CN113222289B - Prediction method of energy power based on data processing - Google Patents
Prediction method of energy power based on data processing Download PDFInfo
- Publication number
- CN113222289B CN113222289B CN202110613976.6A CN202110613976A CN113222289B CN 113222289 B CN113222289 B CN 113222289B CN 202110613976 A CN202110613976 A CN 202110613976A CN 113222289 B CN113222289 B CN 113222289B
- Authority
- CN
- China
- Prior art keywords
- wind power
- data
- wind
- time sequence
- output
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012545 processing Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 10
- 238000000513 principal component analysis Methods 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 14
- 230000009467 reduction Effects 0.000 claims description 14
- 230000001186 cumulative effect Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000001502 supplementing effect Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 25
- 238000005070 sampling Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 101150073669 NCAN gene Proteins 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000009469 supplementation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Operations Research (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Algebra (AREA)
- Development Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Evolutionary Biology (AREA)
- Pure & Applied Mathematics (AREA)
- Quality & Reliability (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
Abstract
The invention provides a data processing-based energy power prediction method, which is based on a hybrid integrated prediction model of a data processing and robust variational echo state network. Firstly, a wind speed time sequence with the maximum influence coefficient on wind power output is determined by utilizing a mutual information method, then, historical wind power and wind speed time sequences are fitted by adopting a Gaussian algorithm, error or missing data in original wind power are corrected and supplemented, meanwhile, the time sequences influencing wind power output, namely wind speed, wind direction, temperature, air pressure and air density time sequences, are processed by adopting a Kernel Principal Component Analysis (KPCA) method, principal components capable of reflecting most characteristics of the original time sequences are extracted, and redundancy of the data is eliminated. And finally, taking the processed data as an input training prediction model of the robust variational echo state network to obtain an output weight of the model, thereby obtaining a short-term prediction method of wind power.
Description
Technical Field
The invention belongs to the technical field of wind power prediction of wind power plants, and particularly relates to a wind power prediction method based on Data Processing (DP) and utilizing an Echo State Network (ESN).
Background
In recent years, with increasing environmental pollution and energy consumption, renewable energy and clean energy are increasingly receiving attention. Wind energy is a new energy source with the most development prospect gradually due to the remarkable advantages of safety, reproducibility, no pollution, wide sources and the like, and is widely favored by various countries. The fluctuation and intermittence of wind energy lead to non-stationary and random wind power generation power, which will have adverse effects on the power system, especially large-scale integration of wind power inevitably causes difficulty in power grid dispatching, thereby reducing the reliability of the power grid. Prediction of wind power is considered as an effective approach to solve this problem. Therefore, accurate prediction of wind power is critical to integration of wind power and stable operation of a power system.
Currently, models for short-term wind power prediction mainly include gray theory (GM), extreme Learning Machine (ELM), support Vector Machine (SVM), neural Network (NN), and the like. The short-term wind Power is predicted by using a Least Squares Support Vector Machine (LSSVM) (reference [1]: ZHANG, Y, WANG, P, NI, T, etc. Wind Power Prediction Based on LS-SVM Model with Error Correction [ J ]. Advances in Electrical & Computer Engineering,2017,17 (1): 3-8.) and an Extreme Learning Machine (ELM) (reference [2]: C.-L.Liu, J.-J.wu.short-term wind Power prediction based on extreme learning machine with kernels [ J ]. Journal of Engineering for Thermal Energy & Power,2017,32 (1): 95-100.), but a single prediction method usually generates a large prediction difference due to the change of wind Power, and the short-term wind Power of a wind Power plant is predicted by using a PCA and two-type fuzzy system (reference [3]: li Jun, wang Xinghui. PCA-interval two-type FLS based wind Power prediction model [ J ]: solar school, 2019,40 (03): 608-619), so that the prediction accuracy of the wind Power is improved, but the error of the original data is not considered, and the stability of the prediction is low.
