining based on existing SVR model and new data B To electric load represent data set:

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

Wherein, p_iFor existing SVR model to input data x_iPredictive value, σ increases data set newly for existing SVR model prediction Error criterion variance, N_mFor A_V∪B_VIn the union of m arest neighbors data of each point；

Data subset A is represented based on the electric load after updating_V∪B_V∪N_m, use nested particle swarm optimization more new model Parameter, represents data subset A to electric load_V∪B_V∪N_mThe i-th primary p generated₂(i) be:

p₂(i)=p₁(i)+λ_i×[p_{best_1}-p₁(i)]

Wherein, p₁I () is the random particles generated in a upper parameter space, p_{best_1}For Primary Stage Data A_VOptimized parameter Arrange, [p_{best_1}-p₁(i)] it is the overall contraction factor to last optimized parameter, λ_iFor obeying the random receipts that U (0,1) is distributed Contracting weight；

Setting up support vector regression model, the electric load after being updated represents the model parameter of data and SVR, obtains Load forecast result now.

Described based on the increment SVR load forecasting method representing data reconstruction, it is preferred that also to include: to obtain new every time Power system load data, repeats S3 to update electric load and represents data subset and SVR parameter, and continuous iteration updates power load Lotus data.

In sum, owing to have employed technique scheme, the invention has the beneficial effects as follows:

1, the phase space reconstruction technique that the present invention uses can obtain the phase space load point with physical significance；

2, the present invention uses nested particle swarm optimization to realize the renewal that SVR model parameter is arranged；

3, the present invention utilizes the openness of SVR support vector regression, uses and represents the renewal extraction of data subset reconstructing method Represent data；

4, the present invention uses and represents data reconstruction method, and its meaning is: load data gather initial stage, data volume less time SVR load forecasting model modeling complexity is the least, along with the continuous collection of load data, and load data amount also continuous heap Long-pending, learning difficulty also continues to increase；And the present invention can constantly update and represent data set, SVR model parameter, it is achieved increment type Practising, therefore the complexity of its load prediction is relatively low, precision is higher.

The additional aspect of the present invention and advantage will part be given in the following description, and part will become from the following description Obtain substantially, or recognized by the practice of the present invention.

Accompanying drawing explanation

Above-mentioned and/or the additional aspect of the present invention and advantage are from combining the accompanying drawings below description to embodiment and will become Substantially with easy to understand, wherein:

Fig. 1 is Forecasting Methodology overview flow chart of the present invention.

Detailed description of the invention

Embodiments of the invention are described below in detail, and the example of described embodiment is shown in the drawings, the most from start to finish Same or similar label represents same or similar element or has the element of same or like function.Below with reference to attached The embodiment that figure describes is exemplary, is only used for explaining the present invention, and is not considered as limiting the invention.

In describing the invention, it is to be understood that term " longitudinally ", " laterally ", " on ", D score, "front", "rear", The orientation of the instruction such as "left", "right", " vertically ", " level ", " top ", " end " " interior ", " outward " or position relationship are for based on accompanying drawing institute The orientation shown or position relationship, be for only for ease of and describe the present invention and simplify description rather than instruction or the dress of hint indication Put or element must have specific orientation, with specific azimuth configuration and operation, therefore it is not intended that limit to the present invention System.

In describing the invention, unless otherwise prescribed and limit, it should be noted that term " is installed ", " being connected ", " connect " and should be interpreted broadly, for example, it may be mechanically connected or electrical connection, it is also possible to be the connection of two element internals, can Being to be joined directly together, it is also possible to be indirectly connected to by intermediary, for the ordinary skill in the art, can basis Concrete condition understands the concrete meaning of above-mentioned term.

As it is shown in figure 1, the invention provides a kind of based on increment support vector regression (SVR) load representing data reconstruction Forecasting Methodology, the method can extract optimum from historical data and represent data subset, and after new data stream arrives, the party Method can be constantly updated and represent data, updates SVR Model Selection, reduces training and the storage complexity of Forecasting Methodology, by weight The intelligible knowledge of representative data acquisition of structure, thus implement more accurately, more intelligible prediction, this Forecasting Methodology includes Following steps:

Step 1: use phase space reconstruction technique to carry out analysis of history Power system load data, obtain the embedding of Power system load data Peacekeeping time delay, obtains multi input-mono-output mode data.

Step 2: utilize optimum training subset method and particle swarm optimization, uses support vector regression (SVR) to history electricity Power load data models, and obtains representing the model parameter of data and SVR, obtains electric load now and finally predict the outcome.

