CN107301475A

CN107301475A - Load forecast optimization method based on continuous power analysis of spectrum

Info

Publication number: CN107301475A
Application number: CN201710477986.5A
Authority: CN
Inventors: 杜杰; 彭丽霞; 王雷
Original assignee: Nanjing University of Information Science and Technology
Current assignee: Nanjing University of Information Science and Technology
Priority date: 2017-06-21
Filing date: 2017-06-21
Publication date: 2017-10-27

Abstract

The invention discloses a kind of load forecast optimization method based on continuous power analysis of spectrum, using continuous power spectral analysis method, extract the harmonic compoment sequence implied in electric load time series and isolated residual sequence, harmonic compoment sequence is predicted using the BP neural network optimized based on particle cluster algorithm, predicting the outcome for each harmonic compoment sequence is obtained；The RBF neural optimized using particle cluster algorithm is predicted to the first-order difference sequence of residual sequence, obtain predicting the outcome for residual sequence by difference inverse operation, finally by the average value of average power Load Time Series and each harmonic compoment sequence predict the outcome and predicting the outcome for residual sequence is added acquisition and finally predicted the outcome.The present invention is directed to the periodic characteristics of Power system load data, and Forecast of electric load precision can be greatly improved by setting up forecast model.

Description

Load forecast optimization method based on continuous power analysis of spectrum

Technical field

The invention belongs to technical field of power systems, and in particular to a kind of Electric Load Forecasting based on continuous power analysis of spectrum Survey optimization method.

Background technology

Power system load refers to the summation of all electrical equipments consumption power in system, and also referred to as power system integrates electricity consumption Load.Comprehensive power load adds the loss in power network and the station service in power plant, be exactly in system all generators should send out General power, also referred to as power system generation load.Electric load is the key factor of influence system safe and stable operation.Electric load Prediction refers to by the analysis and research to electric load historical record, considers the various factors of influence electric load change, Such as social development planning, economic situation, meteorological changing factor and festivals or holidays, the development to future electrical energy load is made in advance Estimation.Load forecast is Power System Planning, plan, scheduling, the foundation of electricity consumption.Improve Techniques for Prediction of Electric Loads water It is flat, be conducive to formulating rational power construction planning, be conducive to reasonable arrangement power system operating mode and unit maintenance scheduling, favorably In economizing on coal, fuel-economizing and reduction cost of electricity-generating, be conducive to planned supply and use of electric power to manage, be conducive to improving economic benefit and the society of power system Can benefit.Therefore, load forecast is to realize one of important content of power system management modernization.Due to by weather condition With the influence of the factor such as people's social activities, there is substantial amounts of randomness and non-linear relation in Power system load data, influence electric power The factor of Load Time Series can be divided into inherent enchancement factor and external enchancement factor, and wherein external factor includes meteorological, society Meeting, economic dispatch, and internal factor is the result influenceed by power system internal nonlinearity factor, electric load be in system with The coefficient result of external stochasticity influence factor, the reason for its forecasting inaccuracy is true is not only the shadow of external enchancement factor Ring, it is often more important that determined by system internal motivation feature.

Therefore, a variety of forecasting procedures have been emerged in large numbers, from general statistical model, such as ARIMA time series models, gray model To all kinds of models of mind, such as neural network model, supporting vector machine model, the improvement of algorithm is expected to improve electric load Forecast precision, but most it is basic be still used in study and Generalization Capability of the Forecasting Methodology for data.Electric load Influenceed to have in obvious regularity, but this regularity by human being's production life and there is substantial amounts of randomness again, influence model Study and generalization ability.

The content of the invention

In order to solve the above-mentioned technical problem, the present invention is intended to provide a kind of Electric Load Forecasting based on continuous power analysis of spectrum Optimization method is surveyed, by continuous power analysis of spectrum, the harmonic compoment sequence implied in raw power Load Time Series is extracted simultaneously Isolated residual sequence, because harmonic compoment sequence accounts for former sequence than great, and it is regular strong, therefore can be pre- in high precision Survey, and residual sequence is due to accounting for that former sequence proportion is small thus error is limited, so that ensure that can effectively improve Electric Load Forecasting The precision of report.

Above-mentioned technical purpose is realized, above-mentioned technique effect is reached, the present invention is achieved through the following technical solutions：

A kind of load forecast optimization method based on continuous power analysis of spectrum, including

Crude sampling electric load time series are read in, and when being converted into average power load by forecast space requirement Between sequence, then calculate the anomaly sequence of average power Load Time Series；

Using continuous power spectral analysis method, what is implied in the anomaly sequence of extraction average power Load Time Series is notable Periodic sequence, and isolated residual sequence；

The BP neural network optimized using particle cluster algorithm is predicted to harmonic compoment sequence, obtains each harmonic compoment sequence What is arranged predicts the outcome；

The RBF neural optimized using particle cluster algorithm is predicted to the first-order difference sequence of residual sequence, by Difference inverse operation obtains predicting the outcome for residual sequence；

By predicting the outcome and residual sequence for the average value of average power Load Time Series and each harmonic compoment sequence Predict the outcome addition obtain finally predict the outcome.