The invention provides a hybrid integrated prediction model based on Data Processing (DP) and a Robust Variation Echo State Network (RVESN), which firstly utilizes a mutual information method to determine a wind speed time sequence with the maximum influence coefficient on wind power output, then adopts a Gaussian algorithm to fit historical wind power and wind speed time sequence, corrects and supplements erroneous or missing data in original wind power, and simultaneously adopts a Kernel Principal Component Analysis (KPCA) method to process the time sequence affecting wind power output, namely wind speed, wind direction, temperature, air pressure and air density, extracts principal components capable of reflecting most characteristics of the original time sequence, and eliminates redundancy of data. And finally, taking the processed data as an input training prediction model of the robust variational echo state network to obtain an output weight of the model, thereby obtaining a short-term prediction method of wind power.
Disclosure of Invention
The invention aims to provide a data processing-based energy power prediction method which is used for short-term prediction of wind power and solves the problems that the existing prediction technology does not consider errors of original data, the prediction robustness is low, the stability is poor and the like.
In order to achieve the above purpose, the invention is realized by adopting the following technical scheme.
Step 1, preprocessing data of a wind power time sequence of an original wind power plant, determining a time sequence with maximum mutual information of wind power according to a mutual information method, fitting the time sequence with the wind power time sequence, and correcting and supplementing error and missing wind power original data;
step 2, performing dimension reduction treatment on a time sequence of wind speed, wind direction, temperature, air pressure and air density which are factors influencing wind power output;
step 3, training a prediction model according to the historical wind power data;
and 4, inputting influence factor data of a time period to be predicted, and predicting the wind power of the wind power plant.
Specifically, in the step 1, when data preprocessing is performed, a mutual information method is adopted to obtain a correlation coefficient between original time sequences, and a time sequence with the maximum mutual information with wind power, namely a wind speed time sequence, is determined.
Specifically, when the wind speed time sequence and the wind power time sequence are fitted, a second-order Gaussian algorithm is adopted.
Model function adopting second-order Gaussian algorithmFor wind power and wind speedFitting the time sequence, correcting abnormal values and supplementing missing values of the original data of the wrong and missing wind power, y i Representing the fitted wind power sequence, x i For the original wind speed sequence, y max 、x max And W represents peak height, peak height position and half-width information of the Gaussian function curve, respectively; according to the fitting result, the rule for processing the wrong and missing wind power time series data is as follows: when the wind speed is less than 2m/s, the wind power is processed to be 0; when the wind speed is greater than 2m/s, when errors or blank data occur, the output power is replaced by the calculation result of the second-order Gaussian algorithm model function.
Specifically, in the step 2, when the time sequence is subjected to dimension reduction treatment, a method for analyzing the principal component is adopted.
The time series of wind speed, wind direction, temperature, air pressure and air density are subjected to principal component extraction by adopting a method of nuclear principal component analysis, namely X= { X for a given 5-dimensional original data sample 1 ,...,x k ,...,x 5 }，x k ∈R d The number of samples is 5; according to the nonlinear mapping phi: r is R d F, sample X of original data space X k Projected into a high-dimensional feature space F, resulting in phi (x k ) Sample phi (x k ) V in F k The projection of the direction isv k Feature vector for the k-th set of original samples, < ->For the ith coefficient corresponding to the kth eigenvalue, K (x i X) is the i-th mapping vector of the sample data X in F; the projection represents the kth principal component of X, and then the cumulative contribution rate Q, # of the eigenvalues is calculated>q is the number of main components, m is the total number of main components, lambda i Is a characteristic value; extracting the first 3 principal components with accumulated contribution rate greater than 90%, namely kernel principal components,thereby realizing dimension reduction.
Specifically, the prediction model in the step 3 adopts an echo state network, and outputs a weight matrix W of the echo state network prediction model according to a robust variation reasoning process out Estimating; assuming that u (t) is input at each moment and u (t) is the main component sequence after dimension reduction, the echo state network reservoir moment updates the state, and the state update equation is x (t+1) =f (W) in X u (t+1)), where W in Is a matrix which is randomly initialized when a network is initially established, is fixed, u (t+1) is a principal component sequence input at the current moment, and x (t+1) is a reserve pool state at the current moment; f () is a DR internal neuron activation function using a hyperbolic tangent function tanh; the output state equation of the echo state network is: p (t+1) =f out ×(W out X (u (t+1), x (t+1)), and P (t+1) is the wind power predicted at the current time, f out Is an output layer neuron activation function and adopts a symmetrical sigmoid function.