Optimum training subset method utilizes the openness of support vector regression, chooses a subset of training data, the party It is minimum and can cover the approximation full detail of training data that method enables to element in this subset.

Particle swarm optimization utilizes illumination scan, it is possible to rapid screening goes out a suitable parameters of this model.

Initially represent data set to be chosen by " orthogonal design " thought:

$x^{*}=\underset{x_i\∈A}{\arg max}\left\{\frac{1}{\left|RDS\right|}\underset{x_j\∈RDS}{\Σ}\frac{\left(x_j-x_i\right){\left(x_j-x_i\right)}^T}{\left|x_i\right|\left|x_j\right|}\right\}$

Wherein, RDS is for initially to represent data set, and A is the Power system load data of initial acquisition, y^*For x^*Corresponding power load Lotus digital output value, RDS=RDS ∪ { (x^*,y^*) until the element in RDS reaches to be sized k, i under element in set A Mark, j is element subscript in set RDS, i=1 ..., | A |, j=1 ..., | RDS |, wherein | A | is element number in set A, | RDS | is element number in set RDS；

Initial electrical load represents the optimal size k of data set and is determined by with the convex framework of lower aprons:

$\begin{array}{cc}\underset{k\∈N^{+}}{\min }& F\left(k\right)=ln\left(n\right)\×\frac{k}{n}+\mathrm{\λ}\×MAPE(k,A)\end{array}$

Wherein, n is element number in A, and MAPE (k, A) is that the average absolute percentage ratio of SVR under RDS based on size k misses Difference, λ is the coefficient of balance between model complexity and precision of prediction, N⁺Represent positive integer.

Sub-step 2-1: with nearest time point data as initial point, determines three not with initially representing data set choosing method With size k₁,k₂,k₃Representative data set；

Sub-step 2-2: for three subsets of sub-step 2-1, train SVR with particle swarm optimization, obtain three SVR methods Parameter, among these, SVR is a kind of multilayer feedforward neural network, and it can approach any non-linear continuous function with arbitrary accuracy, There is the strongest fault-tolerance and processing speed quickly, it is adaptable to carry out nonlinear prediction.

Sub-step 2-3: for the given convex framework of approximation, uses 0.618 method to update k₁,k₂,k₃If, max (k₁,k₂, k₃)-min(k₁,k₂,k₃)≤3, obtain optimum SVR and represent data set, terminating；If max is (k₁,k₂,k₃)-min(k₁,k₂, k₃) ＞ 3, return to and perform sub-step 2-1.

Step 3: when new data enters system, utilizes step 1 to obtain multi input-mono-output mode data, utilizes and represent Data reconstruction method updates and represents data, and the representative data subset after renewal is

A_V∪B_V∪N_m

Wherein, A_VFor the representative data set of initial data A, B_VFor the representative obtained based on existing SVR model and new data B Data set:

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

Wherein, p_iFor existing SVR model to input data x_iPredictive value, σ increases data set newly for existing SVR model prediction Error criterion variance, N_mFor A_V∪B_VIn the union of m arest neighbors data of each point.

Based on the representative data subset A after updating_V∪B_V∪N_m, use nested particle swarm optimization to update model parameter, it is right Represent data subset A_V∪B_V∪N_mThe i-th primary p generated₂(i) be:

p₂(i)=p₁(i)+λ_i×[p_{best_1}-p₁(i)]

Wherein, p₁I () is the random particles generated in a upper parameter space, p_{best_1}For Primary Stage Data A_VOptimized parameter Arrange, [p_{best_1}-p₁(i)] it is the overall contraction factor to last optimized parameter, λ_iFor obeying the random receipts that U (0,1) is distributed Contracting weight.

Set up support vector regression (SVR) model, the representative data after being updated and the model parameter of SVR, obtain this Time electric load finally predict the outcome.

Step 4: when new data enters system, perform step 3.

The support vector of support vector regression is applied to the knowledge understanding research of mass data, the method for proposition by the present invention It is capable of the representative data reconstruction that newly-increased data cause, efficiently solves that mass data computational complexity is high, be difficult to extraction knows The problem known, nestedly achieves the renewal of model parameter, provides reference frame for Power System Planning with running.

In the description of this specification, reference term " embodiment ", " some embodiments ", " example ", " specifically show Example " or the description of " some examples " etc. means to combine this embodiment or example describes specific features, structure, material or spy Point is contained at least one embodiment or the example of the present invention.In this manual, to the schematic representation of above-mentioned term not Necessarily refer to identical embodiment or example.And, the specific features of description, structure, material or feature can be any One or more embodiments or example in combine in an appropriate manner.

Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that: not These embodiments can be carried out multiple change in the case of departing from the principle of the present invention and objective, revise, replace and modification, this The scope of invention is limited by claim and equivalent thereof.

Claims (5)

1. one kind based on the increment SVR load forecasting method representing data reconstruction, it is characterised in that including:

S1, uses phase space reconfiguration analysis of history Power system load data, obtains the embedding peacekeeping time delay of Power system load data, obtain Multi input-mono-output mode data；

S2, utilizes optimum training subset method and particle swarm optimization, uses support vector regression to build history Power system load data Mould, obtains electric load and represents the model parameter of data and support vector regression, represent data and support vector regression according to this Model parameter obtain load forecast result now；

S3, obtains new Power system load data, according to this multi input-mono-output mode data, uses and represents data reconstruction method Update this and represent data；And re-execute S3.

The most according to claim 1 based on the increment SVR load forecasting method representing data reconstruction, it is characterised in that institute State S2 to include:

Initial electrical load represents data set and is chosen by equation below:

$x^{*}=\underset{x_j\∈A}{\arg \max }\left\{\frac{1}{\left|RDS\right|}\underset{x_j\∈RDS}{\Σ}\frac{(x_j-x_i){(x_j-x_i)}^T}{\left|x_i\right|\left|x_j\right|}\right\}$

Wherein, RDS is for initially to represent data set, and A is the Power system load data of initial acquisition,ForCorresponding electric load number According to output valve,Until the element in RDS reaches to be sized k, i for element subscript, j in set A For gathering element subscript in RDS, i=1 ..., | A |, j=1 ..., | RDS |, element number during wherein | A | is set A, | RDS | for element number in set RDS；

Initial electrical load represents the optimal size k of data set and is determined by with the convex framework of lower aprons:

$\begin{array}{cc}\underset{k\∈N^{+}}{min}& F\left(k\right)=\ln \left(n\right)\×\frac{k}{n}+\end{array}\mathrm{\λ}\×MAPE(k,A)$

Wherein, n is element number in A, and MAPE (k, A) is the average absolute percentage of support vector regression under RDS based on size k Ratio error, λ is the coefficient of balance between model complexity and precision of prediction, wherein, N⁺Represent positive integer.

The most according to claim 1 based on the increment SVR load forecasting method representing data reconstruction, it is characterised in that institute State S2 also to include:

S2-1: with nearest time point Power system load data as initial point, represents data set choosing method with initial electrical load true Fixed three different size k₁,k₂,k₃Representative data set；

S2-2: three electric loads for S2-1 represent data set, train SVR with particle swarm optimization, obtain three SVR methods Parameter, carry out nonlinear prediction；

S2-3: for the given convex framework of approximation, uses 0.618 method to update k₁,k₂,k₃If, max (k₁,k₂,k₃)-min(k₁, k₂,k₃)≤3, obtain optimum SVR and electric load represents data set, terminate；If max is (k₁,k₂,k₃)-min(k₁,k₂,k₃) ＞ 3, return to and perform S2-1.

The most according to claim 1 based on the increment SVR load forecasting method representing data reconstruction, it is characterised in that institute State and S3 represents data reconstruction method include:

Utilization represents data reconstruction method renewal and represents data, and the electric load after renewal represents data subset and is

A_V∪B_V∪N_m

Wherein, A_VElectric load for initial data A represents data set, B_VFor obtain based on existing SVR model and new data B Electric load represents data set:

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

Wherein, p_iFor existing SVR model to input data x_iPredictive value, σ is the mistake that existing SVR model prediction increases data set newly Difference standard variance, N_mFor A_V∪B_VIn the union of m arest neighbors data of each point；

Data subset A is represented based on the electric load after updating_V∪B_V∪N_m, use nested particle swarm optimization to update model parameter, Electric load is represented data subset A_V∪B_V∪N_mThe i-th primary p generated₂(i) be:

p₂(i)=p₁(i)+λ_i×[p_{best_1}-p₁(i)]

Wherein, p₁I () is the random particles generated in a upper parameter space, p_{best_1}For Primary Stage Data A_VOptimized parameter set Put, [p_{best_1}-p₁(i)] it is the overall contraction factor to last optimized parameter, λ_iFor obeying the random contraction that U (0,1) is distributed Weight；

Setting up support vector regression model, the electric load after being updated represents the model parameter of data and SVR, obtains now Load forecast result.

The most according to claim 1 based on the increment SVR load forecasting method representing data reconstruction, it is characterised in that also Including: obtain new Power system load data every time, repeat S3 to update electric load and represent data subset and SVR parameter, no Disconnected iteration updates Power system load data.