Further, the crude sampling electric load time series are p={ p (i), i=1,2 ..., N }, and wherein N is Raw power load sampled point number；

The average power Load Time Series are p '={ p ' (j), j=1,2 ..., M }, and wherein M is to be spaced by forecast It is required that the sampled point number of the average power load sequence after conversion, p ' average value isOrder

The anomaly sequence of the average power Load Time Series is

Further, the harmonic compoment sequence is { P₁,P₂,…,P_k,…,P_K, wherein K is notable week implicit in P The number of phase sequence, P_k={ P_k(1),P_k(2),…,P_k(M) }, wherein P_k(1),P_k(2),…,P_k(M) it is respectively harmonic compoment Sequence P_kValue；The residual sequence is R=P-P₁-P₂-…-P_K。

Further, the use continuous power spectral analysis method, extracts the anomaly sequence of average power Load Time Series The harmonic compoment sequence implied in row, be specially：Using continuous power spectral method, analysis average power Load Time Series away from The harmonic compoment band of flat sequence, and when corresponding using each harmonic compoment band of frequency domain filtering method extraction of FFT Between sequence, so as to obtain harmonic compoment sequence.

Further, it is described that harmonic compoment sequence is predicted using the BP neural network optimized based on particle cluster algorithm Detailed process be：

(1) according to Kolmogorov theorems, 3 layers of BP neural network model are set up, if input layer number is I,

Hidden layer neuron number is H, and output layer neuron number is O；Wherein, H=2*I+1, O=1；

(2) parameter for needing to optimize is determined, including：The input layer number I and the length of training set of BP neural network Degree

L, in addition to：W=(w (1), w (2) ..., w (q)), q=I*H+H*O+H+O, wherein, w (1)~w (I*H) is BP The input layer of neutral net is to the link weights of hidden layer neuron, and w (I*H+1)~w (I*H+H*O) is the hidden of BP neural network Link weights containing layer to output layer neuron, w (I*H+H*O+1)~w (I*H+H*O+H) is BP neural network hidden layer nerve The threshold value of member, w (I*H+H*O+H+1)~w (I*H+H*O+H+O) is the threshold value of BP neural network output layer neuron；

(3) initialization population X=(X₁,X₂,...,X_Q1), wherein Q₁For the sum of particle, i-th of particle is X_i=(I_i, W_i,L_i), particle rapidity is V_i=(v_Ii,v_Wi,v_Li), wherein I_i、W_i、L_iFor parameter I, W, L, mono- group alternatively solves；

(4) each particle X in colony_i=(I_i,W_i,L_i) parameter that determines, construction BP neural network training set Input and output matrix, wherein for harmonic compoment sequence P_kAnd BP neural network input layer number I_iInitially set up square Battle array Z₁And Z₂, wherein：

For neural metwork training collection length L, Z to be optimized₁In last L_iArrange the input matrix I as training set_train, Z₂In last L_iArrange the output matrix O as training set_train；It regard forecast step-length l as test step-length, Z₁In last l row make For the input matrix I of test set_test, Z₂In last l arrange output matrix O as test set_test；Constructed according to training set BP neural network to the error sum of squares of test set analog result as its fitness value, with the minimum optimization direction of fitness value The quality of each particle, record particle X are judged as evaluation criterion_iCurrent individual extreme value is P_best(i) P in colony, is taken_best(i) Optimal individual is used as overall extreme value G_best；

(5) each particle X in colony_i, its position and speed are updated respectively；

In formula：ω is inertia weight, c₁、c₂For acceleration factor, g is current iteration number of times, r₁、r₂To be distributed in [0,1] Random number；

(6) target function value of each particle now is recalculated, P is updated_bestAnd G (i)_best；

(7) judge whether to reach maximum iteration, terminate optimization process if meeting, acquisition optimizes through particle cluster algorithm Obtained parameter optimal value is (I_best,W_best(w_best(1),w_best(2),...,w_best(q)),L_best), otherwise return to step (4)；

(8) I is pressed_best、W_best(w_best(1),w_best(2),...,w_best(q))、L_bestConstruct BP neural network training set Z₃ With test set Z₄And BP neural network link weights and threshold value are initialized, wherein：

w_best(1)~w_best(I*H) the initial of weights is linked to hidden layer neuron for the input layer of BP neural network Value, w_best(I*H+1)~w_best(I*H+H*O) the first of weights is linked to output layer neuron for the hidden layer of BP neural network Initial value, w_best(I*H+H*O+1)~w_best(I*H+H*O+H) for BP neural network hidden layer neuron threshold value initial value, w_best(I*H+H*O+H+1)~w_best(I*H+H*O+H+O) for BP neural network output layer neuron threshold value initial value, just This sets up BP neural network model, it is trained after the l step predictions that are iterated, and obtain corresponding predict the outcome.

