JP7441553B2 - Short-term wind speed prediction method and system based on rolling time series and support vector machine - Google Patents

Description

The present application belongs to the technical field of wind power generation, and particularly relates to a short-term wind speed prediction method and system based on rolling time series and support vector machine.

In recent years, wind power generation has become connected to power systems on a large scale. Wind power generation converts the kinetic energy of the wind into electrical energy, but wind in the natural world has characteristics of intermittent and wave nature, so wind power generation also has instability and wave nature. With the rapid development of wind power generation, the proportion of wind power generation in the electricity network is increasing, but this also brings many problems and challenges, such as the voltage and frequency of the electricity network. Wind power generators can reduce voltage and frequency fluctuations caused by sudden cut-offs by accurately predicting ultra-short-term wind speeds at wind power plants, and can improve networked operation of wind power plants and control of wind power generators. It plays an important role in increasing the ability of wind power generation networking from the perspective of wind power generator control.

Conventional wind speed prediction methods include continuous prediction method, time series method, Kalman filter method, artificial neural network method, and spatial correlation method. In the continuous prediction method, several recent data points are weighted as the next predicted wind speed value, but such a method has a large prediction error, so it is used as a standard comparison method to evaluate the accuracy of other prediction methods. There are many. The time series method, also known as the ARMA model, uses past data to create a model through model identification, parameter estimation, and model validation, and uses this model to derive a predictive model to achieve the purpose of forecasting the studied time series. achieve. Time-series methods require less modeling information and are easy, convenient, and quick to calculate, but lower-level models often have low prediction accuracy, and higher-level models have difficulty estimating parameters, so time-series methods needs many improvements. The Kalman filter method uses linear minimum variance estimation to extract necessary sequences from the state sequence of a dynamic wind speed system.This method, which is suitable for online wind speed prediction, is based on the statistical characteristics of noise. There is a problem with the estimation. Artificial neural network methods imitate the human brain neural network structure and its functions, are composed of large-scale interconnections of a large number of simple processing nodes, and are used to solve complex nonlinear problems, and its shortcomings This is reflected in overfitting and weak generalization ability. In the spatial correlation method, it is necessary to consider multiple sets of actually measured wind speed data at multiple points at the wind power plant and its surroundings, and to analyze the correlation of wind speeds between each site, and research on this method is constantly being improved. There is.

Chinese Patent Application Publication No. CN10518439A discloses a wind speed and power prediction method for wind power plants based on wavelet decomposition and support vector machine, which is a prior art. Using wavelet packet decomposition technology, the past wind speed time series is decomposed into wavelet packets to obtain the components of the low frequency band, medium frequency band, and high frequency band of the past wind speed time series. A gray support vector machine prediction model is used to predict each component of the past wind speed time series, and then wavelet packet reconstruction is used to obtain short-term wind speed prediction data. Using past wind power generation data and wind speed data from numerical weather forecasts as a training set, a gray support vector machine model is constructed to perform the primary prediction of wind power generation. The obtained wind speed prediction data and wind power generation power prediction data are predicted through the RBF neural network to obtain a final predicted value of wind power generation power. China Invention Patent Application Publication No. CN10744507A discloses a prior art method for predicting short-term wind speed in a wind power plant based on a combination of support vector machines. Analyze the chaotic characteristics of the original wind speed series, calculate the embedding dimension m and delay time τ of the wind speed series using the CC algorithm, and construct input and output datasets, which are divided into a training set and a validation set. On the training set, we build a single prediction model of support vector machine based on different kernel functions, and determine the key parameters using comprehensive learning strategy particle swarm optimization algorithm. In the validation set, by using a combination prediction method of variable weighting coefficients based on the guided ordered weighted harmonic mean operator, determining the weighting coefficients of each single model, making predictions respectively, and performing weighted addition on the prediction results. Get one-step wind speed prediction results. China Invention Patent Application Publication No. CN11344984747A discloses a rolling prediction method for offshore wind power generation, which is a prior art and takes into account time-series wind speed changes in units of seconds. Using data preprocessing, we process the abnormal data, complete the normalization of offshore wind speed and wind power generation, and then build a wind speed prediction model on the time scale of seconds through the differential smoothing power series, and finally A rolling LSTM storage network is constructed to realize prediction of time-series data on a second-by-second basis.

However, the accuracy and stability of wind speed prediction are low in the above conventional technologies, CN105184391A cannot interpret the chaotic characteristics of the wind speed series, and CN10744507A does not perform rolling prediction of the wind speed series, improving accuracy. CN11344984747A does not analyze the chaotic characteristics of the wind speed series.

