CN116701868A - A Probabilistic Prediction Method for Short-term Wind Power Range - Google Patents

Abstract

The invention relates to the technical field of power system operation and planning, in particular to a short-term wind power section probability prediction method, which utilizes hidden information in deep learning mining data and nonlinear characteristics in a wind power sequence to generate a prediction probability interval, selects a nonlinear weight method to improve the optimization performance of a particle swarm algorithm, namely an IPSO algorithm solves part of problems existing in the traditional algorithm, improves convergence rate, and then a CNN-LSTM hybrid algorithm is selected by the hybrid artificial intelligence algorithm to construct an IPSO-CNN-LSTM algorithm prediction model based on combination of SVM and fractional regression, short-term wind power probability prediction is completed after training, wherein the CNN network can extract potential characteristics of the CNN network from sample data by using convolution kernels, long-term components can be captured by the LSTM of the long-term memory network, gradient disappearance and explosion of the existing partial algorithm are avoided, and the wind power probability prediction efficiency is improved.

Description

A Probabilistic Prediction Method for Short-term Wind Power Range

technical field

The invention relates to the technical field of power system operation and planning, in particular to a short-term wind power probability prediction method.

Background technique

With the development of social science and technology, the role of energy is increasing day by day, and the rapid development is accompanied by the continuous consumption of traditional fossil energy, and renewable energy has begun to enter the research of various scholars. In the power industry, optimizing the traditional energy structure, increasing the proportion of renewable energy, and improving grid-connected efficiency have become key research directions for global energy development. The characteristics of wind energy, such as cleanness and abundant reserves, have been favored by scholars. However, the instability and uncertainty of wind speed also cause large fluctuations in the power generation efficiency of wind farms, making it difficult to schedule wind power in advance. Accurate power prediction can not only effectively alleviate the impact of wind power uncertainty, but also affect the power generation efficiency. The safe operation of the system provides a guarantee. At present, the new trend in the development of wind power forecasting is the artificial intelligence method, but the traditional single-point forecasting model cannot quantify the irregularity and uncertainty of wind power, and the shallow learning model cannot fully extract the deep nonlinear features in the wind power sequence; single forecasting It is difficult for the model to capture the changing law in the wind power sequence and achieve satisfactory prediction results. Therefore prior art urgently needs new scheme to solve this problem.

Contents of the invention

The purpose of the present invention is to provide a short-term wind power probability prediction method, which aims to solve the technical problems that the traditional point prediction method is difficult to mine the hidden information in the sub-sequence of the wind power data set and the single prediction model is difficult to capture the variation law in the wind power sequence .

To achieve the above object, the present invention provides a short-term wind power probability prediction method, comprising the following steps:

Step 1: Collect historical meteorological data and process to obtain the initial data set for wind power probability prediction;

Step 2: Screen and supplement the initial data set to obtain a wind power data set, and perform normalization processing;

Step 3: Divide the data of the wind power dataset into a training set and a test set, construct a prediction model based on the IPSO-CNN-LSTM algorithm combining SVM and quantile regression, and perform training and prediction;

Step 4: Adjust the model parameters according to the prediction results and errors until the results are close to the test set, and complete the short-term wind power probability prediction.

Optionally, the process of collecting historical meteorological data and processing to obtain the initial data set for wind power probability forecasting is specifically to obtain regional historical meteorological data, and to query the local log, the record information of the meteorological station, and the record information of other climate monitoring systems to compare the original meteorological data and The wind power situation is processed, and the initial data set of wind power probability prediction is obtained according to the historical predicted power and the wind speed fluctuation of the numerical weather prediction NWP prediction results of the wind farm.

Optionally, the numerical weather prediction (NWP) data adopts the annual meteorological data of a selected wind farm, and the parameters include wind speed, wind direction, temperature and air pressure.

Optionally, the process of screening and supplementing the initial data set to obtain the wind power data set, and performing normalization processing is specifically to screen out the meteorological factors with the strongest correlation with wind power, and filter out the relevant data for abnormal Part of the data is processed, screened by clustering algorithms or manual methods, and the screened overall data set is divided into several conditional subsets; the abnormal data in each subset is supplemented by cleaning or interpolation methods, and finally the wind power data is obtained The set is normalized.