Further, the RBF neural of the use particle group optimizing is carried out to the first difference sequence of residual sequence Predict, detailed process is：

(1) determine to need Optimal Parameters, including：RBF neural input layer number I and training set length L；

(2) population is initializedWherein Q₂For the sum of particle, i-th of particle is X_i=(I_i, L_i), particle rapidity isWherein I_i,L_iFor parameter I, L, mono- group alternatively solves；

(3) each particle in colonyThe parameter of determination, constructs the input of RBF neural training set And output matrix, wherein for residual sequence R and RBF neural input layer number I_iInitially set up matrix Z₅And Z₆, Wherein：

For neural metwork training collection length L, Z to be optimized₅In last L_iArrange the input matrix I as training set_train, Z₆In last L_iArrange the output matrix O as training set_train；It regard forecast step-length l as test step-length, Z₅In last l row make For the input matrix I of test set_test, Z₆In last l arrange output matrix O as test set_test；Constructed according to training set RBF neural to the error sum of squares of test set analog result as its fitness value, with the minimum optimization side of fitness value To the quality that each particle is judged as evaluation criterion, record particle X_iCurrent individual extreme value is P_best(i) P in colony, is taken_best (i) optimal individual is used as overall extreme value G_best；

(4) each particle X in colony_i, its position and speed are updated respectively；

In formula：ω is inertia weight, c₁、c₂For acceleration factor, g is current iteration number of times, and r₁、r₂To be distributed in [0, 1] random number；

(5) target function value of each particle now is recalculated, P is updated_bestAnd G (i)_best；

(6) judge whether to reach maximum iteration, terminate optimization process if meeting, acquisition optimizes through particle cluster algorithm Obtained parameter optimal value is (I_best,L_best), otherwise return to step (3).

(7) I is pressed_bestAnd L_bestConstruct RBF neural training set Z₇With test set Z₈, wherein：

RBF neural network model set up with regard to this, it is trained after the l step predictions that are iterated, and obtain corresponding prediction As a result.

Further, inertia weight ω=0.5, acceleration factor c₁=c₂=1.49445.

Beneficial effects of the present invention：

(1) the electric load harmonic compoment sequence extracted through continuous power analysis of spectrum, therefore can be high due to regular strong Precision is predicted, and harmonic compoment sequence proportion in original power load sequence is larger, therefore has established higher The basis of accuracy prediction；The residual sequence after periodic signal is eliminated on the one hand due to the proportion in overall electric load sequence Less, on the other hand become steady due to having carried out first difference computing during processing, its predicated error is relative to be had Limit, thus it is proposed by the invention by electric load sequence through continuous power analysis of spectrum, be decomposed into multiple harmonic compoment sequences and Single residual sequence, so the method being predicted respectively to each harmonic compoment sequence and residual sequence can greatly improve it is whole Body prediction effect.

(2) influence differed for prediction performance is selected for neural network structure, the present invention is directed to electric load sequence The characteristics of harmonic compoment sequence and residual sequence for isolating, BP neural network and RBF neural is respectively adopted, and for god Structural parameters through network, training set scale is optimized using particle cluster algorithm, significantly improves the generalization of neutral net Can, finally improve precision of prediction.

Brief description of the drawings

Fig. 1 is the flow chart of the load forecast optimization method based on continuous power analysis of spectrum of the present invention；

Fig. 2 is raw power load sequence chart；

Fig. 3 is the continuous power spectrum analysis result figure of the anomaly sequence of average power Load Time Series；

Fig. 4 is the harmonic compoment sequence and the residual sequence of separation of the anomaly sequential extraction procedures of average power Load Time Series Figure；

Fig. 5 (a) is the one-step prediction result figure of the inventive method；

Fig. 5 (b) predicts the outcome figure for two steps of the inventive method；

Fig. 5 (c) predicts the outcome figure for three steps of the inventive method；

Fig. 6 (a) is to set up particle group optimizing RBF neural one-step prediction result figure for raw power load sequence；

Fig. 6 (b) is to set up the step of particle group optimizing RBF neural two for raw power load sequence to predict the outcome figure；

Fig. 6 (c) is to set up the step of particle group optimizing RBF neural three for raw power load sequence to predict the outcome figure

Fig. 7 (a) is the ARIMA time series models one-step prediction result figures set up for raw power load sequence；

Fig. 7 (b) is that the step of ARIMA time series models two set up for raw power load sequence predicts the outcome figure；

Fig. 7 (c) is that the step of ARIMA time series models three set up for raw power load sequence predicts the outcome figure.