Specifically, the robust variational reasoning process is as follows:
(1) For the actual value of wind power given by the echo state network model, an error function is constructed by combining the power prediction output and is defined as F (W out )＝βE D +αE W ，E D Is the sum of squares of errors, E W The alpha and beta are super parameters for the regularization term introduced to the performance evaluation function according to the regularization rule; known p (W) out ) For the prior probability distribution of the output weights, according to the bayesian theory, for a given sample data y, the posterior probability distribution of the output weights can be obtained as follows:p(y|W out ) For outputting likelihood functions, p (y) is a normalization factor;
(2) The robust variational echo state network takes Gaussian mixture distribution as likelihood function of model, and the form is as follows: p (y) =ηp (y|w) out )+(1-η)p 0 (y|W out ) η is a parameter that varies adaptively with the data sample information; the output likelihood function of the model for all training samples NCan be expressed as
β 0 An initial value of beta, z is a hidden variable, z k A is the kth value of z k Is an element of the pool state matrix a;
(3)，W out is as follows
Is z k About joint distribution->Is not limited to the above-described embodiments.
Specifically, the step 3 of training the prediction model is as follows:
(1) Initializing echo state network related parameters;
(2) Taking the main component sequence after dimension reduction and wind power historical data as input variables, and taking an output weight matrix W of an echo state network prediction model out As parameters to be estimated;
(3) W is based on a variational reasoning method out An estimation is made.
Specifically, in step 4, the first 3 kernel principal components after the analysis of the kernel principal components are input with the time series of wind speed, wind direction, temperature, air pressure and air density as the original data, and the predicted power of wind power is output.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, the error and missing wind power time sequence is corrected and supplemented by adopting data fitting of a Gaussian algorithm according to the relation between wind speed and wind power; the Kernel Principal Component Analysis (KPCA) method extracts principal components of the original 5-dimensional time sequence, reduces data redundancy, and improves the computing capacity of the prediction model. The ESN has strong nonlinear system learning capability, and the calculation of the output weight of the ESN model based on robust variation reasoning enables the prediction model to have stronger robust capability. The simulation result also shows that on the premise of data processing, the wind power is predicted by using the robust variation ESN model, so that higher prediction accuracy can be obtained, and the method has a good application prospect.
Drawings
FIG. 1 is a flow chart of the overall implementation of the present invention.
FIG. 2 is a graph of the data fitting results of step 1 of the present invention.
Fig. 3 is a graph of wind power data corrected in step 1 of the present invention.
Fig. 4 is a main component diagram of KPCA in step 2 of the invention.
Fig. 5 is a graph of predicted results according to an embodiment of the present invention.
Fig. 6 is a prediction error map of an embodiment of the present invention.
Detailed Description
FIG. 1 is a flow chart of a general implementation of the present invention, wherein the input is a data-processed time series of kernel primitives, and the output is wind power prediction data of a wind farm. The main flow of the invention comprises: correcting original wind power data, reducing dimension of influence factor time sequence, training a model and predicting short-term wind power.
The invention is further described below with reference to the drawings and examples.
Step 1, preprocessing data of a wind power time sequence of an original wind power plant, determining a time sequence with maximum mutual information of wind power according to a mutual information method, fitting the time sequence with the wind power time sequence, and correcting and supplementing error and missing wind power original time sequence data, wherein the method specifically comprises the following steps of:
1.1 according to the mutual information method modelX, Y are two discrete random variables, in this example expressed as wind speed, wind direction, temperature, air pressure, air density andtwo-by-two combinations of wind power time sequences, X and Y are elements in X, Y respectively, p (X and Y) are joint probability density functions of the two time sequences, p (X) and p (Y) are edge probability density functions of the original time sequences represented by X and Y respectively; the cross correlation coefficient I (X; Y) between the original time sequences (wind speed, wind direction, temperature, air pressure, air density and wind power) is calculated, and the result is shown in the table 1, so that the time sequence with the maximum information of wind power, namely the wind speed time sequence, is determined.
