FEEMD-GWO-GCF-based short-term wind speed prediction method

Abstract

The invention discloses a short-term wind speed prediction method based on FEEMD-GWO-GCF, and the method comprises the following steps: S1, obtaining the historical wind speed data of a wind power plant, and carrying out the preprocessing of the historical wind speed data, and obtaining a wind speed sequence data set; s2, based on an FEEMD algorithm, dividing the wind speed sequence data set into a plurality of subsequences according to a time sequence, and performing set empirical mode decomposition based on fractal dimensions on the subsequences to obtain a plurality of mode components; s3, on the basis of the all-parameter continued fraction model and the contrast quotient, constructing a universal continued fraction model of any sub-sequence; s4, establishing a GCF model, determining an optimal order of the model based on AIC, and optimizing model structure parameters according to GWO; s5, all the subsequences are substituted into the GCF model for prediction, and predicted values of all the subsequences are obtained; and S6, superposing the predicted values of the subsequences to obtain a final prediction result. The prediction method is high in accuracy, can reduce adverse effects and operation cost of large-scale wind power integration, and maintains stable operation and real-time scheduling of a power system.

Description

Technical Field

[0001] The invention relates to the technical field of wind power generation, and in particular to a short-term wind speed prediction method based on FEEMD-GWO-GCF. Background Art

[0002] Under the background of the national "dual carbon" goal, my country's wind power technology has been accelerated, and the installed capacity and proportion of wind power have increased rapidly. Due to the intermittent and chaotic characteristics of short-term wind speed, large-scale access to wind power will seriously affect the normal operation of the power grid. Accurately predicting wind speed can reduce the adverse effects and operating costs of large-scale wind power grid connection, maintain the stable operation and real-time dispatch of the power system, and provide reliable data support for it. Therefore, improving the accuracy of short-term wind speed prediction has important practical significance and application value.

[0003] Based on different modeling principles, current prediction models are mainly divided into: physical models, statistical models, artificial intelligence models and combined models. Physical models need to consider many factors before modeling to achieve accurate forecasts of wind speed. Due to the complex information collection process, the corresponding cost is high and the amount of calculation is large, so they are suitable for long-term predictions. Statistical models predict wind speed by establishing a mapping relationship between the wind speed at the current moment and the wind speed at the historical moment. However, since the wind speed sequence is non-stationary, it is often difficult to achieve ideal prediction results by simply using statistical models for prediction. Artificial intelligence models are widely used in the field of wind speed prediction because of their powerful learning and nonlinear problem processing capabilities. However, the artificial intelligence prediction model has a complex structure, takes a long time to calculate, is difficult to select hyperparameters, and is prone to fall into local optimal states during predictions, resulting in problems such as overfitting. Compared with the limited role of a single model in extracting wind speed features, the combined model can give full play to the advantages of each model and have a more stable prediction effect.

[0004] In summary, physical methods require a lot of calculations in wind speed forecasting, are inefficient, and are only suitable for long-term forecasting; statistical methods have better prediction effects in stable data, but due to the strong volatility of wind speed series, it is often difficult to achieve ideal prediction results; artificial intelligence methods have strong nonlinear fitting capabilities, but the model structure is complex and the calculation is time-consuming, and overfitting is prone to occur in the forecast. Summary of the invention

[0005] To solve at least one of the above problems, the present invention proposes a short-term wind speed prediction method based on FEEMD-GWO-GCF.

[0006] The technical solution of the present invention is: a short-term wind speed prediction method based on FEEMD-GWO-GCF, comprising the following steps:

[0007] S1. Obtain historical wind speed data of the wind farm and preprocess it to obtain a wind speed series data set;

[0008] S2. Based on the FEEMD algorithm, the custom sequence data set is divided into multiple subsequences according to the time series, and the subsequences are subjected to collective empirical mode decomposition based on fractal dimension to obtain multiple modal components;

[0009] S3. Based on the full parameter continued fraction model and contrast quotient, a general continued fraction model for any subsequence is constructed;

[0010] S4, establish the GCF model, determine the optimal order of the model based on AIC, and optimize the model structure parameters according to GWO;

[0011] S5: Substitute all subsequences into the GCF model for prediction to obtain the predicted value of each subsequence;

[0012] S6: Superimpose the predicted values ​​of each subsequence to obtain the final prediction result.

[0013] Beneficial effects: 1. In the present invention, by improving the EEMD and CEEMD algorithms, the original wind speed sequence data set is decomposed using the ensemble empirical mode decomposition (FEEMD) based on fractal dimension. This improved algorithm can ensure the accuracy of decomposition and the physical meaning of components while maintaining the adaptability of decomposition.