Embodiment

In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to embodiments, to the present invention It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to Limit the present invention.

The application principle of the present invention is explained in detail below in conjunction with the accompanying drawings.

A kind of load forecast optimization method based on continuous power analysis of spectrum of the present invention, using continuous power spectrum point Analysis method, extracts the harmonic compoment sequence implied in electric load time series and isolated residual sequence, using based on grain The BP neural network of swarm optimization optimization is predicted to harmonic compoment sequence, obtains predicting the outcome for each harmonic compoment sequence, The RBF neural optimized using particle cluster algorithm is predicted to the first-order difference sequence of residual sequence, is transported by difference is counter Calculation obtains predicting the outcome for residual sequence, finally by the average value of average power Load Time Series and each harmonic compoment sequence Predict the outcome and the addition acquisition that predicts the outcome of residual sequence finally predicts the outcome.

As shown in figure 1, specifically, comprising the following steps：

S1, reads in crude sampling electric load time series, and is converted into average power by forecast space requirement and bear Lotus time series, then calculates the anomaly sequence of average power Load Time Series；

The crude sampling electric load time series are p={ p (i), i=1,2 ..., N }, and wherein N is raw power Load sampled point number；

The average power load sequence is p '={ p ' (j), j=1,2 ..., M }, and wherein M is by forecast space requirement The sampled point number of average power load sequence after conversion, p ' average value isOrder

The anomaly sequence of the average power Load Time Series is

S2, using continuous power spectral analysis method, extracts what is implied in the anomaly sequence of average power Load Time Series Harmonic compoment sequence, and isolated residual sequence；

The harmonic compoment sequence is { P₁,P₂,…,P_k,…,P_K, wherein K is of the harmonic compoment sequence implied in P Number, P_k={ P_k(1),P_k(2),…,P_k(M) }, wherein P_k(1),P_k(2),…,P_k(M) it is respectively harmonic compoment sequence P_kValue； The residual sequence is R=P-P₁-P₂-…-P_K；Therefore, P=P₁+P₂+…+P_K+R。

Above-mentioned extraction process, it is specific as follows：

Assuming that a discrete-time series is x_t, wherein t=0,1 ..., N-1, common N number of sampled point, time interval δ_t=1, should Continuous power spectrum method of estimation is used, the harmonic compoment band of the discrete-time series is analyzed, and utilize Fourier transform FFT frequency domain Filtering method extracts the corresponding time series of each harmonic compoment band, specifically includes following steps：

(1) continuous power spectrum is determined

X is calculated first_tThe thick Power estimation value of continuous power spectrum：

Wherein：For the thick Power estimation value of the corresponding continuous power spectrum of h wave numbers, h is wave number, h=0,1 ... m, m=N/8, r (τ) is time series x_tLag time length be τ auto-correlation coefficient：

Wherein,It is respectively discrete-time series x with s_tAverage value and standard deviation.

It is rather smooth to (1) formula progress Chinese in order to eliminate the minor swing of thick Power estimation value, it is smooth rear for continuous power spectrum (i.e. Fig. 3 is shown in solid) is：

S₀For the corresponding continuous power spectrum of 0 wave number；S_hFor the corresponding continuous power spectrum of h wave numbers, S_mFor m wave numbers correspondence Continuous power spectrum.

(2) analytical cycle is determined

The h wave numbers corresponding cycle is：Consider in (i.e. the corresponding periodic point of Fig. 3 abscissas), the embodiment of the present invention M=N8, then

(3) continuous power spectrum credit assigned

Continuous power spectrum obtained by formula (3) is compared with red noise spectrum, its conspicuousness is judged.

Assuming that continuous power spectrum obtained by formula (3) is a certain non-periodic process spectrum, the corresponding continuous power spectrum of h wave numbers Value S_hWith averagely red noise spectrumThe ratio between defer to by its free degree ν remove χ²Distribution：

Wherein average red noise spectrumFor：

In formula,For the average value of the continuous power spectrum of all wave numbers obtained by being calculated in formula (3), r (1) is x_tFall behind Time span is 1 auto-correlation coefficient, and free degree ν is：

The embodiment of the present invention is chosen under 0.05 significance, whenWhen, the spectrum of the wave number is aobvious Write, then the cyclic swing be it is significant,Line is examined for dotted line in Fig. 3.