TABLE 1 original time series cross-correlation coefficient
1.2, adopting a second-order Gaussian algorithm modely i Representing the fitted wind power sequence, x i For the original wind speed sequence, y max 、x max And W respectively represents peak height, peak height position and half peak width information of a Gaussian function curve, fitting is carried out on wind power and wind speed sequences, and abnormal value correction and missing value supplementation are carried out on error and missing wind power original data. According to the fitting result, the rule of handling the error and missing wind power time sequence in the embodiment is as follows: when the wind speed is less than 2m/s, the wind power is 0; when the wind speed is greater than 2m/s and error or blank data are generated, the output power is corrected and supplemented by the Gaussian fitting function calculation result, and the fitting result and the corrected wind power are shown in fig. 2 and 3. As can be seen from FIG. 2, the original wind power has more sampling points which deviate from the fitted curve unreasonably, the corrected wind power time sequence of FIG. 3 is distributed continuously along with the sampling points, and no sampling data is missed.
And 2, performing dimension reduction processing on the original time sequence of the factors influencing the wind power output.
The factors affecting the wind power output, namely wind speed, wind direction, temperature, air pressure and air, are analyzed by adopting a Kernel Principal Component Analysis (KPCA)The air density time sequence is used for extracting principal components, and for a given 5-dimensional original data sample X= { X 1 ,...,x k ,...,x 5 }，x k ∈R d ,R d Representing a real number. According to the nonlinear mapping phi: r is R d F, sample X of original data space X k Projected into a high-dimensional feature space F, resulting in phi (x k ) Sample phi (x k ) V in F k The projection of the direction isv k Feature vector for the k-th set of original samples, < ->For the ith coefficient corresponding to the kth eigenvalue, K (x i X) is the i-th mapping vector of the sample data X in F. The projection represents the kth principal component of X, and then calculates the cumulative contribution Q of the eigenvalues,q is the number of principal components, n is the total number of principal components, lambda i As the eigenvalues, the first 3 principal components with the cumulative contribution rate greater than 90% are extracted, thereby achieving dimension reduction, as shown in fig. 4. As can be seen from fig. 4, when the number of principal components is 3, the cumulative contribution rate has exceeded 90%, and as the number of principal components is greater than 3, the increase rate of the cumulative contribution rate is small, so the first 3 principal components are selected in this embodiment.
And step 3, training a prediction model according to the historical wind power data.
The prediction model of the invention adopts an Echo State Network (ESN), and based on robust variational reasoning, the output weight matrix W of the echo state network prediction model is adopted out An estimation is made.
Assuming that u (t) is input at each moment, u (t) is the main component sequence after dimension reduction, the ESN reservoir time updates the state, and the state update equation is x (t+1) =f (W in X u (t+1)), where W in Is a matrix that is randomly initialized when the network is initially established and is fixed, u (t+1) is the currentThe input principal component sequence of time x (t+1) is the pool state at the current time. f is the DR internal neuron activation function, using the hyperbolic tangent function tanh.
The output state equation for ESN is: p (t+1) =f out ×(W out X (u (t+1), x (t+1)), and P (t+1) is the wind power predicted at the current time, f out Is an output layer neuron activation function and adopts a symmetrical sigmoid function.
The parameter values of the echo state network are as follows: the scale of the reserve pool is 60, the sparsity of the internal connection weight matrix is 0.02, the spectrum radius is 0.95, and then the output weight matrix W of the echo state network prediction model is obtained according to the robust variation reasoning process out 。
The process of robust variational reasoning is as follows:
1, for the actual value of wind power given by ESN model, combining the power prediction output to construct an error function, which is defined as F (W out )＝βE D +αE W ，E D Is the sum of squares of errors, E W For the regularization term introduced to the performance evaluation function according to the regularization rule, α and β are hyper-parameters. Known p (W) out ) For the prior probability distribution of the output weights, according to the bayesian theory, for a given sample data y, the posterior probability distribution of the output weights can be obtained as follows:p(y|W out ) To output the likelihood function, p (y) is a normalization factor.