Optionally, the Pearson coefficient is used to screen meteorological factor data, and the most relevant meteorological data are retained for subsequent prediction; the clustering algorithm includes DBSCAN and K-means algorithm, and the manual method is to manually process abnormal data.

Optionally, the execution process of step 3 is specifically to divide the obtained data into two groups of training set and test set, and build a prediction model based on the IPSO-CNN-LSTM algorithm combining SVM and quantile regression through historical data in the training set , predict after training through the training set, and perform quantitative analysis on the output side of the model to obtain the upper and lower boundary values of the short-term wind power in quantile form under the given confidence level and the prediction sequence of the day to be predicted, and finally compare with the pre-set test set Compare.

Optionally, the construction process of the IPSO-CNN-LSTM algorithm prediction model based on combining SVM and quantile regression includes the following steps:

Use the IPSO algorithm to optimize the inertia weight ω;

Combine the CNN network with the LSTM network, use the CNN network to extract its potential features from the sample data, and the LSTM network captures the long-term components;

Use the QRNN model to reflect the nonlinear situation of the data, and obtain the probability prediction interval under the given confidence level, and further analyze the obtained data;

The mean absolute error MAE, root mean square error RMSE, interval coverage PICP and interval average bandwidth PINAW are used as the evaluation indicators of model accuracy.

The invention provides a short-term wind power probability prediction method, which uses deep learning to mine the hidden information in the data and the nonlinear characteristics in the wind power sequence, and generates a prediction probability interval, and at the same time selects a nonlinear weight method to improve the particle swarm algorithm The optimization performance of the IPSO algorithm, that is, the IPSO algorithm solves some problems existing in the traditional algorithm, improves the convergence speed, and then mixes the artificial intelligence algorithm to select the CNN-LSTM hybrid algorithm to construct the IPSO-CNN-LSTM algorithm prediction model based on the combination of SVM and quantile regression. After the training, the short-term wind power probability prediction is completed. The CNN network can extract its potential features from the sample data by using the convolution kernel, and the long-term short-term memory network LSTM can capture long-term components, avoiding the existence of some existing algorithms. The phenomenon of gradient disappearance and explosion improves the efficiency of wind power probability prediction.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Figure 1 is a schematic diagram of the convolutional neural network structure.

Figure 2 is a schematic diagram of the structure of the LSTM network memory unit.

Fig. 3 is a schematic diagram of the information transmission process of the LSTM network.

Figure 4 is a flowchart of the CNN-LSTM algorithm.

Fig. 5 is a schematic flow chart of a short-term wind power probability prediction method of the present invention.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

The invention provides a short-term wind power probability prediction method, comprising the following steps:

S1: Collect historical meteorological data and process to obtain the initial data set for wind power probability prediction;

S2: Screen and supplement the initial data set to obtain a wind power data set, and perform normalization processing;

S3: Divide the data of the wind power dataset into training set and test set, construct the IPSO-CNN-LSTM algorithm prediction model based on the combination of SVM and quantile regression, and perform training and prediction;

S4: Adjust the model parameters according to the prediction results and errors until the results are close to the test set, and complete the short-term wind power probability prediction.

The following will be described in conjunction with specific implementation steps:

In step S1, the specific execution process is: obtain regional historical meteorological data, process the original meteorological data and wind conditions by querying local logs, records of meteorological stations, and other climate monitoring systems, and predict power and wind power according to historical The wind speed fluctuation of the NWP prediction results of the field numerical weather prediction is used to obtain the initial data set of wind power probability prediction;

The Numerical Weather Prediction (NWP) data uses the annual meteorological data of a certain wind farm, including wind speed, wind direction, temperature, air pressure, etc.

The specific process of step S2 is as follows: screen out the meteorological factors with the strongest correlation with wind power, screen out the relevant data and process the abnormal data part, screen by clustering algorithm or manual method, and divide the screened overall data set into It is several conditional subsets; the abnormal data in each subset is supplemented by cleaning or interpolation methods, and the final new data set is normalized;

Further, in the above step S2, the Pearson coefficient is used to screen the meteorological factor data, and the most relevant meteorological data is retained for subsequent prediction; the clustering algorithm mainly includes the DBSCAN and K-means algorithm, and the manual method is to manually process abnormal data , cleaning the abnormal data existing after screening or performing supplementary processing by interpolation method to obtain the wind power data set, and normalize the data;

For the data selection of the numerical weather prediction (NWP) of a certain wind farm for one year, the most relevant meteorological factor data is selected by the Pearson coefficient method as the data set for subsequent prediction. The Pearson analysis method measures the correlation of two variables by the distance between the two variables, and its specific function expression formula is:

In the formula, ρxy is the Pearson correlation coefficient. When ρxy>0, the two variables are positively correlated, otherwise, they are negatively correlated. The larger ρ _xy |, the greater the degree of correlation between the two variables, x, y are two n-dimensional vectors;

Step S3: Divide the obtained data into two groups of training set and test set, construct the IPSO-CNN-LSTM algorithm prediction model based on the combination of SVM and quantile regression through historical data in the training set, and make predictions after training through the training set, Quantitative analysis is carried out on the output side of the model to obtain the upper and lower boundary values of short-term wind power in quantile form under the given confidence level and the forecast sequence of the day to be predicted, and finally compared with the pre-set test set;

Specifically, after the data set is divided into training set and test set, the IPSO-CNN-LSTM algorithm prediction model based on the combination of SVM and quantile regression is constructed for training and prediction, in which the data set is first input into the IPSO-CNN- In the LSTM algorithm model, the IPSO algorithm is used to improve the convergence speed, and the optimal inertia weight ω is obtained at the same time. This ω can be used as the weight of the SVM for data output, and then the data set is put into the CNN-LSTM to generate the power prediction value, and the prediction Put the value into a model combining SVM and quantile regression for quantitative analysis, generate a probability distribution, and further obtain a power interval prediction graph under a given confidence level.

Both the training set and the test set are taken from the processed data set, and only the training set data is used alone during training, and the obtained data is compared with the test set for verification; the IPSO-CNN-LSTM algorithm based on quantile regression predicts the model structure as follows:

Firstly, aiming at the problem of slow convergence speed of the model, the particle swarm optimization algorithm (PSO) is used to optimize its hyperparameters, and the hyperparameters to be optimized are mapped to particles. Each particle shares an individual extreme value and compares it with the global situation, and is constantly updated. Position and iterative optimization, which introduces an inertia weight ω, ω is positively correlated with the global search ability of the particle, and negatively correlated with the local search ability, while the traditional particle swarm optimization algorithm ω is a fixed value, the local search ability of the particle and the global search ability The ability is not outstanding, and it is easy to fall into the local optimum. Therefore, a nonlinear weight method is proposed to improve the optimization performance of the particle swarm optimization algorithm, that is, the IPSO algorithm. The ω optimization formula is:

ωmax and ωmin are the maximum and minimum weight values, respectively, 0.9 and 0.4; t is the current iteration number; Tmax is the maximum iteration number.

Further, the convolutional neural network (CNN) is combined with the long-term short-term memory neural network (LSTM), and the CNN is used to extract its potential features from the sample data, and the LSTM captures the long-term components. The processed data set is used as the input of the prediction model, and the power value is used as the output. The model is expressed as: P _t =f(p _tn ,k), where Pt is the wind power at time t, pt-n is the historical wind farm data, and k is preprocessed data. Finally, a deep neural network with a single hidden layer is used as the output layer, and the output result is the wind power at time t;

The convolutional neural network (CNN) is composed of convolutional layers and pooling layers alternately. There is a ReLu activation function between each convolutional layer and pooling layer to accelerate the convergence of the model. The data in the model is convoluted The processing of the neural network, after all the features are fused, the feature description of the convolutional neural network is obtained. At this time, the data is passed to the LSTM. Usually, the input data at this time needs to be reshaped into the type processed by the LSTM. After the LSTM gets the new input, determine the required Keeping and discarding, where keeping and discarding is done with the help of sigmoid activation function, you can choose to forget and save important data. The data obtained from the input gate is our updated state. Finally, the information carried is determined by the output gate, and the new state and the hidden state are transferred to the next time step.

The CNN network can extract the relationship of multi-dimensional time series data in the spatial structure. It is composed of convolutional layers and pooling layers. It uses its features such as weight sharing to speed up training and improve generalization performance. The calculation of one-dimensional convolution is:

In the formula is the kth convolutional map of layer l, f is the activation function, N is the number of input convolutional maps, * is the convolution operation, /> is the bias of the l layer corresponding to the kth convolution kernel, the CNN network principle is shown in Figure 1;

The LSTM network is a kind of recurrent neural network (RNN). In the traditional RNN hidden layer, an input gate, a forgetting gate and an output gate are added, and a unit for storing memory is added. The internal memory unit structure and information The transfer process is shown in Figure 2 and Figure 3. In the figure, Ut is the preliminary value to be output to the state of the hidden layer, σ, l, g are activation functions, where σ is the sigmoid function, l and g are the tanh functions, and Ht is The hidden state at time t, ft, it, ot are the forget gate, input gate and output gate, xt is the input information at time t, and yt is the output information at time y. The mathematical calculation process of LSTM network is:

Forget gate: f _t = σ(W _fx x _t +W _fx h _t-1 +W _fc C _t-1 +b _f );

Input gate: i _t ＝σ(W _ix x _t +W _ih h _t-1 +W _ic C _t-1 +b _i ); U _t ＝g(W _cx x _t +W _ch h _t-1 +b _c );

C _t =C _t-1 f _t +U _t i _t ;

Output gate: o _t = σ(W _ox x _t +W _oh h _t-1 +W _oc C _t-1 +b _o );

The forget gate helps the LSTM decide which information will be deleted from the memory cell state; the input gate it decides the new information to be stored in the new cell state Ct, and the output gate is used to calculate the output value Ht. In the formula, Ct-1ft means to determine how much information is forgotten from Ct-1, Utit means how much information is added to the new unit state Ct, Wix, Wfx, Wox, Wcx are weight matrixes connecting input information xt; Wic, Wfc, Woc is the diagonal matrix connecting the output value ct of the neuron activation function and the gate function; Wih, Wfh, Woh, Wch are the weight matrixes connecting the output signal Ht; bi, bf, bo, bc are the input gate, forgetting gate, output The bias corresponding to the gate and the prepared output value Ut. The algorithm flow after fitting is shown in Figure 4.

Further, the quantile regression of neural network (QRNN) model is used to reflect the nonlinearity of the data, and the data obtained by the CNN-LSTM algorithm model is further analyzed. Let the explanatory variable of the QRNN algorithm be X=[x1,x2...,xn], the response variable be Y, and the nonlinear response model of X to Y is:

In the formula, J is the number of nodes in the hidden layer, S is the number of nodes in the input layer, g _i (τ) is the input result of the hidden layer, QY(τ丨X) is the conditional quantile of the quantile point τ, f(1) is the conversion function between the input layer and the hidden layer, taking the tanh function, is the weight and bias of the input layer, /> b ⁽²⁾ (τ) is the weight and bias of the output layer, and f(2) is the activation function between the hidden layer and the output layer. Estimate the weights and biases of the QRNN model according to the minimized loss function:

In the formula, EYτ is the loss function, Yi is the explained quantity of the i-th sample, I( ) is the indicator function, and its value range is:

Assuming that there are N hidden layer neurons in the CNN-LSTM network, the output vector at time t is Use this output vector as the input of the fully connected layer to get the output value:

In the formula, ωk and b are the weights and offsets between the output of the hidden layer and the input of the fully connected layer, so as to obtain the power prediction image under the given confidence.

Furthermore, mean absolute error (MAE) and root mean square error (RMSE) are used as evaluation indicators of model accuracy, and their expressions are respectively:

where xt, is the actual value and predicted value of power after normalization, and N is the number of test verification. The smaller the obtained MAE and RMSE values, the higher the prediction accuracy of the model and the better the prediction performance.

Furthermore, prediction interval coverage (PICP) and prediction interval average bandwidth (PINAW) are used as the evaluation indicators of interval prediction accuracy, and their expressions are respectively:

In the formula, N is the number of evaluation samples, if the evaluation target value falls within the evaluation interval, then α _i =1, otherwise it is zero; β is the width of the interval, that is, the difference between the upper bound and the lower bound of the interval. The larger the value of the obtained PICP, the higher the prediction accuracy and reliability of the road; when the PICP value is constant, the smaller the PINAW value, the better the prediction effect.

Finally, the model parameters are compared and adjusted according to the image data until the predicted data is similar to the test set, and the short-term wind power probability prediction is completed. The specific process is shown in Figure 5.

In summary, the present invention has the following beneficial effects:

1. The use of multivariate forecasting models improves the calculation efficiency and calculation accuracy, and is also conducive to the stable operation of the power system and the promotion of abandoned wind absorption, improving energy utilization;

2. Conducive to improving the competitiveness of wind power in the power industry;

3. It is helpful for relevant electric power departments to determine reasonable maintenance and inspection time for wind turbines and improve the power generation efficiency of wind turbines.

What is disclosed above is only a preferred embodiment of the present invention, and of course it cannot limit the scope of rights of the present invention. Those of ordinary skill in the art can understand all or part of the process for realizing the above embodiments, and according to the rights of the present invention The equivalent changes required still belong to the scope covered by the invention.

Claims (7)

1. A short-term wind power section probability prediction method is characterized in that, comprising the following steps: Step 1: Collect historical meteorological data and process to obtain the initial data set for wind power probability prediction; Step 2: Screen and supplement the initial data set to obtain a wind power data set, and perform normalization processing; Step 3: Divide the data of the wind power dataset into training set and test set, construct the IPSO-CNN-LSTM algorithm prediction model based on SVM and quantile regression, and perform training and prediction; Step 4: Adjust the model parameters according to the prediction results and errors until the results are close to the test set, and complete the short-term wind power probability prediction. 2. the short-term wind power section probability prediction method as claimed in claim 1, is characterized in that, The process of collecting historical meteorological data and processing the initial data set for wind power probability prediction, specifically obtaining regional historical meteorological data, and processing the original meteorological data and wind conditions by querying local logs, records of meteorological stations, and records of other climate monitoring systems , according to the historical predicted power and the wind speed fluctuation of the numerical weather prediction NWP prediction results of the wind farm, the initial data set of wind power probability prediction is obtained. 3. short-term wind power section probability prediction method as claimed in claim 2, is characterized in that, The data of the numerical weather prediction NWP adopts the annual meteorological data of a certain wind farm selected, and the parameters include wind speed, wind direction, temperature and air pressure. 4. short-term wind power section probability prediction method as claimed in claim 3, is characterized in that, The process of screening and supplementing the initial data set to obtain the wind power data set, and performing normalization processing, is specifically to screen out the meteorological factors with the strongest correlation with wind power, and process the abnormal data after filtering out the relevant data , through clustering algorithm or manual screening, the screened overall data set is divided into several conditional subsets; the abnormal data in each subset is supplemented by cleaning or interpolation methods, and the final wind power data set is normalized processing. 5. short-term wind power section probability prediction method as claimed in claim 4, is characterized in that, The Pearson coefficient is used to screen meteorological factor data, and the most relevant meteorological data are retained for subsequent prediction; the clustering algorithm includes DBSCAN and K-means algorithm, and the manual method is to manually process abnormal data. 6. short-term wind power section probability prediction method as claimed in claim 5, is characterized in that, The execution process of step 3 is specifically to divide the obtained data into two groups of training set and test set, construct the IPSO-CNN-LSTM algorithm prediction model based on the combination of SVM and quantile regression through the historical data in the training set, and pass the training set Prediction is carried out after training, and quantitative analysis is carried out on the output side of the model to obtain the upper and lower boundary values of short-term wind power in quantile form under a given confidence level and the prediction sequence of the day to be predicted, and finally compared with the pre-set test set. 7. short-term wind power section probability prediction method as claimed in claim 6, is characterized in that, The construction process of the IPSO-CNN-LSTM algorithm prediction model based on the combination of SVM and quantile regression comprises the following steps: Use the IPSO algorithm to optimize the inertia weight ω; Combine the CNN network with the LSTM network, use the CNN network to extract its potential features from the sample data, and the LSTM network captures the long-term components; Use the QRNN model to reflect the nonlinear situation of the data, and obtain the prediction probability interval under different confidence levels, and further analyze the obtained data; The mean absolute error MAE, root mean square error RMSE, interval coverage PICP and interval average bandwidth PINAW are used as the evaluation indicators of model accuracy.