(4) the corresponding time series of extracting cycle band

The determination of half period zones：Notable continuous power spectrum selected by step (3) is taken each side to be less than red noise for first The periodic point of detection line, constitutes half period zones, and this half period zones is harmonic compoment band, and left side first is examined less than red noise in wherein Fig. 3 The point of survey line is set to the lower bound that right side first in the upper bound of half period zones, Fig. 3 is set to half period zones less than the point of red noise measuring line.

The extraction of the corresponding time series of half period zones：The embodiment of the present invention is ground using Chinese Academy of Sciences's measurement with geophysics Study carefully developed earth science data processing routine storehouse WHIGG F90LIB (WFL), pass through the Fourier transform FFT's using the software Frequency domain filtering subprogram, carrys out the corresponding time series of extracting cycle band, and the subprogram is：

CALL FFT_FILTER(N,X,DT,PER1,PER2,FIL_METHOD,XOUT)

Wherein N is total sampled point number, and X is x_t, DT is sampling time interval δ t, and PER1 is extracting cycle band lower bound, PER2 is the extracting cycle band upper bound, and FIL_METHOD is filtering type, and " BAND " is taken here, is referred to as the banding cycle, and XOUT is The corresponding time series of harmonic compoment band of extraction.

S3, is predicted using the BP neural network based on particle group optimizing to harmonic compoment sequence, and its detailed process is；

(1) according to Kolmogorov theorems, 3 layers of BP neural network can be realized and Any Nonlinear Function is forced Closely, therefore, the embodiment of the present invention sets up 3 layers of BP neural network model, if input layer number is I, hidden layer neuron Number is H, and output layer neuron number is O；Wherein, H=2*I+1, O=1；

(2) parameter for needing to optimize is determined, including：The input layer number I and the length of training set of BP neural network L is spent, in addition to：W=(w (1), w (2) ..., w (q)), q=I*H+H*O+H+O, wherein, w (1)~w (I*H) is BP nerve nets The input layer of network to hidden layer neuron link weights, w (I*H+1)~w (I*H+H*O) for BP neural network hidden layer extremely The link weights of output layer neuron, w (I*H+H*O+1)~w (I*H+H*O+H) is the threshold of BP neural network hidden layer neuron Value, w (I*H+H*O+H+1)~w (I*H+H*O+H+O) is the threshold value of BP neural network output layer neuron；

S4, is predicted, specific mistake using the RBF neural of particle group optimizing to the first-order difference sequence of residual sequence Cheng Wei：

(3) each particle X in colony_i(I_i,L_i) determine parameter, construct RBF neural training set input And output matrix, wherein for residual sequence R and RBF neural input layer number I_iInitially set up matrix Z₅And Z₆, Wherein：

S5, by predicting the outcome and residual error sequence for the average value of average power Load Time Series and each harmonic compoment sequence The addition that predicts the outcome of row, which is obtained, finally to predict the outcome.

Embodiment two

According to the step S1-S5 in embodiment one, the raw power duration of load application sequence of the hour rank of certain power network collection is taken Row, referring specifically to Fig. 2, due to the short-period forecast that the purpose of present example is hour rank, therefore without negative to raw power Lotus data, which make any adjustments, just can directly use, i.e. p ' (i)=p (i), i=1,2 ..., N.Taken in the present embodiment before p ' (i) 1680 points are training data, 50 points of prediction thereafter, and examination algorithm has by index of percentage ratio error MAPE Effect property, i.e.,：

Wherein, Y (i) and p ' (i) are respectively Electric Load Forecasting measured value and sampled value, and l is prediction step.

Fig. 3 show the anomaly sequence P of average power Load Time Series continuous power spectrum analysis result, finds the electricity Net electric load sequence has 12 and 24 hours 2 harmonic compoment bands for being extreme point, takes its extreme point each side first The individual periodic point less than detection line, constitutes half period zones, and this half period zones is harmonic compoment band, and described detection line is the void in Fig. 3 Line, in the present embodiment, 2 harmonic compoment bands are respectively [21.8,26.7] and [11.4,12.6], using Fourier transform FFT Frequency domain filtering method, extract this corresponding time series of 2 half period zones, respectively P₁、P₂, and obtain corresponding residual error sequence R is arranged, thus P=P₁+P₂+ R, is shown in Fig. 4.It can be seen that, 2 harmonic compoment sequences it is regular extremely strong, can be with the prediction of degree of precision； On the other hand, although the predicated error for residual error is inevitable, but is computed, residual error R energy (variance) accounting P energy (variance) is 28.56%, is declined significantly, therefore, what the predicated error for residual error far smaller than directly will be predicted for P Error.