2, the robust variational echo state network takes Gaussian mixture distribution as likelihood function of model, and the form is as follows: p (y) =ηp (y|w) out )+(1-η)p 0 (y|W out ) η is a parameter that varies adaptively with the data sample information. The output likelihood function of the model can be expressed as for all training samples Nβ 0 An initial value of beta, z is a hidden variable, z k A is the kth value of z k For elements of pool state matrix AAnd (5) plain.
3, outputting weight W out Is as followsIs z k About joint distribution->Is not limited to the above-described embodiments.
The method comprises the following steps of:
1, initializing ESN related parameters; 2, taking the main component sequence after dimension reduction and wind power historical data as input variables, and taking an output weight matrix W of the network prediction model out As parameters to be estimated; 3, based on a variation reasoning method, W out An estimation is made. When a new wind power sample needs to be predicted, processing new sample data, and inputting a prediction model to obtain predicted wind power.
And 4, inputting influence factor data of a time period to be predicted, and predicting the wind power of the wind power plant.
The sampling interval of the actually measured wind power original data set is 15min, the first 90% of data is taken as training data, and the rest data are taken as test samples (120 sample points). Selecting Mean Absolute Error (MAE), root Mean Square Error (RMSE) and Mean Absolute Percent Error (MAPE) to evaluate the prediction result, wherein the expression is N is the number of test samples, P i And P' are the actual value and the predicted value of the wind power respectively.
In order to verify the feasibility and effectiveness of the prediction model of the invention for the prediction of wind power, the proposed prediction model (DP-RVESN) is compared and analyzed with a prediction model (RVESN) without a data processing link and an echo state network prediction model (DP-BESN) based on Bayesian regression. The prediction results are shown in fig. 5, and the relative errors of the prediction models are shown in fig. 6. Figure 5 shows that the predicted result of the DP-RVESN model is closer to the actual value, and the predicted curve is more similar to the actual curve. Figure 6 shows that the relative error of DP-RVESN is minimal in all models. In summary, it can be seen that the prediction model provided by the invention has a more accurate prediction value, and is suitable for short-term prediction of wind power of a wind power plant.
Claims (4)
1. The method for predicting the energy power based on the data processing is characterized by comprising the following steps of:
step 1, preprocessing data of a wind power time sequence of an original wind power plant, determining a time sequence with maximum mutual information of wind power according to a mutual information method, fitting the time sequence with the wind power time sequence, and correcting and supplementing error and missing wind power original data;
step 2, performing dimension reduction treatment on a time sequence of wind speed, wind direction, temperature, air pressure and air density which are factors influencing wind power output;
step 3, training a prediction model according to the historical wind power data;
step 4, inputting influence factor data of a time period to be predicted, and predicting wind power of a wind power plant;
step 1, when data preprocessing is carried out, a mutual information method is adopted to obtain a correlation coefficient among original time sequences, and a time sequence with the maximum mutual information with wind power, namely a wind speed time sequence, is determined;
when the wind speed time sequence and the wind power time sequence are simulated, a second-order Gaussian algorithm is adopted;
model function adopting second-order Gaussian algorithmFitting the wind power and wind speed time sequence, correcting abnormal values and supplementing missing values of the original data of the wrong and missing wind power, y i Representing the fitted wind power sequence, x i For the original wind speed sequence, y max 、x max And W represents peak height, peak height position and half-width information of the Gaussian function curve, respectively; according to the fitting result, the rule for processing the wrong and missing wind power time series data is as follows: when the wind speed is less than 2m/s, the wind power is processed to be 0; when the wind speed is greater than 2m/s and error or blank data occur, the output power is replaced by the calculation result of the second-order Gaussian algorithm model function;
