A method for ultra-short-term wind power prediction based on secondary decomposition and IWOA-LSSVM

Abstract

The present application relates to the technical field of wind power prediction, in particular to a kind of ultra-short-term wind power prediction method based on secondary decomposition and IWOA-LSSVM, the method comprises: using CEEMD to carry out primary decomposition to original wind power sequence, and the sample entropy of all components after primary decomposition is calculated;Part of sub-sequence with large sample entropy is secondarily decomposed using VMD;The LSSVM model parameter is optimized using improved whale optimization algorithm, and IWOA-LSSVM prediction model is established to predict all components obtained by secondary decomposition;All component prediction results are superimposed, and the final wind power prediction result is obtained, and the error of prediction result is verified.Using the method, the problems of high complexity of original wind power sequence, slow convergence speed of optimization algorithm and low model prediction accuracy are effectively solved, and the method has the advantages of high accuracy and good application prospect.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, specifically to an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM. Background Technology

[0002] In recent years, with the development of new energy sources, the installed capacity of renewable energy power generation, represented by wind power and photovoltaic power, has continued to increase in the power system. Wind power is characterized by strong randomness, significant intermittency, large fluctuation amplitude, irregular fluctuation frequency, and anti-peak shaving. These uncertainties increase the potential risks to the stable operation of the power grid and the difficulty of peak shaving. Accurate wind power forecasting not only helps grid dispatchers formulate reasonable power generation plans, but also provides strategies for wind farm dispatching and energy storage, reducing the power generation cost of wind farms.

[0003] Generally, wind power prediction methods can be divided into two main categories: physical methods and statistical methods. Physical prediction methods study wind speed changes using physical laws and boundary conditions, typically exhibiting good performance in long-term wind speed prediction. This method requires consideration of geographical information such as topography and location, meteorological information such as wind direction, air pressure, temperature, and humidity, and must be organically combined with physical numerical weather prediction (PNF) for wind power prediction. However, its computational complexity is high, the calculation time is long, and the update speed of PNF is slow, making it difficult to obtain accurate wind power prediction values. Statistical methods are currently the more widely studied power prediction methods. Compared to physical methods, statistical methods offer superior prediction results. Statistical methods primarily utilize historical data from wind farms, combined with intelligent algorithms to build prediction models, demonstrating excellent performance in short-term power prediction.

[0004] Due to the strong volatility, nonlinearity, and complexity of wind power sequences, many existing technologies employ EMD (Empirical Mode Decomposition) and EEMD (Ensemble Empirical Mode Decomposition) to reduce their complexity. However, EMD decomposition is prone to mode aliasing, while EEMD decomposition is subject to noise interference. Meanwhile, numerous prediction models also have their own shortcomings. For example, the BP neural network (a multi-layer feedforward network trained using the backpropagation algorithm) has a simple structure and unstable prediction accuracy. Optimizing model parameters using optimization algorithms can improve the model's prediction accuracy, but many optimization algorithms also have their own deficiencies. For example, the convergence speed of the genetic optimization algorithm needs to be improved.

[0005] Therefore, there is an urgent need for an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM. Summary of the Invention

[0006] To address the challenges of high complexity in analyzing original wind power sequences, slow convergence of optimization algorithms, and low model prediction accuracy, this invention provides an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM.

[0007] To achieve the above objectives, this invention provides an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM, the method comprising:

[0008] S1. Use CEEMD to decompose the original wind power sequence once, and calculate the sample entropy of all components after the first decomposition.

[0009] S2. Perform secondary decomposition of the subsequences with high sample entropy using VMD.

[0010] S3. The improved whale optimization algorithm is used to optimize the LSSVM model parameters, and the IWOA-LSSVM prediction model is established to predict all components obtained by the second decomposition.

[0011] S4. Superimpose all component prediction results to obtain the final wind power prediction result, and verify the error of the prediction result.

[0012] Preferably, in step S1, the original wind power sequence is decomposed using CEEMD, and the sample entropy of all components after the first decomposition is calculated, specifically including:

[0013] S1.1, in the original wind power sequence x i Positive and negative noise signals are added to (t) respectively. and This generates a new set of signals;

[0014] S1.2 Perform EMD decomposition on the generated new set of signals;

[0015] S1.3. Take the average value of the several IMF components obtained from the decomposition;

[0016] S1.4 Calculate the sample entropy of all components.

[0017] Preferably, in step S1.2, performing EMD decomposition on the generated new set of signals specifically includes:

[0018] S1.2.1. The upper and lower envelopes U of the sequence are obtained using cubic spline interpolation. i (t) and L i (t);

[0019] Among them, U i (t)≥X i (t)≥L i (t);

[0020] S1.2.2 Calculate the mean m of the upper and lower envelopes. i (t):

[0021]

[0022] S1.2.3 Extracting local detail information h of the signal i (t):

[0023] h i (t)=X i (t)-m i (t)

[0024] S1.2.4, Determine h i (t) Whether conditions a and b are satisfied at the same time. If conditions a and b are satisfied at the same time, proceed to S1.2.5; otherwise, return to execute step S1.2.1. Wherein, condition a is that the number of points is equal to or differs from the number of zero-crossing points in the sequence by one, and condition b is that at any time point, the average value of the upper and lower envelopes is zero.

[0025] S1.2.5 Calculate the remaining data r i (t):

[0026] r i (t)=X i (t)-h i (t)

[0027] S1.2.6 Repeat steps S1.2.1-S1.2.5 to obtain N IMF components until equation (1) is satisfied and the decomposition stops.

[0028]

[0029] Where M is the number of screening attempts and n is the number of samples;

[0030] S1.2.7 The final EMD decomposition result is as follows:

[0031]

[0032] Where N is the total number of IMFs that are decomposed.

[0033] Preferably, in step S2, the secondary decomposition of the subsequences with high sample entropy using VMD specifically includes:

[0034] S2.1 Constructing the variational optimization function:

[0035]

[0036]

[0037] Among them, {u k} represents the decomposition components, {ω k} represents the center frequency, * represents convolution, δ(t) is the impulse function, and K is the number of decompositions. To find the partial derivative with respect to t;

[0038] S2.2 Solving the variational optimization function using the Lagrange method:

[0039]

[0040] Where α is the penalty factor, α>0, λ is the operator, and <·,·> denote the inner product;

[0041] S2.3 Combining the alternating direction multiplier algorithm and Fourier transform, the final iterative expression is obtained:

[0042]

[0043]

[0044]

[0045] Where the step size τ > 0.

[0046] Preferably, in step S3, the improved whale optimization algorithm is used to optimize the LSSVM model parameters, and the IWOA-LSSVM prediction model is established to predict all components obtained by the secondary decomposition. Specifically, this includes:

[0047] S3.1 Improve the whale optimization algorithm;

[0048] S3.2. An improved whale optimization algorithm is used to optimize the penalty factor γ and kernel function parameter σ in LSSVM.

[0049] S3.3 Establish the IWOA-LSSVM model to predict the wind power subsequences after the second decomposition.

[0050] Preferably, in step S3.1, improving the whale optimization algorithm specifically includes:

[0051] a. Introducing the Cauchy mutation strategy, the specific mathematical description formula is as follows:

[0052]

[0053] Where x(t) is the position of the whale before mutation, x(t+1) is the position after mutation, and r is a random value in the range [0,1].

[0054] b. Introduce an adaptive weighting strategy, the specific mathematical description of which is:

[0055]

[0056] x(t+1)=wx * (t)-AD

[0057] x(t+1)=D'e bl cos(2πl)+wx * (t)

[0058] Where w is the adaptive weight and Max_iter is the optimization algebra.

[0059] Preferably, in step S3.1, improving the whale optimization algorithm further includes:

[0060] c. Introduce a random difference mutation strategy, the specific mathematical description of which is as follows:

[0061] x(t+1)=r1(x * (t)-x(t))+r2(x'(t)-x(t))

[0062] Where x'(t) is a randomly selected individual from the population, and r1 and r2 are random values ​​in the range [0,1].

[0063] Preferably, the solution steps for LSSVM in step S3.2 are as follows:

[0064] 1) Set up a wind power training sample set, perform spatial mapping on the samples in the set, and establish the optimal decision function:

[0065] y(x)=ωφ(x)+b

[0066] Where ω is the weight and b is the bias. For spatial mapping;

[0067] 2) Based on the minimum theory of structural risk mitigation, the optimized optimal decision function is obtained:

[0068]

[0069] y i =φ(x i )ω+b+ξ i i = 1, ..., l

[0070] Where, ξ i Let c be a slack variable, c be a penalty factor, and l be the length of the training samples.

[0071] 3) Use the Lagrange multiplier method to solve for the optimal decision function optimized in step S3.2.2, and obtain the final linear regression equation:

[0072]

[0073] Where K(x) i ,x j ) is the kernel function.

[0074] Preferably, in step S4, all component prediction results are superimposed to obtain the final wind power prediction result, and the prediction result is verified for error, specifically including:

[0075] S4.1. Superimpose all wind power subsequence prediction results to obtain the final wind power prediction result;

[0076] S4.2 Calculate the values ​​of RMSE, MAE and MAPE, and verify the error of the prediction results.

[0077] Preferably, the expressions for RMSE, MAE, and MAPE are as follows:

[0078]

[0079] Based on the above technical solution, this method effectively solves the problems of high complexity and difficulty in analyzing the original wind power sequence, slow convergence speed of the optimization algorithm, and low prediction accuracy of the model. It not only helps grid dispatchers to formulate reasonable power generation plans, but also provides strategies for wind farm scheduling and energy storage, reducing the power generation cost of wind farms, and has good application prospects. Attached Figure Description

[0080] Figure 1 This is a flowchart of an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM;

[0081] Figure 2 This is the overall flowchart of the ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM;

[0082] Figure 3 This is a schematic diagram of the original sequence of wind power in an embodiment of the present invention;

[0083] Figure 4 This is a schematic diagram of the original wind power sequence after a first decomposition by CEEMD in an embodiment of the present invention;

[0084] Figure 5 This is a schematic diagram of VMD secondary decomposition for subsequences with large sample entropy after the first decomposition in this invention.

[0085] Figure 6This is a comparison chart of the convergence curves of the improved whale optimization algorithm in this invention and other optimization algorithms;

[0086] Figure 7 This is a comparison chart of the prediction results and the actual values ​​using the method of the present invention in an embodiment of the present invention;

[0087] Figure 8 This is a comparison chart of the prediction results of the method of the present invention and other methods in the embodiments of the present invention. Detailed Implementation

[0088] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0089] This invention provides an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA (Improved Whale Optimization Algorithm)-LSSVM (Least Squares Support Vector Machine), such as... Figure 1-2 As shown, the method includes the following steps:

[0090] S1. The original wind power sequence is decomposed once using CEEMD (Complementary Set Empirical Mode Method), and the sample entropy of all components after the first decomposition is calculated.

[0091] S2. Perform secondary decomposition on the subsequences with high sample entropy using VMD (Variational Mode Decomposition).

[0092] S3. The improved whale optimization algorithm is used to optimize the LSSVM model parameters, and the IWOA-LSSVM prediction model is established to predict all components obtained by the second decomposition.

[0093] S4. Superimpose all component prediction results to obtain the final wind power prediction result, and verify the error of the prediction result.

[0094] Based on the above technical solution, this method effectively solves the problems of high complexity and difficulty in analyzing the original wind power sequence, slow convergence speed of the optimization algorithm, and low prediction accuracy of the model. It not only helps grid dispatchers to formulate reasonable power generation plans, but also provides strategies for wind farm scheduling and energy storage, reducing the power generation cost of wind farms, and has good application prospects.

[0095] According to a preferred embodiment of the present invention, in step S1, the original wind power sequence is decomposed using CEEMD, and the sample entropy of all components after the first decomposition is calculated, specifically including:

[0096] S1.1, in the original wind power sequence x i Positive and negative noise signals are added to (t) respectively. and A new set of signals is generated;

[0097] S1.2 Perform EMD decomposition on the generated new set of signals;

[0098] Specifically, this step of performing EMD decomposition on the newly generated set of signals includes:

[0099] S1.2.1. The upper and lower envelopes U of the sequence are obtained using cubic spline interpolation. i (t) and L i (t);

[0100] Among them, U i (t)≥X i (t)≥L i (t);

[0101] S1.2.2 Calculate the mean m of the upper and lower envelopes. i (t):

[0102]

[0103] S1.2.3 Extracting local detail information h of the signal i (t):

[0104] h i (t)=X i (t)-m i (t)

[0105] S1.2.4, Determine h i (t) Whether conditions a and b are satisfied at the same time. If conditions a and b are satisfied at the same time, proceed to S1.2.5; otherwise, return to execute step S1.2.1. Wherein, condition a is that the number of points is equal to or differs from the number of zero-crossing points in the sequence by one, and condition b is that at any time point, the average value of the upper and lower envelopes is zero.

[0106] S1.2.5 Calculate the remaining data r i (t):

[0107] r i (t)=X i (t)-h i (t)

[0108] S1.2.6 Repeat steps S1.2.1-S1.2.5 to obtain N IMF components until equation (1) is satisfied and the decomposition stops.

[0109]

[0110] Where M is the number of screening attempts and n is the number of samples;

[0111] S1.2.7 The final EMD decomposition result is as follows:

[0112]

[0113] Where N is the total number of IMFs that are decomposed.

[0114] S1.3. Take the average value of the several IMF components obtained from the decomposition;

[0115] S1.4 Calculate the sample entropy of all components.

[0116] In this embodiment of the invention, the CEEMD decomposition process includes EMD decomposition. Step S1.3 represents taking the average of several IMF components obtained from the EMD decomposition to obtain a certain component of the CEEMD decomposition. Step S1.4 represents calculating the sample entropy of all components obtained from the CEEMD decomposition after obtaining all CEEMD decomposition components. Specifically, in this embodiment, wind power data of a 1.5MW wind turbine in a wind farm in Hebei Province was selected for 8 days (sampling period of 5 minutes) from December 1st to December 8th, 2017. A total of 2304 points were collected as the research object, and the last 48 points were used as test data. The collected wind power data curve is shown below. Figure 3 As shown in the figure. The graph is a result of the first decomposition of the original wind power sequence by CEEMD. Figure 4 As shown.

[0117] According to a preferred embodiment of the present invention, in step S2, the secondary decomposition of the subsequences with high sample entropy using VMD specifically includes:

[0118] S2.1 Constructing the variational optimization function:

[0119]

[0120]

[0121] Among them, {u k} represents the decomposition components, {ω k} represents the center frequency, * represents convolution, δ(t) is the impulse function, and K is the number of decompositions. To find the partial derivative with respect to t;

[0122] S2.2 Solving the variational optimization function using the Lagrange method:

[0123]

[0124] Where α is the penalty factor, α>0, λ is the operator, and <·,·> denote the inner product;

[0125] S2.3 Combining the Alternating Direction Multiplier Algorithm (ADMM) and Fourier transform, the final iterative expression is obtained:

[0126]

[0127]

[0128]

[0129] Where the step size τ > 0.

[0130] In this embodiment of the invention, based on the original wind power data selected in step S1 above, after the first CEEMD decomposition, since the entropy values ​​of the first four components are significantly greater than those of other components, the first four components with larger entropy values ​​are selected for the second VMD decomposition. After the second decomposition, the original wind power sequence can be converted into several relatively stable components. At the same time, all components after the second decomposition are normalized. A schematic diagram of the VMD decomposition of one component is shown below. Figure 5 As shown.

[0131] According to a preferred embodiment of the present invention, in step S3, the improved whale optimization algorithm is used to optimize the LSSVM model parameters, and the IWOA-LSSVM prediction model is established to predict all components obtained by the quadratic decomposition. Specifically, this includes:

[0132] S3.1 Improve the whale optimization algorithm;

[0133] Specifically, the improvements to the whale optimization algorithm in this step include:

[0134] a. Introducing the Cauchy mutation strategy, the specific mathematical description formula is as follows:

[0135]

[0136] Where x(t) is the position of the whale before mutation, x(t+1) is the position after mutation, and r is a random value in the range [0,1].

[0137] b. Introduce an adaptive weighting strategy, the specific mathematical description of which is:

[0138]

[0139] x(t+1)=wx * (t)-AD

[0140] x(t+1)=D'e bl cos(2πl)+wx * (t)

[0141] Where w is the adaptive weight and Max_iter is the optimization algebra.

[0142] c. Introduce a random difference mutation strategy, the specific mathematical description of which is as follows:

[0143] x(t+1)=r1(x * (t)-x(t))+r2(x'(t)-x(t))

[0144] Where x'(t) is a randomly selected individual from the population, and r1 and r2 are random values ​​in the range [0,1].

[0145] S3.2. An improved whale optimization algorithm is used to optimize the penalty factor γ and kernel function parameter σ in LSSVM.

[0146] The solution steps for LSSVM in step S3.2 are as follows:

[0147] 1) Set up a wind power training sample set, perform spatial mapping on the samples in the set, and establish the optimal decision function:

[0148] y(x)=ωφ(x)+b

[0149] Where ω is the weight and b is the bias. For spatial mapping;

[0150] 2) Based on the minimum theory of structural risk mitigation, the optimized optimal decision function is obtained:

[0151]

[0152] y i =φ(x i )ω+b+ξ i i = 1, ..., l

[0153] Where, ξ i Let c be a slack variable, c be a penalty factor, and l be the length of the training samples.

[0154] 3) Use the Lagrange multiplier method to solve for the optimal decision function optimized in step S3.2.2, and obtain the final linear regression equation:

[0155]

[0156] Where K(x) i ,x j ) is the kernel function.

[0157] S3.3 Establish the IWOA-LSSVM model to predict the wind power subsequences after the second decomposition.

[0158] In this embodiment of the invention, the improved whale optimization algorithm is used to optimize the LSSVM model parameters. The concept of establishing an IWOA-LSSVM prediction model to predict all components obtained from the quadratic decomposition is as follows: the training sets of all components obtained from the quadratic decomposition are sequentially input into the LSSVM model for training. During training of each component, the improved whale optimization algorithm is used to optimize the penalty factor and kernel function parameters in the LSSVM model to obtain the optimal prediction model. Then, the test sets of each component are input into the trained IWOA-LSSVM model for prediction to obtain the prediction results for all components. The convergence curve comparison between the improved whale optimization algorithm (IWOA) for optimizing the penalty factor γ and kernel function parameter σ in the LSSVM model and other optimization algorithms is shown in the figure below. Figure 6 As shown. The Whale Optimization Algorithm (WOA) mainly consists of three stages: surrounding the prey, using a bubble net to catch prey, and searching for prey. The specific steps are as follows:

[0159] (1) Surrounding the prey. The whale optimization algorithm considers the nearest individual as prey, which is the optimal solution. Then, other whales in the group will gradually move closer to the current optimal solution to surround the prey, while updating their positions. The update formula is as follows:

[0160] D = |Cx * (t)-x(t)|

[0161] x(t+1)=x * (t)-AD

[0162] Where t is the current iteration number; x * (t) represents the position of the current optimal solution; D represents the distance between the current search individual and the optimal solution; x(t) represents the position of the current search individual; A and C are coefficient vectors.

[0163] A = 2ar - a

[0164] C = 2r

[0165] Where a is the convergence factor, which decreases linearly from 2 to 0; r is a random value in the range [0,1].

[0166] (2) Bubble Net Hunting. Whales employ two hunting mechanisms: shrinking circling hunting and spiral bubble net hunting. The position update when using the shrinking circling hunting mechanism is shown in the following formula:

[0167] x(t+1)=x * (t)-AD

[0168] When using a spiral bubble net mechanism for predation, the position update is as follows:

[0169] D'=|x * (t)-x(t)|

[0170] x(t+1)=D'e bl cos(2πl)+x * (t)

[0171] Where D' is the distance between the current search individual and the current optimal solution; b is the spiral parameter; and l is a random number with a value range of [-1, 1].

[0172] Since humpback whales have two methods for approaching prey to hunt, the WOA (Wide Position Update) determines whether to use the shrinking circling mechanism or the spiral bubble net mechanism based on the probability p. The position update formula is shown below:

[0173]

[0174] In the formula: p is the probability of predation mechanism, and the value range is a random number between [0,1].

[0175] As the number of iterations t increases, the parameter A and the convergence factor a gradually decrease. If |A| < 1, then each whale gradually surrounds the current optimal solution, which belongs to the local optimization stage in WOA.

[0176] (3) Searching for prey. When the absolute value of A is less than 1, the prey location is the optimal individual location, and other whales in the pod will continuously move towards this location. When the absolute value of A is greater than 1, the current optimal individual will be reselected until the optimal solution is obtained. The process is as follows:

[0177] D”=|Cx rand (t)-x(t)|

[0178] x(t+1)=x rand (t)-AD”

[0179] Where D” represents the distance between the current searched individual and the random individual; x rand (t) represents the current position of the random individual.

[0180] According to a preferred embodiment of the present invention, in step S4, all component prediction results are superimposed to obtain the final wind power prediction result, and error verification of the prediction result specifically includes:

[0181] S4.1. Superimpose all wind power subsequence prediction results to obtain the final wind power prediction result;

[0182] S4.2 Calculate the values ​​of RMSE, MAE and MAPE, and verify the error of the prediction results.

[0183] The expressions for RMSE, MAE, and MAPE are as follows:

[0184]

[0185] In the embodiments of the present invention, see Figure 7-8 To better demonstrate the superiority of the model proposed in this invention, the prediction results of the four prediction models are verified by error verification. The RMSE, MAE, and MAPE of the predicted values ​​of each prediction model are calculated. RMSE is the root mean square error, MAE is the mean absolute error, and MAPE is the mean relative percentage error. The final results are shown in Table 1 below.

[0186] Table 1: Evaluation Indicators of Model Prediction Error

[0187]

[0188] As shown in the table above, the wind power prediction model of this invention has the lowest prediction error. Specifically, directly using the original wind power sequence for LSSVM modeling yields the worst prediction performance, with a prediction error of 15.24%. The IWOA-LSSVM model with optimized model parameters has a prediction error of 14.37%. The CEEMD-IWOA-LSSVM model, constructed using CEEMD primary decomposition, has a prediction error of 8.96%. The wind power prediction model based on CEEMD-VMD-IWOA-LSSVM, which uses CEEMD primary decomposition, sample entropy analysis, and VMD secondary decomposition for the still relatively complex subsequences, has a prediction error of 4.41%.

[0189] The present invention provides an ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM. This method effectively solves the problems of high complexity of the original wind power sequence, slow convergence speed of the optimization algorithm, and low prediction accuracy of the model. It not only helps grid dispatchers to formulate reasonable power generation plans, but also provides strategies for wind farm scheduling and energy storage, reducing the power generation cost of wind farms, and has good application prospects.

[0190] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings; however, the present invention is not limited thereto. Within the scope of the inventive concept, various simple modifications can be made to the technical solutions of the present invention. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately. However, these simple modifications and combinations should also be considered as the content disclosed in the present invention and are all within the protection scope of the present invention.

Claims

1. A method for ultra-short-term wind power prediction based on quadratic decomposition and IWOA-LSSVM, characterized in that, The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM includes: S1. Use CEEMD to decompose the original wind power sequence once, and calculate the sample entropy of all components after the first decomposition. S2. Perform secondary decomposition of the subsequences with high sample entropy using VMD. S3. The improved whale optimization algorithm is used to optimize the LSSVM model parameters, and the IWOA-LSSVM prediction model is established to predict all components obtained by the second decomposition. S4. Superimpose all component prediction results to obtain the final wind power prediction result, and verify the error of the prediction result. In step S3, the improved whale optimization algorithm is used to optimize the LSSVM model parameters, and the IWOA-LSSVM prediction model is established to predict all components obtained from the quadratic decomposition. Specifically, this includes: S3.1 Improve the whale optimization algorithm; S3.

2. An improved whale optimization algorithm is used to adjust the penalty factor in LSSVM. and kernel function parameters Optimize; S3.3 Establish the IWOA-LSSVM model to predict the wind power subsequences after the second decomposition. The steps for solving LSSVM in step S3.2 are as follows: 1) Set up a wind power training sample set, perform spatial mapping on the samples in the set, and establish the optimal decision function: in, As weight, For deviation, For spatial mapping; 2) Based on the minimum theory of structural risk mitigation, the optimized optimal decision function is obtained: in, As slack variables, As a penalty factor, The length of the training samples; 3) Use the Lagrange multiplier method to solve for the optimal decision function optimized in step S3.2.2, and obtain the final linear regression equation: in, This is the kernel function.

2. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 1, characterized in that, In step S1, the original wind power sequence is decomposed using CEEMD, and the sample entropy of all components after the first decomposition is calculated, specifically including: S1.1, in the original wind power sequence Positive and negative noise signals are added to the middle respectively. and This generates a new set of signals; S1.2 Perform EMD decomposition on the generated new set of signals; S1.

3. Take the average value of the several IMF components obtained from the decomposition; S1.4 Calculate the sample entropy of all components.

3. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 2, characterized in that, In step S1.2, the EMD decomposition of the generated new set of signals specifically includes: S1.2.1 Obtain the upper and lower envelopes of the sequence using cubic spline interpolation. and ; in, ; S1.2.2 Calculate the mean of the upper and lower envelopes. : S1.2.3 Extracting local detail information of the signal : S1.2.4, Judgment If conditions a and b are satisfied simultaneously, proceed to step S1.2.5; otherwise, return to step S1.2.

1. Condition a is that the number of points is equal to or differs from the number of zero-crossing points in the sequence by one, and condition b is that the average value of the upper and lower envelopes is zero at any given time. S1.2.5 Calculate the remaining data : S1.2.6 Repeat steps S1.2.1-S1.2.5 to obtain N IMF components until equation (1) is satisfied and the decomposition stops; Equation (1) in, For the number of filters, The number of samples; S1.2.7 The final EMD decomposition result is as follows: Where N is the total number of IMFs that are decomposed.

4. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 1, characterized in that, In step S2, the secondary decomposition of the subsequences with high sample entropy using VMD specifically includes: S2.1 Constructing the variational optimization function: in, To decompose the components, Indicates the center frequency, * represents convolution. Let be the impulse function. To determine the number of decompositions, To Find the partial derivative; S2.2 Solving the variational optimization function using the Lagrange method: in, As a penalty factor, , For operators, Indicates the inner product; S2.3 Combining the alternating direction multiplier algorithm and Fourier transform, the final iterative expression is obtained: Among them, step size .

5. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 1, characterized in that, In step S3.1, the improvements to the whale optimization algorithm specifically include: a. Introducing the Cauchy mutation strategy, the specific mathematical description formula is as follows: in, This represents the position of the individual whale before its mutation. This is the position after the mutation. It is a random value with a range of [0,1]. b. Introduce an adaptive weighting strategy, the specific mathematical description of which is: in, For adaptive weights, To find the optimal algebra.

6. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 5, characterized in that, In step S3.1, the improvement of the whale optimization algorithm also specifically includes: c. Introduce a random difference mutation strategy, the specific mathematical description of which is as follows: in, It involves randomly selecting individuals from the population. and It is a random value with a range of [0,1].

7. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 1, characterized in that, In step S4, all component prediction results are superimposed to obtain the final wind power prediction result, and the prediction result is verified for error, specifically including: S4.

1. Superimpose all wind power subsequence prediction results to obtain the final wind power prediction result; S4.2 Calculate the values ​​of RMSE, MAE and MAPE, and verify the error of the prediction results.

8. The ultra-short-term wind power prediction method based on quadratic decomposition and IWOA-LSSVM according to claim 7, characterized in that, The expressions for RMSE, MAE, and MAPE are: 。

Images