Although neutral net has powerful nonlinear fitting ability and quick learning ability, how to select appropriate Neural network model, determines that structure, training set and the test set of neutral net still gather mainly by artificial experience or examination, its is pervasive Property is poor.Found by the harmonic compoment sequence analysis to extraction, although it has obvious mechanical periodicity feature and sequence light Sliding smooth-going, but wherein the amplitude and phase of sequence, with time trickle change, are more suitable for the stronger BP neural network of fault-tolerant ability, And residual sequence is presented after first-order difference around the fluctuation of 0 axle, more suitable for RBF neural, therefore the embodiment of the present invention pair The harmonic compoment sequence P of extraction₁、P₂Using the BP neural network model optimized based on particle cluster algorithm, and for residual sequence R Then using the RBF neural optimized based on particle cluster algorithm.

To P₁、P₂Using the BP neural network model optimized based on particle cluster algorithm, the scope of input layer number is taken For [5,14], the length of training set is [50,1650], and the scope of neural network weight and threshold value is [- 3,3], population population Scale is 50, iteration 30 times.For R then using the RBF neural optimized based on particle cluster algorithm, input layer is taken Several scopes is [5,20], and the length of training set is [50,1650], and population population scale is 50, iteration 30 times.Shown in table 1 To carry out during the prediction of 3 steps, for harmonic compoment sequence P₁、P₂With residual error R input layer number I and training set length L two The optimum results of individual parameter, for P₁、P₂The BP neural network weights of foundation and the optimum results of threshold value due to parameter it is excessive without List one by one.

Table 1

The present embodiment carried out total prediction step be 50 1 step, 2 steps and 3 step prognostic experiments, predict the outcome as Fig. 5 (a)- (c) shown in, table 2 counts for predicated error.It can be seen that, with the increase of prediction step, overall precision of prediction has declined, but overall Error is less than 5%, predicts the outcome and is more satisfied with.

Table 2

	1 step is predicted	2 steps are predicted	3 steps are predicted
				MAPE	0.0399	0.0436	0.0434

Contrast experiment 1

In order to verify influence of the optimisation strategy proposed by the present invention to experimental result, contrast experiment 1 is to raw power load Sequence p ' directly carries out first difference computing, and the RBF neural of particle cluster algorithm optimization is set up afterwards, input layer is taken The scope of number is [5,25], and the length of training set is [50,1650], and population population scale is 50, iteration 30 times.The institute of table 3 When being shown as carrying out the prediction of 3 steps, the RBF neural parameter optimization result set up for raw power load sequence p '.

Table 3

Likewise, contrast experiment 1 carried out total prediction step be 50 1 step, 2 steps and 3 step prognostic experiments, predict the outcome as Shown in Fig. 6 (a)-(c), table 4 counts for predicated error, and contrast table 2 is visible, and the mean error of its 1~3 step prediction increases than table 2 60.44%.

Table 4

	1 step is predicted	2 steps are predicted	3 steps are predicted
				MAPE	0.0395	0.0708	0.0933

If not carrying out first difference computing to p ', input layer number I and the training of RBF neural are arbitrarily chosen Collect length L, final predicated error difference can be very big, and the embodiment of the present invention chooses two groups of difference I and L to final predicated error Influence be illustrated, as shown in table 5.

Table 5

The mean error of the contrast experiment of two groups of different parameters its 1~3 step prediction adds 47.68% He than table 2 170.61%.The bad selection for showing neural network parameter of this group of contrast experiment's effect for neutral net learning ability and It is extensive to cause tremendous influence so that directly to use neural net model establishing effect and bad.

Contrast experiment 2

Difference ARMA model (Autoregressive is set up for raw power load sequence IntegratedMoving Average Model, ARIMA) model.100 sampled data points before future position are chosen, pass through AIC Criterion determines the structure that rank method determines ARIMA models, likewise, contrast experiment 2 carried out total prediction step be 50 1 step, 2 steps and 3 step prognostic experiments, predict the outcome as shown in Fig. 7 (a)-(c), and table 6 counts for predicated error, and contrast table 2 is visible, and its 1~3 step is pre- The mean error of survey adds 136.25% than table 2.