It is an urgent issue in this technical field to overcome the deficiencies of the prior art and provide a more accurate and stable wind speed prediction method for the chaotic characteristics of wind speed series.

In order to overcome the shortcomings of the above prior art, the present invention provides a short-term wind speed prediction method and system based on rolling time series and support vector machine, which specifically adopts the following technical solutions.

The short-term wind speed prediction method based on rolling time series and support vector machine is
Step S1 of inputting the wind speed time series;
step S2 of reconstructing the phase space;
Step S3 of applying ARMA-AR-GARCH rolling prediction;
It includes step S4 of applying a minimum support vector machine, and step S5 of outputting a predicted value of wind speed.

Also,
The step S1 of inputting the wind speed time series is as follows:
inputting an original wind speed time series Pb;
The step S2 of reconstructing the phase space includes:
Reconstructing the phase space for the original wind speed time series Pb according to the following equation (1), and obtaining an intermediate value series for reconstructing the spatial phase,

Data processing is facilitated by performing preprocessing on the intermediate value sequence that reconstructs the obtained spatial phase.

In addition, the time series model ARMA (p, q) is the following equation (2) used to predict a stationary series,

The time series model ARMA(p, q) is based on the assumption that the average value of model residuals is 0 and the variance of the residuals is constant.

In addition, the GARCH model is the following equation (3),

Also, the AR(s)-GARCH(m,n) model is the following equation (4),
The time series model ARMA (p, q) and the AR (s)-GARCH (m, n) model are combined to create an ARMA-AR-GARCH model as shown in equation (5) below.

Further, the step S4 of applying the minimum support vector machine is
Organize the learning samples, perform normalization processing, and calculate the wind speeds at times t-3, t-2, t-1, and t predicted by the ARMA-AR-GARCH model, and the actual wind speeds at times t-1 and t-2. selecting wind speeds of as six input vectors;
a and b are calculated using an estimation function, and the estimation function is the following formula (6),

By substituting a time series that already has a wind speed into training, fx is obtained from the wind speed so far, whereas t-3, which was predicted by the previous ARMA-AR-GARCH model corresponding to the reverse push time, Since the corresponding variables of the wind speeds at times t-2, t-1, and t and the actual wind speeds at times t-1 and t-2 are also known, a and b are calculated using an estimation function, and
Substitute parameters a and b into equation (6), determine the specific format of the LS-SVM model,
The six input vectors are input to the LS-SVM model to obtain the predicted wind speed at time t.

Further, as the time variable t changes, step S4 is continuously repeated to obtain a predicted wind speed time series of the ARMA-AR-GARCH portion.

The present invention further relates to a short-term wind speed prediction system based on rolling time series and support vector machine, which is used to realize the method, the system comprising:
an input module for inputting wind speed time series;
a reconstruction module used to reconstruct the phase space;
a rolling prediction module used to apply ARMA-AR-GARCH rolling prediction;
It includes an application module used to apply the minimum support vector machine, and an output module that outputs the predicted value of wind speed.

The invention further relates to a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the method for predicting short-term wind speed based on the rolling time series and support vector machine described above.

The present invention is a short-term wind speed prediction method based on a rolling time series and a support vector machine. After reproducing the chaotic characteristics of wind speed using a spatial phase reconstruction method, the reaction wind speed of a time series model and a support vector machine is Combining the linear and nonlinear characteristics of the series has the beneficial effect of increasing the accuracy and reliability of wind speed prediction.

1 is a flowchart illustrating the method of the present invention. FIG. 2 is a schematic diagram showing a combination of ARMAX-AR-GARCH rolling prediction and support vector machine model of the present invention. FIG. 2 is a schematic diagram showing the ARMA-AR-GARCH rolling prediction model of the present invention. FIG. 2 is a schematic diagram showing the flow of LS-SVM of the present invention.

The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to explain the technical solutions of the present invention more clearly, and cannot limit the protection scope of the present invention. It is noted that all of the following detailed descriptions are exemplary and are intended to provide further explanation to the present application.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It should be noted that the terminology used herein is not intended to limit the exemplary embodiments of the present application, but is used only to describe particular embodiments. As used herein, the singular terms are intended to include the plural unless the context clearly dictates otherwise. It is also to be understood that when the terms "comprising" and/or "consisting" are used herein, it indicates the presence of features, steps, operations, devices, components, and/or combinations thereof.

The short-term wind speed prediction method based on rolling time series and support vector machine first reconstructs the spatial phase of the original data for the chaotic characteristics of the wind speed series, and the obtained variable input is ARMA-AR-GARCH. Using a time series model, we predict the linear intermediate value of wind speed and clarify the linear relationship of the wind speed series. At the same time, a rolling prediction strategy is used to ensure the real-time nature of the model. The variables obtained by reconstructing the wind speed linearity and spatial phase are input to a least squares support vector machine to obtain the final predicted wind speed and clarify the nonlinear relationship of the wind speed series.

