A short-term offshore wind power combination forecasting method
Through the LSTM-LSSVM combined prediction model and Fourier series residual correction, the problem of insufficient model robustness in offshore wind power prediction is solved, and higher prediction accuracy and system stability are achieved.
Patent Information
- Application Number
- CN202410277481.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-12
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-03-12
AI Technical Summary
In the offshore wind power prediction, mathematical statistical models are not robust enough to nonlinear wind power power data, while traditional machine learning algorithms are difficult to effectively express the information in the data, resulting in inaccurate prediction results.
The LSTM-LSSVM combination prediction model is adopted, and the LSTM model is optimized through the improved seagull algorithm and the whale algorithm are optimized, and the weight is calculated by the optimal combination prediction method, combined with Fourier series residual correction, the prediction accuracy is improved.
It improves the accuracy and stability of offshore wind power power prediction, enhances the safety and efficiency of the system, and overcomes the shortcomings of a single model.
Smart Images

Figure CN118040678B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of short-term offshore wind power prediction, and in particular to a short-term offshore wind power combination prediction method. Background Art
[0002] At present, the world is facing a series of severe challenges such as increasingly severe energy resource constraints, prominent ecological and environmental problems, and climate change. As a renewable and clean energy source, wind energy has broad development prospects, especially offshore wind power development has great potential. However, affected by severe climate and extreme weather, offshore wind speed shows strong randomness and intermittent characteristics, which brings challenges to the design of offshore wind turbines. At the same time, it will cause the power of offshore wind turbines to fluctuate violently, and in severe cases, it will cause the wind turbine to shut down, affecting the safe and stable operation of the system. Therefore, studying the offshore wind power prediction method and improving the accuracy of offshore wind power prediction in the short term are of great significance to determine the typical working conditions of offshore wind farms, improve the efficiency of offshore wind farms, and ensure the safe and stable operation of offshore wind power systems.
[0003] Currently, there are two main methods for predicting offshore wind power: one is an algorithm based on mathematical statistical models, and the other is an algorithm based on machine learning. The mathematical statistical model algorithm stands out for its significant advantages such as its concise and clear model structure, efficient computing power, and fast prediction speed. However, for complex nonlinear wind power data, due to the lack of robustness of statistical algorithms, it is difficult to obtain accurate prediction results using statistical models, resulting in a significant increase in errors. For nonlinear wind power data, machine learning algorithms can make more accurate predictions, but the amount of wind power data is huge and the situation is more complicated. Traditional machine learning algorithm models can set fewer parameters and cannot effectively express the information hidden in the data, and cannot obtain accurate prediction results. Summary of the invention
[0004] The purpose of the present invention is to provide a short-term offshore wind power combination prediction method, so as to overcome the shortcomings of the prior art and ultimately achieve accurate prediction of short-term offshore wind power.
[0005] To achieve the above object, the present invention provides a short-term offshore wind power combination prediction method, comprising:
[0006] Obtain offshore wind power data, input the offshore wind power data into the LSTM-LSSVM combined prediction model, and obtain a short-term offshore wind power prediction result;
[0007] Among them, the LSTM-LSSVM combined prediction model is obtained by training a training data set, and the LSTM-LSSVM combined prediction model includes an LSTM prediction model based on an improved seagull algorithm and an LSSVM prediction model based on an improved whale algorithm. When the LSTM prediction model and the LSSVM prediction model are combined, corresponding weights are assigned, and the training data set includes historical offshore wind power data.
[0008] Preferably, training the LSTM-LSSVM combined prediction model comprises:
[0009] Inputting the training data set into the LSTM prediction model based on the improved Seagull algorithm, and outputting a short-term offshore wind power prediction result 1;
[0010] Inputting the training data set into the LSSVM prediction model based on the improved whale algorithm, and outputting a second short-term offshore wind power prediction result;
[0011] Based on the short-term offshore wind power forecast result 1 and the short-term offshore wind power forecast result 2, the weight of the LSTM-LSSVM combined forecast model is calculated by an optimal combined forecasting method;
[0012] The weight of the LSTM-LSSVM combined prediction model is obtained according to the weight, and the LSTM-LSSVM combined prediction model is corrected using the Fourier series residual.
[0013] Preferably, the LSTM prediction model based on the improved Seagull algorithm uses the improved Seagull algorithm to optimize parameters of the long short-term memory network LSTM model.
[0014] Preferably, outputting the short-term offshore wind power prediction result 1 includes:
[0015] Initialize the LSTM model parameters and the parameters of the improved Seagull HSOA algorithm;
[0016] The mean square error obtained by training the LSTM model is used as a fitness value, and the fitness value is updated in real time;
[0017] Within the iteration range, calculating the position of the seagull, and updating the position of the seagull according to the fitness value;
[0018] According to the optimal parameter combination, prediction is performed through the LSTM model to output the short-term offshore wind power prediction result one.
[0019] Preferably, the LSSVM prediction model based on the improved whale algorithm uses the whale algorithm BWOA to optimize the parameters of the least squares support vector machine LSSVM.
[0020] Preferably, outputting the second short-term offshore wind power prediction result includes:
[0021] Collect and preprocess offshore wind power data, and divide the data into training and test sets;
[0022] Initialize the parameters of the whale algorithm BWOA to optimize the parameters and obtain the optimal prediction model;
[0023] The test set is introduced into the optimal prediction model to perform offshore wind power prediction, and the second short-term offshore wind power prediction result is output.
[0024] Preferably, the method for calculating the weight of the LSTM-LSSVM combined prediction model is:
[0025] Construct the objective function based on the minimum error sum of squares criterion:
[0026]
[0027] Among them, P is the objective function, E is the error matrix of the combined prediction error, R is an n-dimensional column vector whose elements are all 1, and T is the transposed sign.
[0028] Preferably, the LSTM-LSSVM combined prediction model is modified, including:
[0029] Assume that the actual offshore wind power is l j , the offshore wind power forecast value is Define the error sequence E j for:
[0030]
[0031] Expanding the error sequence based on Fourier series yields:
[0032]
[0033] Among them, a0, a i and b i (i=1,2,3,…) is the Fourier series coefficient, k a is the series of Fourier polynomials, T a is the period of the periodic function E(j), i is 1 to k a The ordinal number of , j is an ordinal number from 2 to n;
[0034] Arrange formula (3) into a matrix form:
[0035] Q a M a =E a (4)
[0036] In the formula,
[0037]
[0038] M a =[a0,a1,b1,a2,b2,…,a ka ,b ka ] T (6)
[0039] E a ={E(2),E(3),…,E(n)} (7)
[0040] Where, T a =n-1, n is the number of sequence elements of the predicted value and actual value sequence, M a is the residual Fourier coefficient vector, a ka and b ka For the kth a Fourier series coefficients of the series, Q a is the Fourier polynomial matrix, M a is the Fourier series coefficient vector, E a is the residual sequence;
[0041] By the least squares method, we can obtain:
[0042]
[0043] In the formula, Q a The transpose of
[0044] The required values a0,a1,b1,a2,b2,…,a ka ,b ka Substituting, let j=n+1, j=n+2,…, we can obtain the prediction error value of the combined model for the actual value, thereby obtaining the prediction value after residual correction.
[0045] Preferably, the predicted value after the residual correction is:
[0046]
[0047]
[0048] In the formula, is the predicted value of the model before residual correction, is the predicted value after residual correction, E a (j) is the jth residual value, is the first predicted value after residual correction, is the first predicted value before residual correction.
[0049] Compared with the prior art, the present invention has the following advantages and technical effects:
[0050] The present invention calculates the weights of the two prediction models, LSTM and LSSVM, after parameter optimization by using the minimum square error criterion to obtain the optimal combined weight, and calculates the prediction results of each model as the combined prediction result. Compared with the prediction results of a single prediction model, the combined prediction has higher accuracy. After that, considering that the power sequence has a certain periodicity, the Fourier series residual is used to correct the combined prediction result, which further improves the prediction accuracy of offshore wind power, and is of great significance in improving the efficiency of offshore wind farms and ensuring the safe and stable operation of offshore wind power systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0052] Figure 1 LSTM structure diagram of an embodiment of the present invention;
[0053] Figure 2 This is a HSOA-LSTM model prediction flow chart of an embodiment of the present invention;
[0054] Figure 3 It is a BWOA-LSSVM model process framework diagram of an embodiment of the present invention;
[0055] Figure 4 A comparison diagram of the effects of the combined prediction model and a single model in an embodiment of the present invention;
[0056] Figure 5 The R coefficient analysis diagram of the model prediction results of the embodiments of the present invention, wherein (a) is the R value diagram of the HSOA-LSTM model, (b) is the R value diagram of the BWOA-LSSVM model, and (c) is the R value diagram of the combined prediction model. DETAILED DESCRIPTION
[0057] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0058] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0059] A short-term offshore wind power combination forecasting method, comprising:
[0060] Obtain offshore wind power data, input the offshore wind power data into the LSTM-LSSVM combined prediction model, and obtain the short-term offshore wind power prediction results;
[0061] Among them, the LSTM-LSSVM combined prediction model is obtained by training the training data set, and the LSTM-LSSVM combined prediction model includes an LSTM prediction model based on the improved Seagull algorithm and an LSSVM prediction model based on the improved Whale algorithm.
[0062] Specifically, the LSTM-LSSVM combined prediction model is trained, including:
[0063] The training data set is input into the LSTM prediction model based on the improved Seagull algorithm, and the short-term offshore wind power prediction result 1 is output;
[0064] The training data set is input into the LSSVM prediction model based on the improved whale algorithm, and the short-term offshore wind power prediction result 2 is output;
[0065] Based on the short-term offshore wind power forecast result 1 and the short-term offshore wind power forecast result 2, the weights of the LSTM-LSSVM combined forecast model are calculated by the optimal combined forecast method;
[0066] The weight of the LSTM-LSSVM combined prediction model is obtained according to the weight, and the LSTM-LSSVM combined prediction model is corrected using the Fourier series residual.
[0067] The LSTM prediction model based on the improved Seagull algorithm uses the improved Seagull algorithm to optimize the parameters of the long short-term memory network LSTM model. The LSSVM prediction model based on the improved whale algorithm uses the whale algorithm BWOA to optimize the parameters of the least squares support vector machine LSSVM.
[0068] Specifically, the output is a short-term offshore wind power forecast result 1, including:
[0069] Initialize the LSTM model parameters and the parameters of the improved Seagull HSOA algorithm;
[0070] The mean square error obtained by LSTM model training is used as the fitness value, and the fitness value is updated in real time;
[0071] Within the iteration range, the position of the seagull is calculated, and the position of the seagull is updated according to the fitness value;
[0072] According to the optimal parameter combination, the LSTM model is used to make predictions and output the short-term offshore wind power prediction result 1.
[0073] Output short-term offshore wind power forecast results 2, including:
[0074] Collect and preprocess offshore wind power data, and divide the data into training and test sets;
[0075] Initialize the parameters of the whale algorithm BWOA to optimize the parameters and obtain the optimal prediction model;
[0076] The test set is introduced into the optimal prediction model to perform offshore wind power prediction, and a second short-term offshore wind power prediction result is output.
[0077] The method for calculating the weights of the LSTM-LSSVM combined prediction model is:
[0078] Construct the objective function based on the minimum error sum of squares criterion:
[0079]
[0080] Among them, P is the objective function, E is the error matrix of the combined prediction error, R is an n-dimensional column vector whose elements are all 1, and T is the transposed sign.
[0081] Solved by Lagrangian method, the optimal combination weight vector is: The minimum sum of squares of the combined prediction error is the objective function Under the constraint R T When W=1, that is, when the sum of the weights is 1, the sum of squares of the combined prediction error is minimized, that is, the objective function is minimized.
[0082] Modifications to the LSTM-LSSVM combined prediction model include:
[0083] Assume that the actual offshore wind power is l j , the offshore wind power forecast value is Define the error sequence E j for:
[0084]
[0085] Expanding the error sequence based on Fourier series yields:
[0086]
[0087] Among them, a0, a i and b i (i=1,2,3,…) is the Fourier series coefficient, k a Fourier polynomial series, numerically [] represents the nearest integer not exceeding, T a is the period of the periodic function E(j), numerically T a =n-1, i is 1 to k a The ordinal number of , j is an ordinal number from 2 to n;
[0088] Arrange formula (3) into a matrix form:
[0089] Q a M a =E a (4)
[0090] In the formula,
[0091]
[0092] M a =[a0,a1,b1,a2,b2,…,a ka ,b ka ] T (6)
[0093] E a ={E(2),E(3),…,E(n)} (7)
[0094] Where, T a =n-1, n is the number of sequence elements of the predicted value and actual value sequence, M a is the residual Fourier coefficient vector, a ka and b ka For the kth a Fourier series coefficients of the series, Q a is the Fourier polynomial matrix, M a is the Fourier series coefficient vector, E a is the residual sequence;
[0095] By the least squares method, we can obtain:
[0096]
[0097] In the formula, Q a The transpose of
[0098] The required values a0,a1,b1,a2,b2,…,a ka ,b ka Substituting, let j=n+1, j=n+2,..., we can obtain the prediction error value of the combined model for the actual value, thereby obtaining the prediction value after residual correction.
[0099] Furthermore, the predicted value after residual correction is:
[0100]
[0101]
[0102] In the formula, is the predicted value of the model before residual correction, is the predicted value after residual correction, E a (j) is the jth residual value, is the first predicted value after residual correction, is the first predicted value before residual correction.
[0103] In order to more clearly express the technical solution of the present invention, the following specific embodiments are provided to introduce the solution:
[0104] Obtain offshore wind power data, input the offshore wind power data into the LSTM-LSSVM combined prediction model, and obtain the short-term offshore wind power prediction results;
[0105] Among them, the LSTM-LSSVM combined prediction model is obtained by training the training data set, and the LSTM-LSSVM combined prediction model includes an LSTM prediction model based on the improved Seagull algorithm and an LSSVM prediction model based on the improved Whale algorithm.
[0106] Training the LSTM-LSSVM combined prediction model includes:
[0107] Substitute the training data into the LSTM prediction model based on the improved Seagull algorithm and output the short-term offshore wind power prediction result 1;
[0108] Substitute the training data into the LSSVM prediction model based on the improved whale algorithm and output the short-term offshore wind power prediction result 2;
[0109] According to the prediction results of the training data, the weights of the two prediction models are calculated based on the minimum error square sum criterion;
[0110] According to the calculation results of the weights, the weights of the two models in the LSTM-LSSVM combined prediction model are obtained, and the model is corrected using the Fourier series residual.
[0111] Specifically, the LSTM prediction model based on the improved Seagull algorithm uses the improved Seagull algorithm to optimize the parameters of the long short-term memory network (LSTM).
[0112] In the LSTM prediction model, the hidden layer of LSTM consists of one or more memory units, each of which includes a forget gate, an input gate and an output gate. Its structure is as follows: Figure 1shown.
[0113] exist Figure 1 In the network topology, the corresponding formula is:
[0114] Forget Gate:
[0115] f t =σ(W f ×[h t-1 ,x t ]+B f )
[0116] Input Gate:
[0117] i t =σ(W i ×[h t-1 ,x t ]+B i )
[0118]
[0119] Output Gate:
[0120] O t =σ(W o ×[h t-1 ,x t ]+B o )
[0121]
[0122] h t =o t ×tanh(c t )
[0123] Where W f is the weight matrix of the forget gate; B f is the bias term; σ(·) represents the sigmoid function. t is the result of the input gate; W i is a weight matrix, which is an input gate; B i Represents the input bias, which is the starting point of the input; Represents the current input unit status; W c is a matrix with weights; B c Represents the unit state bias; the matrix [h t-1 ,x t ] consists of two vectors, the output value h at the previous moment t-1 and the current input value x t ; σ(·) represents the sigmoid function, and tanh(·) is a double tangent function. W ois the weight matrix of the output gate; B o is the bias term of the output gate.
[0124] f t is the output of the forget gate, x t is the current input information, i t is the input gate state, is the temporary memory unit state, O t is the output result of information processed by the output gate, c t is the updated memory cell state, c t-1 To update the memory unit state at the previous moment, h t is the updated hidden layer state.
[0125] The optimization work of improving the Seagull algorithm for the LSTM model is mainly divided into four parts:
[0126] (1) Determine the LSTM network topology according to the above formula and initialize all relevant parameters.
[0127] (2) Select the parameter search range based on the random value determined by the parameter to be optimized.
[0128] (3) Calculate the fitness value corresponding to the individual seagull. The fitness value is set as the root mean square error (RMSE) of the model validation data set. If the calculated fitness is the smallest in the world, it is set as the best fitness of this time, that is, the best position of the seagull, and compared with the global best fitness value. If the value is smaller than the global best fitness value, it is replaced.
[0129] (4) Start iteration. Use the HSOA algorithm to continuously update the three hyperparameters that need to be optimized, and repeat steps (3) to (4) until the maximum number of iterations is reached.
[0130] The flow chart of HSOA-LSTM prediction model is as follows: Figure 2 As shown, the specific steps are as follows:
[0131] (1) Initialize LSTM parameters (T iter ,P R , P1, P2) and HSOA algorithm parameters, including population number N = 5, maximum number of iterations M = 20, parameter U b Upper limit, parameter L b lower limit, dimension D and initialize sample data.
[0132] (2) The mean square error obtained from LSTM training is used as the fitness value, and the data of the fitness value is calculated. The fitness value is updated in real time as the seagull continues to hunt. Within the iteration range, the position of the seagull is calculated. If the current new position is better, the old position is updated.
[0133] (3) Based on the optimal parameter combination, the LSTM model is used to make predictions and output the short-term offshore wind power prediction result 1.
[0134] Specifically, the LSSVM prediction model based on the improved whale algorithm uses the whale algorithm to optimize the parameters of the least squares support vector machine (LSSVM).
[0135] In the LSSVM prediction model, the normalization parameter reflects the distribution characteristics of the training sample data, and the penalty factor determines the size of the training error and the strength of the generalization ability. They are two important hyperparameters that affect the prediction performance of the LSSVM model. In the traditional LSSVM offshore wind power prediction, the penalty factor and the normalization parameter are often selected based on experience. In order to avoid blindly selecting hyperparameters and improve the prediction accuracy of the model, this embodiment uses the whale algorithm (BWOA) to optimize the two parameters, such as Figure 3 shown.
[0136] The optimization work of the whale algorithm for the LSSVM model is mainly divided into three parts:
[0137] (1) Collect offshore wind power and preprocess it, dividing the data into training and test sets.
[0138] (2) Initialize the BWOA parameters and perform parameter optimization to obtain the optimal prediction model.
[0139] (3) The test set data is introduced into the model to predict offshore wind power and output the short-term offshore wind power prediction result 2.
[0140] The LSTM prediction model based on the improved Seagull algorithm and the LSSVM prediction model based on the improved Whale algorithm are combined to obtain the LSTM-LSSVM combined prediction model, so as to predict the short-term offshore wind power.
[0141] Specifically, in the process of combining the two prediction methods into a prediction method based on LSTM-LSSVM, the weights of the two prediction methods can be calculated based on the minimum criterion of the sum of squared errors.
[0142] Combined prediction can be expressed as: for a prediction object, first use n (n ≥ 2) prediction methods, take the actual value at the tth moment as R t , the prediction value and error of the i-th method at the tth moment are P it and δ it ,in:
[0143] δ it =R t -P it ,(i=1,2,…,n,t=1,2,…,N)
[0144] Assume that the weight of the prediction result of the i-th method in the combined model is ω i The following conditions are met:
[0145]
[0146] Similar to ω i Normalization is done to maintain the accuracy of the combined model predictions.
[0147] The predicted value at time t is recorded as P t , the prediction error is recorded as δ t ,So:
[0148]
[0149]
[0150] According to the rules for improving prediction accuracy, the combination prediction method is divided into two types: the optimal combination prediction method and the combination method of forecasting science. The optimal combination prediction method first constructs an objective function based on the forecast object, and then solves the weight coefficient that makes the objective function reach the maximum or minimum value. The weight coefficient obtained in this way must be optimal. The combination method of forecasting science solves the formula of the weight coefficient based on the basic principles of forecasting science, that is, the variance of the error of each method in the combination prediction method is inversely proportional to the weight coefficient.
[0151] This embodiment selects the optimal combination prediction method and constructs the objective function based on the minimum error square sum criterion:
[0152] Let P be the sum of squares of the prediction errors of the combined prediction model, then:
[0153]
[0154] This optimization model can be expressed as:
[0155]
[0156] Let W = (ω1,ω2,...,ω n ) T ,δ i =(δ i1 ,δ i2 ,...,δ iN ) T , it can be seen that W represents the weight coefficient vector, δ i is the prediction error vector of the i-th prediction method in the combined prediction model, if:
[0157]
[0158] but:
[0159] E=(E ij ) n×n
[0160] Where E is the error matrix of the combined prediction error.
[0161] In general, it is necessary to select n (n < N) sets of error vectors for single predictions that are linearly independent, i.e., δ1, δ2, ..., δ n Linearly independent, E is a reversible positive definite matrix. Then:
[0162]
[0163]
[0164] In the formula, R T =(1,1,...,1) 1×n . Therefore, the optimization model can be expressed in the form of the following matrix:
[0165]
[0166] Introducing the Lagrange multiplier λ, solving for W yields:
[0167]
[0168] The optimal weight vector can be obtained from the above formula, so the objective function is:
[0169]
[0170] After obtaining the weights of the two models in the LSTM-LSSVM combined prediction model, the model is corrected using the Fourier series residual.
[0171] Fourier is a periodic function that can extract the periodic information hidden in the sample data and reduce the prediction error caused by random factors in the sequence data. Therefore, the Fourier series residual correction model can apply Fourier, which has the characteristics of periodic functions, to error correction and improve the problem of large exponential growth rate sequence errors in the offshore wind power forecast series in different periods.
[0172] Assume that the actual offshore wind power is l j , the offshore wind power forecast value is The error sequence is defined as:
[0173]
[0174] Expanding the error sequence using Fourier series yields:
[0175]
[0176] The above formula is a trigonometric series, where a0, a i and b i (i=1,2,3,…) is the Fourier series coefficient. The above formula can be organized into a matrix form to obtain:
[0177] Q a M a =E a
[0178] In the formula,
[0179]
[0180] M a =[a0,a1,b1,a2,b2,…,a ka ,b ka ] T
[0181] E a ={E(2),E(3),…,E(n)}
[0182] Where, T a =n-1, M a is the residual Fourier coefficient vector.
[0183] The solution can be obtained by the least squares method:
[0184]
[0185] The obtained a0,a1,b1,a2,b2,…,a ka ,b ka Substituting j=n+1, j=n+2,..., we can find the prediction error of the combined model for the actual value, and thus find the prediction value after residual correction.
[0186] The predicted value after residual correction is:
[0187]
[0188]
[0189] In the formula, is the predicted value of the model before residual correction.
[0190] The offshore wind power data is brought into the LSTM-LSSVM combined prediction model to obtain the short-term offshore wind power prediction results.
[0191] In order to further understand the performance of the prediction model, the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are selected as the evaluation indicators of the model, and the calculation method is as follows:
[0192]
[0193]
[0194]
[0195] In order to prove the effectiveness of the method, the following simulation scenarios are set up for comparative analysis:
[0196] The data set used in this embodiment is an offshore wind power data set of a certain place, which includes wind speed data and wind power data of a certain place. The complete data set includes data from January 1, 2018 to December 31, 2018. This embodiment uses data from February 1, 2018 to March 30, 2018 for training, and uses wind power data at each moment on March 31, 2018 for prediction verification. In order to verify the accuracy of the combined prediction method, its prediction results are compared with the single prediction results of LSTM and LSSVM respectively. The prediction comparison results are as follows: Figure 4 Indicated.
[0197] Depend on Figure 4 It can be seen that the prediction effect of the minimum sum of squared errors combined model after Fourier series residual correction is significantly improved compared with the single prediction model. Among them, it is very consistent with the true power value from 0:00 to 4:00 in the morning and around 12:00 noon, and the effect is worse than that of the single model around 5:00 and 18:00, which may be related to the Fourier series residual correction method. If the error changes periodically, the correction effect of the prediction model is better. If the relative error does not change periodically, the model prediction error increases after the Fourier series error correction. However, it can be clearly seen in Table 1 that the prediction error of the combined prediction is significantly lower than that of the two single prediction models HSOA-LSTM and BWOA-LSSVM.
[0198] Table 1
[0199]
[0200] Figure 5The R coefficient analysis diagram of the combined prediction model and two single prediction models, among which (a) is the R value diagram of the HSOA-LSTM model, (b) is the R value diagram of the BWOA-LSSVM model, and (c) is the R value diagram of the combined prediction model. The regression value R represents the correlation between the predicted output and the target output. The closer the R value is to 1, the closer the relationship between the predicted and output data is. The closer the R value is to 0, the greater the randomness of the relationship between the predicted and output data is. The R value of the combined prediction is 0.99742, the R value of the HSOA-LSTM model is 0.99553, and the R value of the BWOA-LSSVM model is 0.99687. It can be concluded that compared with the single prediction model, the relationship between the predicted value and the actual value of the combined prediction model is closer, and it has a better prediction effect.
[0201] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A short-term offshore wind power combination forecasting method, characterized in that: include: Obtain offshore wind power data, input the offshore wind power data into the LSTM-LSSVM combined prediction model, and obtain a short-term offshore wind power prediction result; The LSTM-LSSVM combined prediction model is obtained by training a training data set, the LSTM-LSSVM combined prediction model includes an LSTM prediction model based on an improved seagull algorithm and an LSSVM prediction model based on an improved whale algorithm, corresponding weights are assigned when the LSTM prediction model and the LSSVM prediction model are combined, and the training data set includes historical offshore wind power data; Training the LSTM-LSSVM combined prediction model includes: Inputting the training data set into the LSTM prediction model based on the improved Seagull algorithm, and outputting a short-term offshore wind power prediction result 1; Inputting the training data set into the LSSVM prediction model based on the improved whale algorithm, and outputting a second short-term offshore wind power prediction result; Based on the short-term offshore wind power forecast result 1 and the short-term offshore wind power forecast result 2, the weight of the LSTM-LSSVM combined forecast model is calculated by an optimal combined forecasting method; Obtaining the weight of the LSTM-LSSVM combined prediction model according to the weight, and correcting the LSTM-LSSVM combined prediction model using Fourier series residuals; The method for calculating the weights of the LSTM-LSSVM combined prediction model is: Construct the objective function based on the minimum error sum of squares criterion: Where P is the objective function, E is the error matrix of the combined prediction error, R is an n-dimensional column vector whose elements are all 1, and T is the transposed symbol; The LSTM-LSSVM combined prediction model is modified, including: Assume that the actual offshore wind power is l j The offshore wind power forecast is Define the error sequence E j for: The error sequence is expanded based on the Fourier series: Among them, a0, a i and b i is the Fourier series coefficient, k a is the series of Fourier polynomials, T a is the period of the periodic function E(j) and T a =n-1, i is 1 to k a The ordinal number of , j is an ordinal number from 2 to n; Arrange formula (3) into matrix form: Q a M a =E a (4) In the formula, <h2 style=";text-align:left;direction:ltr">M<h2 style=";text-align:left;direction:ltr"> a <h2 style=";text-align:left;direction:ltr"> (a0,a1,b1,a2,b2,…,a)<h2 style=";text-align:left;direction:ltr"> ka <h2 style=";text-align:left;direction:ltr"> ,b<h2 style=";text-align:left;direction:ltr"> ka <h2 style=";text-align:left;direction:ltr"> ]<h2 style=";text-align:left;direction:ltr"> T <h2 style=";text-align:left;direction:ltr"> (6) Ea={E(2),E(3),…,E(n)}(7) In the formula, n is the number of sequence elements of the predicted value and actual value sequence, M a is the residual Fourier coefficient vector, a ka and b ka For the kth a Fourier series coefficients of the series, Q a is the Fourier polynomial matrix, M a is the Fourier series coefficient vector, E a is the residual sequence; By the least squares method, we can obtain: In the formula, Q a The transpose of The required values a0,a1,b1,a2,b2,…,a ka ,b ka Substitute, let j = n + 1, j = n + 2, ..., to obtain the prediction error value of the combined model for the actual value, thereby obtaining the prediction value after residual correction; The predicted value after the residual correction is: In the formula, is the predicted value of the model before residual correction, is the predicted value after residual correction, E a (j) is the jth residual value, is the first predicted value after residual correction, is the first predicted value before residual correction.
2. The short-term offshore wind power combination forecasting method according to claim 1 is characterized in that: The LSTM prediction model based on the improved Seagull algorithm uses the improved Seagull algorithm to optimize the parameters of the long short-term memory network LSTM model.
3. The short-term offshore wind power combination prediction method according to claim 2 is characterized in that: Outputting the short-term offshore wind power forecast result 1, including: Initialize the LSTM model parameters and the parameters of the improved Seagull HSOA algorithm; The mean square error obtained by training the LSTM model is used as a fitness value, and the fitness value is updated in real time; Within the iteration range, calculating the position of the seagull, and updating the position of the seagull according to the fitness value; According to the optimal parameter combination, prediction is performed through the LSTM model to output the short-term offshore wind power prediction result one.
4. The short-term offshore wind power combination forecasting method according to claim 1 is characterized in that: The LSSVM prediction model based on the improved whale algorithm uses the whale algorithm BWOA to optimize the parameters of the least squares support vector machine LSSVM.
5. The short-term offshore wind power combination forecasting method according to claim 4 is characterized in that: Outputting the second short-term offshore wind power forecast result, including: Collect and preprocess offshore wind power data, and divide the data into training and test sets; Initialize the parameters of the whale algorithm BWOA to optimize the parameters and obtain the optimal prediction model; The test set is brought into the optimal prediction model to perform offshore wind power prediction, and the second short-term offshore wind power prediction result is output.
Citation Information
Patent Citations
Short-term Load Forecasting Method Based on TCN and IPSO-LSSVM Combined Model
AU2020104000A4
Wind power combination prediction method based on optimal weight coefficient
CN105303250A
Long-term power load prediction method based on residual correction grey prediction model
CN110163447A
Grey model and support vector machine-based short-term wind power prediction method
CN112001537A
Short-term wind power prediction method and device
CN112686464A