Table 6

	1 step is predicted	2 steps are predicted	3 steps are predicted
				MAPE	0.0169	0.1486	0.1343

In summary：

The electric load harmonic compoment sequence extracted through continuous power analysis of spectrum is regular strong, thus can with it is high-precision enter Row prediction, and harmonic compoment sequence proportion in former sequence is larger, therefore established the basis of degree of precision prediction；Pick Except the residual sequence after harmonic compoment sequence due in former sequence proportion less, therefore predicated error is relatively limited.This hair It is bright it is proposed by electric load sequence through continuous power analysis of spectrum, be decomposed into multiple harmonic compoment sequences and single residual error sequence Row, and then the method being predicted respectively to each harmonic compoment sequence and residual sequence can greatly improve overall prediction effect Really.

The general principle and principal character and advantages of the present invention of the present invention has been shown and described above.The technology of the industry Personnel are it should be appreciated that the present invention is not limited to the above embodiments, and the simply explanation described in above-described embodiment and specification is originally The principle of invention, without departing from the spirit and scope of the present invention, various changes and modifications of the present invention are possible, these changes Change and improvement all fall within the protetion scope of the claimed invention.The claimed scope of the invention by appended claims and its Equivalent thereof.

Claims

1. a kind of load forecast optimization method based on continuous power analysis of spectrum, it is characterised in that：Including

Crude sampling electric load time series are read in, and average power duration of load application sequence is converted into by forecast space requirement Row, then calculate the anomaly sequence of average power Load Time Series；

Using continuous power spectral analysis method, the harmonic compoment implied in the anomaly sequence of average power Load Time Series is extracted Sequence, and isolated residual sequence；

The BP neural network optimized using particle cluster algorithm is predicted to harmonic compoment sequence, obtains each harmonic compoment sequence Predict the outcome；

By the average value of average power Load Time Series and each harmonic compoment sequence predict the outcome and residual sequence it is pre- Results added acquisition is surveyed finally to predict the outcome.

2. a kind of load forecast optimization method based on continuous power analysis of spectrum according to claim 1, its feature It is：

The average power Load Time Series are p '={ p ' (j), j=1,2 ..., M }, and wherein M is by forecast space requirement The sampled point number of average power load sequence after conversion, p ' average value isOrder

The anomaly sequence of the average power Load Time Series is

3. a kind of load forecast optimization method based on continuous power analysis of spectrum according to claim 2, its feature It is：

The harmonic compoment sequence is { P₁,P₂,…,P_k,…,P_K, wherein K is the number of the harmonic compoment sequence implied in P, P_k ={ P_k(1),P_k(2),…,P_k(M) }, wherein P_k(1),P_k(2),…,P_k(M) it is respectively harmonic compoment sequence P_kValue；

The residual sequence is R=P-P₁-P₂-…-P_K。

4. a kind of load forecast optimization side based on continuous power analysis of spectrum according to any one of claim 1-3 Method, it is characterised in that：In the use continuous power spectral analysis method, the anomaly sequence for extracting average power Load Time Series Implicit harmonic compoment sequence, be specially：Using continuous power spectral method, the anomaly sequence of average power Load Time Series is analyzed The harmonic compoment band of row, and extract the corresponding time sequence of each harmonic compoment band using the frequency domain filtering method of FFT Row, so as to obtain harmonic compoment sequence.

5. a kind of load forecast optimization method based on continuous power analysis of spectrum according to claim 3, its feature It is：The detailed process being predicted using the BP neural network optimized based on particle cluster algorithm to harmonic compoment sequence For：

(1) according to Kolmogorov theorems, 3 layers of BP neural network model are set up, if input layer number is I, hidden layer Neuron number is H, and output layer neuron number is O；Wherein, H=2*I+1, O=1；

(2) parameter for needing to optimize is determined, including：The input layer number I of the BP neural network and length L of training set, Also include：W=(w (1), w (2) ..., w (q)), q=I*H+H*O+H+O, wherein, w (1)~w (I*H) is BP neural network Input layer is to the link weights of hidden layer neuron, and w (I*H+1)~w (I*H+H*O) is that the hidden layer of BP neural network is extremely exported The link weights of layer neuron, w (I*H+H*O+1)~w (I*H+H*O+H) is the threshold value of BP neural network hidden layer neuron, w (I*H+H*O+H+1)~w (I*H+H*O+H+O) is the threshold value of BP neural network output layer neuron；

(3) population is initializedWherein Q₁For the sum of particle, i-th of particle is X_i=(I_i,W_i,L_i), Particle rapidity isWherein I_i、W_i、L_iFor parameter I, W, L, mono- group alternatively solves；

(4) each particle X in colony_i=(I_i,W_i,L_i) determine parameter, construct BP neural network training set input And output matrix, wherein for harmonic compoment sequence P_kAnd BP neural network input layer number I_iInitially set up matrix Z₁ And Z₂, wherein：