Referring to FIG. 1, the short-term wind speed prediction method based on rolling time series and support vector machine of the present invention is as follows:
Step S1 of inputting the wind speed time series;
step S2 of reconstructing the phase space;
Step S3 of applying ARMA-AR-GARCH rolling prediction;
It includes step S4 of applying a minimum support vector machine, and step S5 of outputting a predicted value of wind speed.

The step S1 of inputting the wind speed time series includes a step of inputting the original wind speed time series Pb.

The present invention uses a spatial phase reconstruction algorithm to perform chaotic characteristic processing on wind speed data to obtain an intermediate value series for reconstructing the spatial phase.
Preprocess the intermediate value sequence to reconstruct the spatial phase,
Using the method with AIC as the reference search, set the search ranges of the order terms (p, q) of the ARMA model to (2-21) and (2-12), respectively, and select the order that minimizes AIC, Find solutions for order term parameters p and q.
An AR partial order is used to search so that AIC takes the minimum value.
Select (1, 1) for GARCH order selection,
To confirm the ARMA-AR-GARCH time series model, all parameters are constructed.

As shown in FIG. 2, the output of the ARMA-AR-GARCH time series model is obtained using the preprocessed wind speed series as input, and the output variables are summarized into predicted wind speeds.

As shown in FIG. 3, it is continuously repeated according to the transition of the time variable t to obtain a predicted wind speed time series of the ARMA-AR-GARCH portion.

The wind speed time series has characteristics of nonlinearity and randomness, and directly using the original series Pb for training and prediction will affect the prediction accuracy of the model, so we introduced a method to reconstruct the phase space. do.

The step S2 of reconstructing the phase space includes:
Reconstructing the phase space for the original wind speed time series Pb according to the following equation (1), and obtaining an intermediate value series for reconstructing the spatial phase,

Data processing is facilitated by performing preprocessing on the obtained intermediate value sequence for reconstructing the spatial phase.

Further, the step S3 of applying the ARMA-AR-GARCH rolling prediction,

In addition, the time series model ARMA (p, q) is the following equation (2) used to predict a stationary series,

The time series model ARMA(p, q) is based on the assumption that the average value of model residuals is 0 and the variance of the residuals is constant.

In addition, the GARCH model is the following equation (3),

The increase in the GARCH model takes into account the m-order correlation of the heteroscedastic function, and can effectively capture the heteroscedasticity of the wind speed series.

The AR(s)-GARCH(m,n) model configured in this way is the following equation (4),
The time series model ARMA (p, q) and the AR (s)-GARCH (m, n) model are combined to create an ARMA-AR-GARCH model as shown in equation (5) below.

We adopted a method using AIC as the reference search, set the search ranges of the three order terms (p, q) of the ARMA model to (2-21) and (2-12), respectively, and calculated the order at which AIC takes the minimum value. select.

To determine the order of this partial AR, the order that is searched so that AIC takes the minimum value is adopted. The order of GARCH is selected as (1, 1).

A least squares support vector machine (LS-SVM) has the advantage of strong generalization ability and global optimization, and is suitable for use in predicting nonlinear relationships. Since the performance of the radial basis function is excellent, the radial basis function is used as the kernel function.
The modeling steps of the least squares support vector machine are shown in FIG.

The step S4 of applying the minimum support vector machine includes:
Organize the learning samples, perform normalization processing, and calculate the wind speeds at times t-3, t-2, t-1, and t predicted by the ARMA-AR-GARCH model, and the actual wind speeds at times t-1 and t-2. wind speeds as six input vectors. The output variable is the predicted wind speed corresponding to time t.

Before training, the combination of parameters with the minimum cross-validation error is retained by continuously adjusting two parameters, the kernel parameter and the regularization parameter, through a cross-validation method.

Based on KKT conditions,
By substituting a time series that already has a wind speed into training, fx is obtained from the wind speed so far, whereas t-3, which was predicted by the previous ARMA-AR-GARCH model corresponding to the reverse push time, Since the corresponding variables of the wind speeds at times t-2, t-1, and t and the actual wind speeds at times t-1 and t-2 are also known, a and b are calculated using an estimation function.

Substitute parameters a and b into equation (6), determine the specific format of the LS-SVM model,
When predicting the wind speed at time t, the six input vectors are input to the LS-SVM model to obtain the predicted wind speed at time t.

As the time variable t changes, step S4 is continuously repeated to obtain a predicted wind speed time series for the ARMA-AR-GARCH portion.

The invention further relates to a short-term wind speed prediction system based on rolling time series and support vector machines, which is used to realize the method, the system comprising:
an input module for inputting wind speed time series;
a reconstruction module used to reconstruct the phase space;
a rolling prediction module used to apply ARMA-AR-GARCH rolling prediction;
It includes an application module used to apply the minimum support vector machine, and an output module that outputs the predicted value of wind speed.

The present invention further relates to a computer readable storage medium on which a computer program is stored, and when this program is executed by a processor, the short-term wind speed prediction method based on the rolling time series and support vector machine described above can be realized.

The present invention first reconstructs the spatial phase of the original data for the chaotic characteristics of the wind speed series, and the obtained variable input is used to calculate the intermediate value of the wind speed linearity using the ARMA-AR-GARCH time series model. Predict and reveal linear relationships in wind speed series. At the same time, a rolling prediction strategy is used to ensure the real-time nature of the model. The variables obtained by reconstructing the wind speed linearity and spatial phase are input to a least squares support vector machine to obtain the final predicted wind speed and clarify the nonlinear relationship of the wind speed series.

The above are only preferred embodiments of the present invention, and those skilled in the art will be able to make some improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications are also within the scope of the present invention. It should be pointed out that this should be considered as a scope of protection.

Claims (7)

A method for predicting short-term wind speed based on a computer-implemented rolling time series and support vector machine, comprising:
Step S1 of inputting the wind speed time series;
step S2 of reconstructing the phase space of the wind speed time series ;
Step S3 of applying ARMA-AR-GARCH rolling prediction;
a step S4 of applying a least squares support vector machine; and a step S5 of outputting a predicted value of wind speed ;
The step S1 of inputting the wind speed time series is as follows:
inputting an original wind speed time series Pb;
The step S2 of reconstructing the phase space includes:
Reconstructing the phase space for the original wind speed time series Pb according to the following equation (1), and obtaining an intermediate value series for reconstructing the spatial phase,
The step S3 of applying the ARMA-AR-GARCH rolling prediction includes:
The step S4 of applying a least squares support vector machine includes:
Organize the learning samples, perform normalization processing, and calculate the wind speeds at times t-3, t-2, t-1, and t predicted by the ARMA-AR-GARCH model, and the actual wind speeds at times t-1 and t-2. selecting wind speeds of as six input vectors;
α and b are calculated using an estimation function, and the estimation function is the following equation (6),
The constraints are
Lagrange function
Based on the KKT (Karush-Kuhn-Tucker) condition
By substituting a time series with wind speed into training, f(x) in the above equation (6) is obtained from the wind speed so far, whereas t-3 predicted by the ARMA-AR-GARCH model Since the corresponding variables of the wind speed at times , t-2, t-1, and t, and the actual wind speeds at times t-1 and t-2 are also known, α and b are calculated using an estimation function. ,
Substitute the parameters α and b into equation (6), determine the specific form of the LS-SVM model,
A short-term wind speed prediction method based on a rolling time series and a support vector machine, characterized in that the six input vectors are input to the LS-SVM model to obtain a predicted wind speed at time t . The time series model ARMA (p, q) is the following equation (2) used to predict a stationary series,
The rolling method according to claim 1 , wherein the time series model ARMA (p, q) is based on the assumption that the average value of model residuals is 0 and the variance of the residuals is constant. A method for predicting short-term wind speed based on time series and support vector machines. The GARCH model is the following equation (3),
The AR(s)-GARCH(m,n) model is the following equation (4),
The time series model ARMA (p, q) and the AR (s)-GARCH (m, n) model are combined to create an ARMA-AR-GARCH model as shown in equation (5) below.
4. The short-term wind speed prediction method based on a rolling time series and support vector machine according to claim 3 , wherein: The rolling time series and support vector machine according to claim 1 , characterized in that as the time variable t changes, the step S4 is continuously repeated to obtain the predicted wind speed time series of the ARMA-AR-GARCH portion. short-term wind speed prediction method based on A short-term wind speed prediction system based on rolling time series and support vector machines used for implementing the method according to claims 1 to 5 , said system comprising:
an input module for inputting wind speed time series;
a reconstruction module used to reconstruct the phase space;
a rolling prediction module used to apply ARMA-AR-GARCH rolling prediction;
A short-term wind speed prediction system based on a rolling time series and a support vector machine, comprising: an application module used to apply a least squares support vector machine; and an output module outputting a predicted value of wind speed. A computer readable storage medium on which a computer program is stored,
A computer-readable storage medium, characterized in that when this program is executed by a processor, the short-term wind speed prediction method based on rolling time series and support vector machine according to any one of claims 1 to 5 is realized.