[0014] 2. The present invention constructs a universal continued fraction model based on the contrast quotient theory to effectively capture the changing rules between wind speed sequences. The gray wolf optimization algorithm is used to optimize the model structure parameters; the Akaike information criterion is introduced to determine the optimal order of the universal continued fraction model. By comparing the AIC values ​​of the models under different orders, the order of the universal continued fraction model is determined.

[0015] 3. In order to further improve the prediction accuracy of short-term wind speed, the present invention proposes a short-term wind speed combined prediction method based on FEEMD-GWO-GCF, aiming to reduce the adverse effects and operating costs of large-scale wind power grid connection and maintain the stable operation and real-time scheduling of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a historical wind speed data diagram of this embodiment;

[0017] Figure 2 is the component of the wind speed sequence data set in this embodiment;

[0018] Figure 3 The figure is a comparison chart of the prediction results of the method in this embodiment and other existing methods. DETAILED DESCRIPTION

[0019] The specific implementation modes of the present invention will be described clearly and completely below in combination with examples and drawings. Obviously, the described examples are only some embodiments of the present invention, rather than all embodiments.

[0020] A short-term wind speed prediction method based on FEEMD-GWO-GCF comprises the following steps:

[0021] S1. Obtain historical wind speed data of the wind farm and preprocess it to obtain a wind speed series data set;

[0022] In this step, the historical wind speed data can be obtained based on the actual sampling and recording of the wind farm. Among them, for some historical wind speed data, there may be data missing problems. For this type of problem, we use the average interpolation method to fill it.

[0023] In this embodiment, the historical wind speed data selected are from the Sotavento wind farm in Galicia, Spain, with a sampling interval of 10 minutes and a total of 1000 data. In order to facilitate subsequent experiments, in this embodiment, these 1000 data are divided into a training set and a test set according to an 8:2 ratio, as shown in FIG. Figure 1 shown.

[0024] S2. Based on the FEEMD algorithm, the custom sequence data set is divided into multiple subsequences according to the time series, and the subsequences are subjected to collective empirical mode decomposition based on fractal dimension to obtain multiple modal components;

[0025] S2 includes the following steps:

[0026] S21. Divide the wind speed sequence data set into multiple subsequences according to the time series, and add positive and negative paired white noise signals to each subsequence in sequence to form a noise sequence data set: Formula 1, where They represent the data after adding positive white noise and negative white noise to the data at time t in the i-th group of noise sequence data sets; X(t) represents the data at time t in the wind speed sequence data sets; a i is the amplitude of the white noise signal, i=1,2,…,M; n i (t) represents the t-th i-th group of white noise signals;

[0027] S22. Perform 2M EMD on the noise sequence data set to obtain multiple first-order component sequences: Formula 2, where IMF1 is the first-order component sequence;

[0028] S23, calculate the box-counting dimension size of IMF1, and compare it with the box-counting dimension threshold. If the box-counting dimension of IMF1 is greater than, it is regarded as an abnormal signal, otherwise it is output as a stable signal; in this step, after multiple simulation experiments by the inventor, it is found that for the box-counting dimension threshold, it is more appropriate to set it to 1.22-1.27, and the modal components finally decomposed have satisfactory effects. In this embodiment, the box-counting dimension threshold is set to 1.27. Of course, those skilled in the art can set box-counting dimension thresholds of different sizes according to actual conditions;

[0029] Specifically, in this step, we first use square boxes with a side length of ε to cover the waveform of the sequence. The minimum number of square boxes required is recorded as N(ε). Then, ε is gradually reduced, and the number of boxes N(ε) increases as the side length decreases. When ε approaches 0, the logarithmic rate at which N(ε) increases as ε decreases is the box-counting dimension. The box-counting dimension calculation formula for the time series waveform is as follows:

[0030]

[0031] By calculating the box dimension, the fractal characteristics of the time series are obtained.

[0032] S24. Repeat S21 to S23 until multiple modal components are output. At the same time, for the remaining signal, we perform EMD decomposition without adding noise. This is a conventional operation in this field.

[0033] The final result of this step is Figure 2 As shown, from Figure 2 It can be seen that multiple components are finally output.

[0034] S3. Based on the full parameter continued fraction model and contrast quotient, a general continued fraction model for any subsequence is constructed;

[0035] This step mainly includes the following sub-steps:

[0036] S31. Establish a full parameter continued fraction model:

[0037] The full parameter continued fraction model is a sequence prediction method based on the contrast quotient theory. For time series data sets, f(x i ) is x i Function value of , construct n-term truncated continued fraction:

[0038]

[0039] Let g(x i )=f(x i ),i=0,1,…,n, through recursion, the above formula is converted into g k (x):

[0040]

[0041] According to equation 2 and equation 3, we can get g in equation 5 k (x i ) and g k+1 (x i )’s recursive relation:

[0042]

[0043] In order to solve the problem of complex calculation, the parameters a0, a1, ..., a n-1 Instead of x0,x1,…,x n-1 , which can be expressed as:

[0044]

[0045] Among them, a0, a1, …, a n-1 and b0,b1,…,b n-1 It is the structural parameter of the model, which can be obtained through intelligent optimization algorithm;

[0046] S32. Contrast quotient construction:

[0047] In this step,

[0048] Among them, d is the model order, β k-d To simplify the above contrast quotient, let available:

[0049]

[0050] S33. Construct a general continued fraction model: In the formula, [α k-d ,…,α k-1 β k-d ,…β k-1 ] and order d are parameters of the general continued fraction model;

[0051] S4, establish the GCF model, determine the optimal order of the model based on AIC, and optimize the model structure parameters according to GWO;

[0052] In this step, the process of GWO optimizing the model structure parameters includes the following steps:

[0053] In a wolf pack, the gray wolves are divided into four levels: α, β, δ, and ω. The first three levels are decision-making and leadership. The leadership ω-level gray wolves keep approaching the prey to achieve the optimal solution to the target problem. The optimization process is divided into three stages: encirclement, pursuit, and attack. In GWO, the gray wolf's position update in the encirclement stage is: Where t is the current iteration number, For the prey location, is the position vector of the gray wolf, is a random vector of (0, 1), As the iteration progresses, it decreases linearly from 2 to 0. It represents the simulated attack of gray wolf on prey, and the value is affected by The impact of The swing factor.

[0054] In the iterative process, α, β and δ guide the movement of ω to achieve global optimization. In the D-dimensional space of search, its hunting behavior is described as:

[0055]

[0056] in, is the distance between the current wolf and the three best gray wolves, is the updated position of the gray wolf under the guidance of wolves a, β, and δ.

[0057] S5: Substitute all subsequences into the GCF model for prediction to obtain the predicted value of each subsequence;

[0058] S6: The predicted values ​​of each subsequence are superimposed to obtain the final prediction result. At the same time, in order to illustrate the accuracy of the method of the present invention, the prediction result of this embodiment (FEEMD-GWO-GCF) is compared with the existing GCF, and the FEEMD of this embodiment is replaced by EMD, EEMD, CEEMD, and MEEMD to form EMD-GWO-GCF, EEMD-GWO-GCF, CEEMD-GWO-GCF, and MEEMD-GWO-GCF methods.

[0059] The final prediction results are compared as follows: Figure 3 As shown, from Figure 3 It can be seen that the method of the embodiment of the present invention is closer to the actual value and has the smallest error compared with the existing method.

[0060] The above description is only a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A short-term wind speed prediction method based on FEEMD-GWO-GCF, characterized in that: The following steps are involved: S1. Obtain historical wind speed data of the wind farm and preprocess it to obtain a wind speed series data set; S2. Based on the FEEMD algorithm, the custom sequence data set is divided into multiple subsequences according to the time series, and the subsequences are subjected to collective empirical mode decomposition based on fractal dimension to obtain multiple modal components; S3. Based on the full parameter continued fraction model and contrast quotient, a general continued fraction model for any subsequence is constructed; S4, establish the GCF model, determine the optimal order of the model based on AIC, and optimize the model structure parameters according to GWO; S5: Substitute all subsequences into the GCF model for prediction to obtain the predicted value of each subsequence; S6: Superimpose the predicted values ​​of each subsequence to obtain the final prediction result.

2. The method according to claim 1, characterized in that: In S1, the preprocessing method is: using the average interpolation method to fill in the missing data in the historical wind speed data.

3. The method according to claim 1, characterized in that S2 includes the following steps: S21. Divide the wind speed sequence data set into multiple subsequences according to the time series, and add positive and negative paired white noise signals to each subsequence in sequence to form a noise sequence data set: In the formula, They represent the data after adding positive white noise and negative white noise to the data at time t in the i-th group of noise sequence data sets; X(t) represents the data at time t in the wind speed sequence data sets; a i is the amplitude of the white noise signal, i=1,2,…,M; n i (t) represents the t-th i-th group of white noise signals; S22. Perform 2M EMD on the noise sequence data set to obtain multiple first-order component sequences: Where, IMF1 is the first-order component sequence; S23, calculate the box-counting dimension of IMF1, and compare it with the box-counting dimension threshold. If the box-counting dimension of IMF1 is greater than , it is regarded as an abnormal signal, otherwise it is output as a stable signal; S24. Repeat S21 to S23 until multiple modal components are output.

4. The method according to claim 1, characterized in that: S3 includes the following steps: S31. Establish a full parameter continued fraction model: Where a0, a1, …, a n-1 and b0,b1,…,b n-1 is the structural parameter of the model; S32. Contrast quotient construction: In the formula, d is the model order, β k-d is the parameter; S33. Construct a general continued fraction model: In the formula, [α k-d ,…,α k-1 β k-d ,…β k-1 ] and order d are parameters of the general continued fraction model;

Images