For neural metwork training collection length L, Z to be optimized₁In last L_iArrange the input matrix I as training set_train, Z₂In Last L_iArrange the output matrix O as training set_train；It regard forecast step-length l as test step-length, Z₁In last l row conducts The input matrix I of test set_test, Z₂In last l arrange output matrix O as test set_test；The BP constructed according to training set Neutral net, as its fitness value, is made to the error sum of squares of test set analog result with the minimum optimization direction of fitness value The quality of each particle, record particle X are judged for evaluation criterion_iCurrent individual extreme value is P_best(i) P in colony, is taken_best(i) most Excellent individual is used as overall extreme value G_best；

<mrow> <msubsup> <mi>V</mi> <mi>i</mi> <mrow> <mi>g</mi> <mo>+</mo> <mn>1</mn> </mrow> </msubsup> <mo>=</mo> <msubsup> <mi>&omega;V</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>+</mo> <msub> <mi>c</mi> <mn>1</mn> </msub> <msub> <mi>r</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <msub> <mi>P</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>)</mo> <mo>+</mo> <msub> <mi>c</mi> <mn>2</mn> </msub> <msub> <mi>r</mi> <mn>2</mn> </msub> <mrow> <mo>(</mo> <msub> <mi>G</mi> <mrow> <mi>b</mi> <mi>e</mi> <mi>s</mi> <mi>t</mi> </mrow> </msub> <mo>-</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mi>g</mi> </msubsup> <mo>)</mo> </mrow> <mo>,</mo> </mrow>

In formula：ω is inertia weight, c₁、c₂For acceleration factor, g is current iteration number of times, r₁、r₂For be distributed in [0,1] with Machine number；

(7) judge whether to reach maximum iteration, terminate optimization process if meeting, obtain and obtained through particle cluster algorithm optimization Parameter optimal value be (I_best,W_best(w_best(1),w_best(2),...,w_best(q)),L_best), otherwise return to step (4)；

(8) I is pressed_best、W_best(w_best(1),w_best(2),...,w_best(q))、L_bestConstruct BP neural network training set Z₃And test Collect Z₄And BP neural network link weights and threshold value are initialized, wherein：

w_best(1)~w_best(I*H) for BP neural network input layer to hidden layer neuron link weights initial value, w_best(I*H+1)~w_best(I*H+H*O) the initial of weights is linked to output layer neuron for the hidden layer of BP neural network Value, w_best(I*H+H*O+1)~w_best(I*H+H*O+H) for BP neural network hidden layer neuron threshold value initial value, w_best (I*H+H*O+H+1)~w_best(I*H+H*O+H+O) for BP neural network output layer neuron threshold value initial value, this is built Erect BP neural network model, it is trained after the l step predictions that are iterated, and obtain corresponding predict the outcome.

6. a kind of load forecast optimization method based on continuous power analysis of spectrum according to claim 3, its feature It is：The RBF neural of the use particle group optimizing is predicted to the first difference sequence of residual sequence, detailed process For：

(2) population is initializedWherein Q₂For the sum of particle, i-th of particle is X_i=(I_i,L_i), grain Sub- speed isWherein I_i,L_iFor parameter I, L, mono- group alternatively solves；

(3) each particle X in colony_i(I_i,L_i) parameter that determines, construct the input of RBF neural training set and defeated Go out matrix, wherein for residual sequence R and RBF neural input layer number I_iInitially set up matrix Z₅And Z₆, its In：

For neural metwork training collection length L, Z to be optimized₅In last L_iArrange the input matrix I as training set_train, Z₆In Last L_iArrange the output matrix O as training set_train；It regard forecast step-length l as test step-length, Z₅In last l row conducts The input matrix I of test set_test, Z₆In last l arrange output matrix O as test set_test；The RBF constructed according to training set Neutral net, as its fitness value, is made to the error sum of squares of test set analog result with the minimum optimization direction of fitness value The quality of each particle, record particle X are judged for evaluation criterion_iCurrent individual extreme value is P_best(i) P in colony, is taken_best(i) most Excellent individual is used as overall extreme value G_best；

In formula：ω is inertia weight, c₁、c₂For acceleration factor, g is current iteration number of times, and r₁、r₂To be distributed in [0,1] Random number；

(6) judge whether to reach maximum iteration, terminate optimization process if meeting, obtain and obtained through particle cluster algorithm optimization Parameter optimal value be (I_best,L_best), otherwise return to step (3).

RBF neural network model set up with regard to this, it is trained after the l step predictions that are iterated, and obtain corresponding predict the outcome.

7. a kind of load forecast optimization method based on continuous power analysis of spectrum according to claim 5 or 6, it is special Levy and be：Inertia weight ω=0.5, acceleration factor c₁=c₂=1.49445.