in the step 3, the prediction model adopts an echo state network, and according to a robust variation reasoning process, an output weight matrix W of the echo state network prediction model is obtained out Estimating; assuming that u (t) is input at each moment and u (t) is the main component sequence after dimension reduction, the echo state network reservoir moment updates the state, and the state update equation is x (t+1) =f (W) in X u (t+1)), where W in Is a matrix which is randomly initialized when a network is initially established, is fixed, u (t+1) is a principal component sequence input at the current moment, and x (t+1) is a reserve pool state at the current moment; f () is a DR internal neuron activation function using a hyperbolic tangent function tanh; the output state equation of the echo state network is: p (t+1) =f out ×(W out X (u (t+1), x (t+1)), and P (t+1) is the wind power predicted at the current time, f out Is an output layer neuron activation function, and adopts a symmetrical sigmoid function;
the process of robust variational reasoning is as follows:
(1) For the actual value of wind power given by the echo state network model, an error function is constructed by combining the power prediction output and is defined as F (W out )＝βE D +αE W ，E D Is the sum of squares of errors, E W The alpha and beta are super parameters for the regularization term introduced to the performance evaluation function according to the regularization rule; known p (W) out ) For the prior probability distribution of the output weights, according to the bayesian theory, for a given sample data y, the posterior probability distribution of the output weights can be obtained as follows:p(y|W out ) Is the delivery ofOutputting a likelihood function, wherein p (y) is a normalization factor;
(2) The robust variational echo state network takes Gaussian mixture distribution as likelihood function of model, and the form is as follows: p (y) =ηp (y|w) out )+(1-η)p 0 (y|W out ) η is a parameter that varies adaptively with the data sample information; the output likelihood function of the model can be expressed as for all training samples N
β 0 An initial value of beta, z is a hidden variable, z k A is the kth value of z k Is an element of the pool state matrix a;
(3)，W out is as follows
Is z k About joint distribution->Is not limited to the desired one;
step 3 the step of training a predictive model is as follows:
(1) Initializing echo state network related parameters;
(2) Taking the main component sequence after dimension reduction and wind power historical data as input variables, and taking an output weight matrix W of an echo state network prediction model out As parameters to be estimated;
(3) W is based on a variational reasoning method out An estimation is made.
2. The method for predicting energy power based on data processing as recited in claim 1, wherein step 2 is a method for performing kernel principal component analysis when performing dimension reduction processing on the time sequence.
3. The method for predicting energy power based on data processing as claimed in claim 2, wherein the time series of wind speed, wind direction, temperature, air pressure and air density is subjected to principal component extraction by using a method of kernel principal component analysis, namely, for a given 5-dimensional original data sample x= { X 1 ,...,x k ,...,x 5 }，x k ∈R d The number of samples is 5; according to the nonlinear mapping phi: r is R d F, sample X of original data space X k Projected into a high-dimensional feature space F, resulting in phi (x k ) Sample phi (x k ) V in F k The projection of the direction isv k Feature vector for the k-th set of original samples, < ->For the ith coefficient corresponding to the kth eigenvalue, K (x i X) is the i-th mapping vector of the sample data X in F; the projection represents the kth principal component of X, and then the cumulative contribution rate Q, # of the eigenvalues is calculated>q is the number of main components, m is the total number of main components, lambda i Is a characteristic value; the first 3 principal components with the cumulative contribution rate of more than 90 percent, namely kernel principal components, are extracted, so that dimension reduction is realized.
4. The method for predicting energy power based on data processing as claimed in claim 3, wherein in step 4, the first 3 kernel principal elements after the analysis of the kernel principal components are inputted with the time series of wind speed, wind direction, temperature, air pressure and air density as the raw data, and the predicted power is outputted as wind power.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110613976.6A CN113222289B (en) | 2021-06-02 | 2021-06-02 | Prediction method of energy power based on data processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110613976.6A CN113222289B (en) | 2021-06-02 | 2021-06-02 | Prediction method of energy power based on data processing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113222289A CN113222289A (en) | 2021-08-06 |
| CN113222289B true CN113222289B (en) | 2024-03-12 |
Family
ID=77082342
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110613976.6A Active CN113222289B (en) | 2021-06-02 | 2021-06-02 | Prediction method of energy power based on data processing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113222289B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113610665B (en) * | 2021-08-12 | 2024-01-30 | 东北电力大学 | Wind power generation power prediction method based on multi-delay output echo state network |
| CN115130743A (en) * | 2022-06-22 | 2022-09-30 | 国网山东省电力公司电力科学研究院 | Wind turbine generator regulation rate prediction method and system based on variation inference |
| CN116911467A (en) * | 2023-09-12 | 2023-10-20 | 浙江华云电力工程设计咨询有限公司 | Renewable energy output prediction method, device and storage medium |
| CN117390967B (en) * | 2023-11-24 | 2024-03-29 | 河海大学 | Offshore wind power short-term power prediction method, device, equipment and storage medium |
| CN117434486B (en) * | 2023-12-20 | 2024-03-08 | 智联信通科技股份有限公司 | DC shunt metering error analysis processing method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109523084A (en) * | 2018-11-23 | 2019-03-26 | 东北大学 | A kind of ultrashort-term wind power prediction method based on pivot analysis and machine learning |
| US10852439B1 (en) * | 2020-04-30 | 2020-12-01 | Beihang University | Global ionospheric total electron content prediction system |
| CN112365040A (en) * | 2020-11-03 | 2021-02-12 | 哈尔滨工业大学 | Short-term wind power prediction method based on multi-channel convolution neural network and time convolution network |
| AU2020104000A4 (en) * | 2020-12-10 | 2021-02-18 | Guangxi University | Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model |
-
2021
- 2021-06-02 CN CN202110613976.6A patent/CN113222289B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109523084A (en) * | 2018-11-23 | 2019-03-26 | 东北大学 | A kind of ultrashort-term wind power prediction method based on pivot analysis and machine learning |
| US10852439B1 (en) * | 2020-04-30 | 2020-12-01 | Beihang University | Global ionospheric total electron content prediction system |
| CN112365040A (en) * | 2020-11-03 | 2021-02-12 | 哈尔滨工业大学 | Short-term wind power prediction method based on multi-channel convolution neural network and time convolution network |
| AU2020104000A4 (en) * | 2020-12-10 | 2021-02-18 | Guangxi University | Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113222289A (en) | 2021-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113222289B (en) | Prediction method of energy power based on data processing | |
| Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| Wang et al. | Wind speed forecasting based on multi-objective grey wolf optimisation algorithm, weighted information criterion, and wind energy conversion system: A case study in Eastern China | |
| CN110942194A (en) | Wind power prediction error interval evaluation method based on TCN | |
| CN111144644B (en) | Short-term wind speed prediction method based on variation variance Gaussian process regression | |
| CN110766200A (en) | Method for predicting generating power of wind turbine generator based on K-means mean clustering | |
| CN109412161B (en) | Power system probability load flow calculation method and system | |
| CN111369045A (en) | Method for predicting short-term photovoltaic power generation power | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| Wang et al. | Wind power curve modeling with asymmetric error distribution | |
| CN114757427A (en) | Autoregressive corrected LSTM intelligent wind power plant ultra-short term power prediction method | |
| CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
| CN115275991A (en) | Active power distribution network operation situation prediction method based on IEMD-TA-LSTM model | |
| CN111242353A (en) | Wind power combined prediction modeling and prediction method | |
| CN114298134A (en) | Wind power prediction method and device and electronic equipment | |
| CN115983494A (en) | Short-term wind power prediction method and system for newly-built small-sample wind power plant | |
| CN114676622A (en) | Short-term photovoltaic power prediction method based on self-encoder deep learning model | |
| CN116341717A (en) | Wind speed prediction method based on error compensation | |
| CN115907131A (en) | Method and system for building electric heating load prediction model in northern area | |
| CN115526376A (en) | Multi-feature fusion generation countermeasure network ultra-short-term wind power prediction method | |
| CN113112085A (en) | New energy station power generation load prediction method based on BP neural network | |
| Liu et al. | A Self-Data-Driven Method for Lifetime Prediction of PV Arrays Considering the Uncertainty and Volatility | |
| Paiva et al. | A Gaussian Process Based Approach to Estimate Wind Speed Using SCADA Measurements from a Wind Turbine | |
| Wydra et al. | Automated adaptive-ensemble framework for large wind power prediction in poland using deep learning models |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |