A wind speed prediction method and system based on entropy clustering

Abstract

The present invention relates to the field of wind speed prediction technology, specifically to a wind speed prediction method and system based on entropy clustering, comprising the following steps: performing multi-scale decomposition of a wind speed sequence to generate subsequences using a permutation entropy algorithm, performing sliding window probability distribution calculation on the multiple subsequences, setting a dynamic classification threshold based on a normalized entropy value, and dividing a subsequence set into an entropy classification result set. The present invention dynamically sets the classification threshold through a sliding window probability distribution, overcomes the sensitivity of fixed entropy value division to changes in data complexity, establishes a nonlinear mapping between the original sequence and the entropy component through GRU training, enhances feature extraction depth, propagates low-component error, and uses Bi-RNN, LSTM, and SVR to model high, medium, and low entropy components, respectively, to match feature extraction of different complexities, optimizes fusion weights using chaotic mapping, balances component complementarity through multi-objective optimization, and adjusts confidence intervals with dynamic error benchmarks and feedback to achieve adaptive tracking of error distribution, thereby enhancing prediction reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of wind speed prediction, and in particular to a wind speed prediction method and system based on entropy clustering. Background Art

[0002] The field of wind speed prediction technology encompasses means and methods for monitoring, evaluating, and predicting wind energy resources. Its core approach is to leverage various historical wind speed data to predict wind speeds for a specific time period in the future through time series analysis, mathematical and statistical modeling, and machine learning. This technology is widely used in engineering scenarios such as wind farm site selection, wind power output scheduling, and system stability analysis. Wind speed prediction technology typically involves data collection, feature generation, model development, and error correction. The key is to improve the accuracy and adaptability of wind speed predictions to meet forecasting needs at different temporal and spatial scales.

[0003] Among them, the entropy clustering-based wind speed prediction method involves dividing wind speed data into multiple sub-groups with different statistical characteristics based on information entropy theory, and then modeling and predicting each sub-group separately. This method addresses the diversity and non-stationarity of wind speed data over different time periods. It first uses entropy as a basis for classification, clustering the overall wind speed dataset into multiple subsets that are relatively consistent internally but significantly different from each other. Each subset is then modeled using a specific prediction model, such as support vector regression or random forest regression, and the outputs of each model are combined to form the final prediction value. This method primarily relies on an information entropy partitioning mechanism and an independent modeling process based on sub-group models.

[0004] The existing technology has three defects. The fixed information entropy threshold leads to inaccurate subclass division when the data distribution changes. For example, the wind speed mode switching caused by the day and night temperature difference will cause traditional clustering to produce abnormal data merging. There is a structural mismatch problem when isomorphic models process heterogeneous subclass data. Support vector regression is prone to losing detailed features when processing high-frequency components, and random forests produce redundant calculations for stationary sequences. The static error benchmark cannot capture real-time error offsets. When the prediction environment changes suddenly, the historical error statistics deviate from the actual distribution, resulting in the failure of the confidence assessment of the interval prediction. These defects make the existing methods face the risk of prediction lag and error accumulation under complex meteorological conditions, affecting the reliability of wind power grid-connected scheduling decisions. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a wind speed prediction method and system based on entropy clustering.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a wind speed prediction method based on entropy clustering, comprising the following steps:

[0007] S1: Perform multi-scale decomposition of the wind speed sequence in the training set using the permutation entropy algorithm to generate a set of subsequences. Perform sliding window probability distribution calculation on the multiple subsequences, set a dynamic classification threshold based on the normalized entropy value, and divide the subsequence set into an entropy classification result set.

[0008] S2: Inputting the original wind speed sequence of the training set and the entropy classification result set into the GRU network to perform gated state parameter training, generating a component mapping model, inputting the original wind speed sequence of the test set into the component mapping model to output the entropy component set of the test set;

[0009] S3: Input the high entropy component in the entropy component set of the test set into the bidirectional gated recurrent unit to generate a high entropy prediction result, input the medium entropy component into the long short-term memory model to generate a medium entropy prediction result, and input the low entropy component into the support vector regression model to generate a low entropy prediction result. Initial weights are assigned based on the entropy value ratio, and the weight parameters are iteratively optimized using the chaotic mapping algorithm. The three types of prediction results are weighted and superimposed to generate a wind speed point prediction result.

[0010] As a further solution of the present invention, the entropy classification result set is specifically a high entropy subsequence category, a medium entropy subsequence category, and a low entropy subsequence category, the test set entropy component set includes a high entropy component, a medium entropy component, and a low entropy component, and the wind speed point prediction result specifically refers to a high entropy prediction result, a medium entropy prediction result, a low entropy prediction result and the optimized weight parameter superposition value.

[0011] As a further solution of the present invention, the normalized entropy value is obtained by linear mapping formula Calculate, where Represents the normalized value of the permutation entropy of the current window, represents the permutation entropy value of the i-th sliding window, Represents the maximum value of the permutation entropy value in all sliding windows, Represents the minimum value of the permutation entropy value in all sliding windows. The dynamic classification threshold is the coordinate value corresponding to the extreme point of the second-order derivative of the entropy distribution density curve after mapping. The extreme point represents the location where the density change is most significant and serves as the classification boundary for category division.

[0012] The dynamic classification threshold is the coordinate value corresponding to the extreme point of the second-order derivative of the entropy distribution density curve after mapping;

[0013] The GRU network includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is equal to the time step of the wind speed sequence. The hidden layer has two layers, each containing 128 neurons. The number of neurons in the output layer is 32.

[0014] The chaotic mapping algorithm adopts the Logistic mapping function Generate a random sequence where is the chaotic sequence value of the nth iteration, is the chaotic sequence value of the n+1th iteration, and the fitness function adopts the mean square error form, that is, ,in, represents the mean square error of the prediction results, is the sample size, is the jth true wind speed value, For the j-th predicted wind speed value, the gradient descent method is used to update the weight perturbation amplitude. When optimizing the weight, the initial weight is distributed according to the entropy ratio. Finally, through multiple iterative optimizations, the converged optimal weight parameters are obtained.

[0015] As a further solution of the present invention, the steps of obtaining the entropy classification result set are specifically as follows:

[0016] S101: Obtain a training set wind speed sequence, call a permutation entropy algorithm to set embedding dimension parameters and delay parameters, construct a multidimensional phase space based on the embedding dimension parameters, perform time delay slicing on the sequence using the delay parameters, reconstruct phase space trajectories at multiple time scales, and decompose the original sequence based on similarity differences between the trajectories to generate multi-scale decomposition subsequences;

[0017] S102: Based on the multi-scale decomposition subsequence, set the coverage length parameter and the moving step parameter of the sliding window, successively intercept the local data segments of the subsequence according to the step parameter, count the frequency of occurrence of differentiated wind speed values in each window, divide the frequency by the window length parameter, calculate the probability distribution of the wind speed values corresponding to the multiple windows, and generate a probability distribution matrix;

[0018] The window length parameter is unified in seconds, and the moving step parameter is 1 second;

[0019] S103: extracting the permutation entropy value of each window from the probability distribution matrix, locating the maximum and minimum values of the entropy value sequence, mapping the entropy value to the interval of 0-1, analyzing the distribution density curve of the mapped entropy value, calculating the second-order derivative of the curvature as the entropy value changes, locating the entropy value coordinates corresponding to the derivative extreme point, and setting the coordinate value as the classification threshold, dividing the high entropy and low entropy categories, and generating a dynamic classification threshold result;

[0020] The second-order derivative is calculated by a numerical difference method, and the extreme point determination criterion is that the sign of the derivative changes from positive to negative or from negative to positive.

[0021] As a further solution of the present invention, the step of obtaining the test set entropy component set is specifically as follows:

[0022] S201: Obtain the original wind speed sequence of the training set and the entropy classification result set, construct a GRU network input layer based on the time step characteristics of the wind speed sequence, and at the same time, normalize and initialize the input layer weights according to the multi-category probability distribution in the entropy classification result set, calculate the weight error of the forget gate and the input gate time step by time step using the gradient descent method, iteratively update the gate control unit parameters through chain derivation, and generate a gate control state parameter set;

[0023] The normalization initialization adopts the Xavier method, and the learning rate of the gradient descent method is set to ;

[0024] S202: calling the gate state parameter set, dividing the training set wind speed sequence into subsequence segments, and establishing a mapping index between the subsequence and the hidden layer state based on the entropy classification label, calculating the error gradient between the hidden layer state and the label through back propagation, updating the output gate weight and the fully connected layer parameters, and generating the component mapping model parameters;

[0025] S203: Segment the test set wind speed sequence into subsequence segments based on the component mapping model parameters, input the GRU network to calculate the hidden layer state vector, extract the neuron activation value corresponding to the entropy classification label, splice the activation value sequence by time step, and generate a test set entropy component set.

[0026] As a further solution of the present invention, the steps for obtaining the wind speed point prediction results are specifically as follows:

[0027] S301: calling the test set entropy component set, inputting the high entropy component into the bidirectional gated recurrent unit based on the entropy classification label, expanding the time step forward and reverse hidden state transfer, calculating the output value of the multi-time step gated unit, inputting the medium entropy component into the long short-term memory model, filtering the time-dependent features according to the forget gate and input gate states, inputting the low entropy component into the support vector regression model, mapping the feature space using the radial basis kernel function and solving the regression hyperplane, and generating a component prediction result set;

[0028] The radial basis kernel function bandwidth parameter ,The hyperplane solution adopts the sequential minimum optimization algorithm;

[0029] S302: Based on the initial entropy value ratios of the high entropy, medium entropy, and low entropy components in the component prediction result set, the proportion of the high entropy component prediction result in the total entropy value is calculated, and the proportion is used as the initial weight distribution coefficient. At the same time, the Logistic mapping function in the chaotic mapping algorithm is called to generate a random sequence. The mean square error of the prediction result is used as the fitness function, and the perturbation amplitude of the weight parameter is iteratively updated to generate the optimized weight parameter.

[0030] The disturbance amplitude update step size is ;

[0031] S303: Call the component prediction result set and the optimized weight parameter, perform a scalar product of the high entropy prediction result and the high entropy weight coefficient in the optimized weight, synchronously calculate the product of the medium entropy and low entropy prediction results and the corresponding weights, align and superimpose the three types of weighted results according to the time step, and generate a wind speed point prediction result.

[0032] As a further embodiment of the present invention, the method further comprises:

[0033] S4: Establish a dynamic error benchmark set based on the error standard deviation of multiple entropy classification intervals in the training set, input the wind speed point prediction results and the real-time error of the test set into the error distribution model, adjust the confidence interval width according to the relative relationship between the error and the benchmark value, and output the interval prediction result set.

[0034] As a further solution of the present invention, the interval prediction result set includes a dynamic error benchmark, real-time error distribution data, a confidence interval width adjustment parameter and a probability interval range;

[0035] The dynamic error benchmark set is an error standard deviation sequence arranged in ascending order by entropy classification labels, and the confidence interval width adjustment formula is: ,in, is the width of the current confidence interval, is the proportionality coefficient, is the wind speed prediction error value at time t, is the error benchmark value of the kth entropy classification interval, and The difference between them reflects the uncertainty of the prediction. The confidence interval width is adjusted dynamically based on the difference, and the probability interval range is determined by the error distribution and the width parameter.

[0036] As a further solution of the present invention, the steps of obtaining the interval prediction result set are specifically as follows:

[0037] S401: Obtain prediction error sequences corresponding to multiple entropy classification intervals of the training set, divide the error data by interval, calculate the standard deviation of the error values in each entropy classification interval, and sort the standard deviations in ascending order according to the entropy classification labels to generate a dynamic error benchmark set;

[0038] S402: calling the dynamic error benchmark set, aligning the wind speed point prediction result with the test set real-time error by time step, and calculating the absolute value of the difference between the error value of each time step and the corresponding entropy classification benchmark value based on the normal distribution setting of the error distribution model, and updating the mean and variance parameters of the distribution model using the maximum likelihood estimation method to generate error distribution parameters;

[0039] The initial mean of the normal distribution is , the variance is ;

[0040] S403: Based on the error distribution parameters, according to the proportion of the error value of the current time step exceeding the benchmark value, the confidence interval width is linearly expanded or reduced according to the proportional coefficient, the mean of the distribution model is used as the center of the interval, and the upper and lower bounds are calculated in combination with the adjusted width to generate the interval prediction result.

[0041] A wind speed prediction system based on entropy clustering, wherein the wind speed prediction system based on entropy clustering is used to execute the above-mentioned wind speed prediction method based on entropy clustering, and the system comprises:

[0042] The decomposition and clustering module is used to decompose the wind speed sequence of the training set into a subsequence set through the permutation entropy algorithm, divide the subsequences into subsequences using a sliding window, calculate the probability distribution of wind speed values within the window, generate a dynamic classification threshold based on the normalized entropy function, divide the subsequences into a high entropy component set, a medium entropy component set, and a low entropy component set, integrate them into an entropy classification result set, and pass the entropy classification result set to the component mapping module;

[0043] The middle entropy component set is a set of subsequences whose normalized entropy values are in the middle 30% interval of the dynamic classification threshold;

[0044] A component mapping module is used to obtain the original wind speed sequence of the training set and the entropy classification result set, generate a component mapping model through GRU network training, input the wind speed sequence of the test set into the model to output the entropy component set of the test set, and pass the entropy component set of the test set to the weight prediction module;

[0045] A weight prediction module is used to call the test set entropy component set, input the high entropy component set into a bidirectional gated recurrent unit to generate a high entropy prediction value, input the medium entropy component set into a long short-term memory model to generate a medium entropy prediction value, input the low entropy component set into a support vector regression model to generate a low entropy prediction value, assign initial weights based on the entropy value ratio, optimize the weight parameters using a chaotic mapping algorithm, superimpose and generate wind speed point prediction results, and pass the wind speed point prediction results to an interval correction module;

[0046] The interval correction module is used to establish a dynamic error benchmark set based on the prediction error standard deviation of the entropy classification interval of the training set, call the wind speed point prediction results and the real-time error of the test set, calculate the relative difference between the real-time error and the benchmark value through the error distribution model, adjust the confidence interval width, and output the interval prediction result set.

[0047] Compared with the prior art, the advantages and positive effects of the present invention are:

[0048] In the present invention, a dynamic classification threshold is constructed by combining the sliding window probability distribution calculation to overcome the sensitivity of the fixed entropy value division to the change of data complexity, so that the subsequence division is more in line with the actual fluctuation pattern of the wind speed. The GRU network gate parameter training establishes a nonlinear mapping relationship between the original sequence and the entropy component, enhances the feature extraction depth, and reduces the component classification error transmission. High, medium and low entropy components are targetedly modeled using bidirectional recurrent units, long short-term memory and support vector regression respectively to match the time series feature extraction requirements of data of different complexity and avoid the structural limitations of a single model. The chaotic mapping algorithm optimizes the weight parameters, balances the complementarity of the prediction results of each component through multi-objective optimization, and improves the stability of the fusion prediction. The dynamic error benchmark set is combined with real-time error feedback to adjust the confidence interval, realizes adaptive tracking of the prediction error distribution, and enhances the timeliness and reliability of the interval prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0050] Figure 2 Flowchart of the steps for obtaining the entropy classification result set of the present invention;

[0051] Figure 3 Flowchart of the steps for obtaining the entropy component set of the test set of the present invention;

[0052] Figure 4 Flowchart of the steps for obtaining the wind speed point prediction results of the present invention;

[0053] Figure 5 This is a flow chart of the steps for obtaining the interval prediction result set of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0056] Example 1: Please refer to Figure 1 The present invention provides a technical solution: a wind speed prediction method based on entropy clustering, comprising the following steps:

[0057] S1: Perform multi-scale decomposition of the wind speed sequence in the training set using the permutation entropy algorithm to generate a set of subsequences. Perform sliding window probability distribution calculation on the multiple subsequences, set a dynamic classification threshold based on the normalized entropy value, and divide the subsequence set into an entropy classification result set.

[0058] S2: Input the original wind speed sequence of the training set and the entropy classification result set into the GRU network for gated state parameter training to generate a component mapping model. Input the original wind speed sequence of the test set into the component mapping model to output the entropy component set of the test set.

[0059] S3: Input the high entropy component of the test set entropy component set into the bidirectional gated recurrent unit to generate a high entropy prediction result, input the medium entropy component into the long short-term memory model to generate a medium entropy prediction result, and input the low entropy component into the support vector regression model to generate a low entropy prediction result. Initial weights are assigned based on the entropy value ratio, and the weight parameters are iteratively optimized using the chaotic mapping algorithm. The three types of prediction results are weighted superimposed to generate the wind speed point prediction result.

[0060] S4: A dynamic error benchmark set is established based on the error standard deviation of multiple entropy classification intervals in the training set. The wind speed point prediction results and the real-time error of the test set are input into the error distribution model. The confidence interval width is adjusted according to the relative relationship between the error and the benchmark value, and the interval prediction result set is output.

[0061] The entropy classification result set is specifically divided into high entropy subsequence category, medium entropy subsequence category, and low entropy subsequence category. The test set entropy component set includes high entropy component, medium entropy component, and low entropy component. The wind speed point prediction result specifically refers to the high entropy prediction result, medium entropy prediction result, low entropy prediction result and the optimized weight parameter superposition value. The interval prediction result set includes dynamic error benchmark, real-time error distribution data, confidence interval width adjustment parameter and probability interval range.

[0062] The normalized entropy value is obtained by linear mapping formula Calculate, where Represents the normalized value of the permutation entropy of the current window, represents the permutation entropy value of the i-th sliding window, Represents the maximum value of the permutation entropy value in all sliding windows, Represents the minimum value of the permutation entropy value in all sliding windows. The dynamic classification threshold is the coordinate value corresponding to the extreme point of the second-order derivative of the entropy distribution density curve after mapping. The extreme point represents the location where the density change is most significant and serves as the classification boundary for category division.

[0063] The dynamic classification threshold is the coordinate value corresponding to the extreme point of the second-order derivative of the entropy distribution density curve after mapping;

[0064] The GRU network consists of an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is equal to the time step of the wind speed sequence. The hidden layer has two layers and each layer contains 128 neurons. The number of neurons in the output layer is 32.

[0065] The chaotic mapping algorithm uses the Logistic mapping function Generate a random sequence where is the chaotic sequence value of the nth iteration, is the chaotic sequence value of the n+1th iteration, and the fitness function adopts the mean square error form, that is, ,in, represents the mean square error of the prediction results, is the sample size, is the jth true wind speed value, For the j-th predicted wind speed value, the gradient descent method is used to update the weight perturbation amplitude. When optimizing the weight, the initial weight is distributed according to the entropy ratio. Finally, through multiple iterative optimizations, the converged optimal weight parameters are obtained.

[0066] The dynamic error benchmark set is the error standard deviation sequence arranged in ascending order by entropy classification labels, and the confidence interval width adjustment formula is: ,in, is the width of the current confidence interval, is the proportionality coefficient, is the wind speed prediction error value at time t, is the error benchmark value of the kth entropy classification interval, and The difference between them reflects the uncertainty of the prediction. The confidence interval width is adjusted dynamically based on the difference, and the probability interval range is determined by the error distribution and the width parameter.

[0067] See also Figure 2 , the specific steps for obtaining the entropy classification result set are:

[0068] S101: Obtain the wind speed sequence of the training set, call the permutation entropy algorithm to set the embedding dimension parameters and delay parameters, construct a multidimensional phase space based on the embedding dimension parameters, slice the sequence by time delay using the delay parameters, reconstruct the phase space trajectories of multiple time scales, and decompose the original sequence based on the similarity differences between the trajectories to generate multi-scale decomposition subsequences;

[0069] The training set wind speed sequence is a set of wind speed observations arranged in chronological order. The wind speed data used here comes from 10 minutes of data collected continuously from wind turbine No. 1 of a wind farm during a specific period. The sampling frequency is 1 Hz, that is, one wind speed value is recorded per second, totaling 600 data points. The numerical unit is uniformly meters per second (m / s). Some examples are shown in Table 1.

[0070] Table 1: Wind speed sequence of training set

[0071] ;

[0072] As shown in Table 1, the specific values of the first 10 sampling points of the wind speed sequence in the training set are displayed.

[0073] When calling the permutation entropy algorithm, first set the embedding dimension parameter , its value refers to the sequence length and the need to capture details, here it is set , which can capture the main fluctuation mode of wind speed while avoiding overfitting noise, and then set the delay parameter , its value should match the characteristic time scale of the sequence. For wind speed data with a sampling frequency of 1 Hz, set seconds, that is, using adjacent continuous data points for reconstruction.

[0074] According to the embedding dimension parameter Construct a three-dimensional phase space. The specific execution process is: for the wind speed sequence , from which a vector sequence is constructed ,in The range is from 1 to ,because Seconds and , so the vector is , taking the data in Table 1 as an example, the first vector is , and the second vector is , and so on, until the end of the sequence, by delay parameter The time delay slicing of the sequence by seconds has been reflected in the above vector construction, that is, the elements of each vector are the data points in the original sequence delayed by 1 second. In this way, the phase space trajectories of multiple time scales are reconstructed, which actually obtains Three-dimensional vectors, the sequence of points formed by these vectors in three-dimensional space is the phase space trajectory.

[0075] Based on the similarity difference between the order of the internal element values of each vector in these phase space trajectories (i.e., the arrangement pattern), the original sequence is decomposed. This step uses the variational mode decomposition (VMD) method to convert the original wind speed sequence into Decompose into modal components and a residual component, setting the number of decomposition layers , these modal components It represents fluctuations with different frequency and amplitude characteristics, which is a multi-scale decomposition subsequence.

[0076] S102: Based on the multi-scale decomposition subsequence, set the coverage length parameter and the moving step parameter of the sliding window, successively intercept the local data segments of the subsequence according to the step parameter, count the frequency of occurrence of differentiated wind speed values in each window, divide the frequency by the window length parameter, calculate the probability distribution of the wind speed values corresponding to the multiple windows, and generate a probability distribution matrix;

[0077] The window length parameter is unified in seconds, and the moving step parameter is 1 second;

[0078] Based on one of the multi-scale decomposition subsequences obtained in step S101, select the first modal component After processing, the subsequence is numerically represented as a set of fluctuation data, and its unit is still meter per second (m / s).

[0079] Set the coverage length parameter of the sliding window , the physical duration covered is uniformly 10 seconds, because The sampling frequency is consistent with the original sequence, which is 1Hz, so Corresponding to 10 consecutive data points, set the moving step parameter , its physical duration is 1 second, corresponding to moving 1 data point, according to the step parameter Successive subsequence truncation The local data segment of the first window is , the data captured by the second window is , and so on until the end of the subsequence.

[0080] Count the frequency of occurrence of differentiated wind speed values in each window, and examine the data segment of the first window. The specific value is (These values are ), where the differentiated wind speed values are -0.2m / s, 0.1m / s, 0.3m / s, and 0.4m / s. Their frequencies are: -0.2 appears 4 times, 0.1 appears 3 times, 0.3 appears 2 times, and 0.4 appears 1 time. These frequencies are combined with the window length parameter (number of data points) is divided to calculate the probability of each wind speed value in the window. Specifically, the probability of the value -0.2 is , the probability of value 0.1 is , the probability of value 0.3 is , the probability of value 0.4 is , for all This statistics and calculation is performed on each window to generate the probability distribution of wind speed values corresponding to each window. These distributions together constitute a probability distribution matrix, where rows represent windows, columns represent differentiated wind speed values appearing in the window, and matrix elements are corresponding probabilities.

[0081] S103: Extract the permutation entropy value of each window from the probability distribution matrix, locate the maximum and minimum values of the entropy value sequence, map the entropy value to the range of 0-1, analyze the distribution density curve of the mapped entropy value, calculate the second-order derivative of the curvature as the entropy value changes, locate the entropy value coordinates corresponding to the derivative extreme point, set the coordinate value as the classification threshold, divide the high entropy and low entropy categories, and generate a dynamic classification threshold result;

[0082] The second-order derivative is calculated by numerical difference method, and the extreme point is determined by the sign of the derivative changing from positive to negative or from negative to positive.

[0083] Extract the corresponding permutation entropy value from the local data segment intercepted from each window determined in step S102. The specific calculation is: take the subsequence data segment within the window, such as the length of Data segment , using the embedding dimension set in S101 and delay parameters Seconds, the phase space of this 10-point data segment is reconstructed to produce Three-dimensional vectors: , determine the arrangement pattern of its elements for each three-dimensional vector, there are Possible arrangements (such as , etc.), count the number of times each arrangement pattern appears in these 8 vectors ( ), calculate its frequency , then the permutation entropy of the window is Calculated as (when hour, ), repeat this process for all windows to get a permutation entropy value sequence ,in is the total number of windows.

[0084] Take the maximum value of this entropy sequence With minimum value Positioning, assuming that the minimum permutation entropy value of the entire sequence is 0.15 and the maximum permutation entropy value is 0.95, then , , each entropy value By linear normalization formula Mapped to the 0-1 interval, an original entropy value is The point whose normalized value is , analyze the normalized entropy values after these mappings The distribution density curve of , and the kernel density estimation method is used to obtain a smooth probability density function , calculate this density function about The second derivative of , calculated by the central difference method: ,in is a small step size selected, .

[0085] Positioning the second derivative The entropy coordinates corresponding to the extreme points, the criterion for determining the extreme points is the second-order derivative The sign of changes and its first-order derivative (i.e. ) is zero or close to zero (in actual operation, it is often simplified to find The zero point of the sign change, or directly find local maxima and minima of the original density function. Set the inflection points corresponding to these inflection points The value is the classification threshold. The specific setting of the threshold refers to the position of the identified inflection point. If after calculation and analysis, the first significant inflection point is identified at the normalized entropy value of 0.33 (the second-order derivative turns from positive to negative, corresponding to the point where the density function slope increases fastest and then begins to slow down), it is set as the first classification threshold. , the second significant inflection point is identified at the normalized entropy value of 0.67 (the second-order derivative turns from negative to positive, corresponding to the point where the density function slope decreases fastest and then begins to increase), and is set as the second classification threshold , then the rule for dividing high entropy and low entropy categories is: normalized entropy value The sequence segment of is defined as low entropy; The sequence segment of is defined as medium entropy; The sequence segment is defined as high entropy, thereby generating a dynamic classification threshold result. The "dynamic" is reflected in the fact that the threshold is automatically determined according to the entropy distribution characteristics of the current training data.

[0086] See also Figure 3 , the specific steps for obtaining the test set entropy component set are:

[0087] S201: Obtain the original wind speed sequence and entropy classification result set of the training set, construct a GRU network input layer based on the time step characteristics of the wind speed sequence, and at the same time, normalize and initialize the input layer weights according to the multi-category probability distribution in the entropy classification result set. Use the gradient descent method to calculate the weight error of the forget gate and the input gate time step by time step, iteratively update the gate control unit parameters through chain derivation, and generate a gate control state parameter set;

[0088] Normalization initialization uses the Xavier method, and the learning rate of the gradient descent method is set to ;

[0089] Obtain the original wind speed sequence of the training set (the complete 600 data points shown in Table 1) and the entropy classification result set generated in S103, which is the entropy category (low, medium, high) corresponding to each 10-second data segment (window) in the training set, based on the time step characteristics of the wind speed sequence (i.e., each time point Wind speed value ) Build the input layer of the Gated Recurrent Unit (GRU) network with an input dimension of 1 (only wind speed values).

[0090] At the same time, according to the entropy classification result set obtained in step S103, the proportion of each category (low, medium, and high entropy) in the entire training set is counted, and the proportion of low entropy data segments is set to , the proportion of medium entropy data segment is , the proportion of high entropy data segments is ,in are the number of low, medium and high entropy data segments, respectively. The weights from the input layer to the hidden layer of the GRU network and the weights of the recurrent connections are initialized using the Xavier normalization method. This method adapts the initial value of the weight to the size of the layer and helps to stabilize the propagation of the gradient. The specific weights are uniformly distributed. Extract from is the number of input neurons, is the number of output neurons, the category ratio calculated above It is used for sample weighting or loss function adjustment in subsequent model training steps, rather than directly for the Xavier initialization process here.

[0091] Use gradient descent method to optimize GRU network parameters and set learning rate , this is an empirical value, usually arrive The specific task is to use the GRU network to make some predictions or characterization learning of the wind speed sequence (such as predicting the wind speed at the next moment), and calculate the GRU internal reset gate by time step by comparing the difference (loss) between the network output and the true target value. and update gate The parameters (weight matrix and the bias vector ) and the parameters of the candidate hidden states (weight matrix and the bias vector ) relative to the gradient of the loss function, the parameters of these gating units are iteratively updated through the backpropagation through time (BPTT) algorithm and the chain derivation rule to generate a set of gating state parameters, namely the trained GRU network weights and biases.

[0092] S202: Calling the gated state parameter set, dividing the training set wind speed sequence into subsequence segments, and establishing a mapping index between the subsequence and the hidden layer state based on the entropy classification label. Calculating the error gradient between the hidden layer state and the label through back propagation, updating the output gate weight and the fully connected layer parameters, and generating the component mapping model parameters.

[0093] The gated state parameter set trained in S201 is called, i.e. the optimized GRU network. The original wind speed sequence of the training set is converted into a window of length defined in S102. seconds (10 data points) and step size The data is segmented into 100 seconds (1 data point) to obtain subsequence segments consistent with the entropy classification in S103.

[0094] Based on the entropy classification label (low, medium, high) of each training subsequence segment obtained in S103, a mapping index is established between the subsequence segment and the hidden layer state obtained after processing by the current GRU network. The specific operation is: input each subsequence segment into the GRU network, extract the hidden state vector of the last time step of the subsequence segment , this vector It is used as the feature representation of the subsequence segment and establishes a correspondence with its known entropy label (such as "low entropy").

[0095] On this basis, a classification layer is constructed, usually a fully connected layer, whose input is the hidden state of the GRU output , the output dimension is 3 (corresponding to the three entropy categories of low, medium, and high), and is followed by a Softmax activation function to convert the output into the probability of each category. The error gradient between the predicted entropy label output by the classification layer and the actual entropy label is calculated through the back propagation algorithm, and the loss function adopts the cross entropy loss. This error gradient is not only used to update the weights of the fully connected classification layer and bias , and further backpropagation is performed to fine-tune the parameters of the GRU network (especially the part related to the output, or the parameters of the entire GRU network are jointly tuned at this stage) to generate the final component mapping model parameters, which can receive a wind speed subsequence segment and output its corresponding entropy classification.

[0096] S203: Split the test set wind speed sequence into subsequence segments based on the component mapping model parameters, input the GRU network to calculate the hidden layer state vector, extract the neuron activation value corresponding to the entropy classification label, splice the activation value sequence by time step, and generate the test set entropy component set.

[0097] Based on the component mapping model parameters (including GRU parameters and subsequent classification layer parameters) obtained through training in S202, the test set wind speed sequence is used. This test set is a new wind speed data that has not been used in training. The same window length as that used in training is used. Seconds and Step Size The test sequence is divided into test subsequence segments.

[0098] Each test subsequence segment is input into the GRU network part of the component mapping model trained in S202, and its learned parameters are used to calculate the hidden layer state vector of the subsequence segment, usually taking the hidden state of the last time step. As a representative. Then, this hidden state The input is sent to the subsequent fully connected classification layer in the model, which outputs the activation values of the three neurons. After the Softmax function, the probability distribution corresponding to the three entropy categories of low, medium, and high is obtained. ,in is the index of the test subsequence segment. For example, for a test subsequence segment, its output probability may be , indicating that the probability of this segment being judged as high entropy is 0.60.

[0099] These probability vectors are concatenated by time step (or subsequence order), or the entropy classification label for each subsequence is determined based on the maximum probability principle (e.g., "high entropy" in the above example). If the probability vector concatenation method is used, the test set entropy component set is a sequence of multiple three-dimensional probability vectors; if the label determination method is used, it is an entropy label sequence, generating a test set entropy component set for use in subsequent steps.

[0100] See also Figure 4 , the specific steps for obtaining the wind speed point forecast results are:

[0101] S301: Calling the test set entropy component set, inputting the high entropy component into the bidirectional gated recurrent unit based on the entropy classification label, expanding the time step forward and reverse hidden state transfer, calculating the output value of the multi-time step gated unit, inputting the medium entropy component into the long short-term memory model, filtering the time-dependent features according to the forget gate and input gate states, inputting the low entropy component into the support vector regression model, mapping the feature space using the radial basis kernel function and solving the regression hyperplane, and generating a component prediction result set;

[0102] Radial basis kernel function bandwidth parameter ,The hyperplane solution adopts the sequential minimum optimization algorithm;

[0103] The test set entropy component set obtained in step S203 is called, and the set is the predicted entropy classification label (low, medium, high) of each 10-second subsequence segment in the test set.

[0104] For the wind speed subsequence segments in the test set that are judged as high entropy by the component mapping model, their original wind speed data (these 10 data points) are input into a bidirectional gated recurrent unit (BiGRU) network specially trained for high entropy sequence prediction. The BiGRU network processes the sequence through two GRU layers, forward and reverse, and at each time step , the forward GRU layer outputs the hidden state , the reverse GRU layer outputs the hidden state , these two states are usually concatenated or added ( or ), an output layer (such as a fully connected layer) is used to calculate the predicted value of the time step or the overall prediction result of the sequence, thereby generating a prediction result sequence of high entropy components High entropy sequences usually fluctuate violently, and BiGRU can better capture their bidirectional dependency features.

[0105] For wind speed subsequences that are judged to have medium entropy, the original wind speed data is input into a long short-term memory model (LSTM) network. LSTM effectively filters and memorizes the long-range time-dependent features in the sequence through its internal input gate, forget gate, and output gate structure. It is suitable for medium-complexity wind speed changes with a certain degree of continuity and trend, and generates a prediction result sequence of medium entropy components. .

[0106] For the wind speed subsequence segments that are judged to be low entropy, the original wind speed data (or the statistical features extracted from them, such as mean, variance, etc.) are input into a support vector regression (SVR) model. This SVR model uses the radial basis kernel function (RBFKernel), and its expression is: ,in The bandwidth parameter of the kernel function affects the complexity and smoothness of the model. It is determined by cross-validation and other methods. , It is a sample and The square of the Euclidean distance between them. SVR implicitly maps the input features to a high-dimensional feature space through this kernel function, and searches for a regression hyperplane in the space so that the deviation of most sample points from the hyperplane is within a certain tolerance. The hyperplane is solved using the Sequential Minimization Optimization (SMO) algorithm, which is an efficient method for solving quadratic programming problems and generates a sequence of prediction results with low entropy components. Low entropy sequences are usually relatively stable, and SVR can better fit their trends.

[0107] Finally, the prediction values of the three models for the test subsequence segments of their respective entropy categories are aggregated to form a component prediction result set containing the prediction results of all test points.

[0108] S302: Based on the initial entropy value ratios of the high entropy, medium entropy, and low entropy components in the component prediction result set, the proportion of the high entropy component prediction result in the total entropy value is calculated, and the proportion is used as the initial weight distribution coefficient. At the same time, the Logistic mapping function in the chaos mapping algorithm is called to generate a random sequence. The mean square error of the prediction result is used as the fitness function, and the perturbation amplitude of the weight parameter is iteratively updated to generate the optimized weight parameter.

[0109] The perturbation amplitude update step is ;

[0110] Based on the component prediction result set obtained in step S301 (i.e., the sequence , , ), as well as the initial proportions of different entropy components in the overall wind speed sequence obtained during the training phase or through analysis of historical data. These proportions can be used as initial weights. Suppose the high entropy component accounts for 0.20, the medium entropy component accounts for 0.45, and the low entropy component accounts for 0.35 (consistent with the training set statistics in S201), then the initial weights are set to , the sum of these weights is 1.

[0111] Call the Logistic mapping function in the chaos mapping algorithm, and its iterative formula is , set the control parameters (This value makes the mapping chaotic), the initial value Randomly select from the interval (0,1), excluding 0 and 1, , by iterating this function to generate a pseudo-random chaotic sequence .

[0112] The mean square error (MSE) between the combined prediction result and the true wind speed value is used as the fitness function. The MSE is calculated as follows: ,in is the number of prediction points involved in the evaluation, It is The weight parameters are updated by iterative optimization algorithms (such as particle swarm optimization or genetic algorithm, which is simplified to a search based on chaotic perturbation) , and constrain And each item is non-negative. In each iteration, the chaotic value generated by the Logistic map is used to slightly perturb the current weight, and the update step size of the perturbation amplitude is set to The specific perturbation method can be: starting from the current weight, using the chaotic sequence to generate a perturbation direction and size, generating a new candidate weight set, if the MSE generated by the new weight set is lower than the current optimal MSE, then accept this new set of weights. The iterative process continues until the MSE converges or reaches the preset maximum number of iterations (such as 200 times), generating the optimized weight parameters , a set of example optimization weights is .

[0113] S303: Call the component prediction result set and optimize the weight parameters, perform a scalar product of the high entropy prediction result and the high entropy weight coefficient in the optimized weight, synchronously calculate the product of the medium entropy and low entropy prediction results and the corresponding weights, align and superimpose the three types of weighted results according to the time step, and generate the wind speed point prediction result.

[0114] Call the component prediction result set generated in step S301, that is, the prediction sequence of the high entropy part , the predicted sequence of the medium entropy part , and the predicted sequence of the low entropy part , and the optimized weight parameters generated in step S302 .

[0115] The prediction results of the high entropy component And the optimized high entropy weight coefficient Perform scalar product operations to obtain weighted high entropy prediction contributions . Simultaneously, the prediction results of the entropy component And the corresponding entropy weight coefficient Perform scalar multiplication to obtain the weighted mid-entropy prediction contribution Similarly, the prediction results for the low entropy component are And the corresponding low entropy weight coefficient Perform scalar multiplication to obtain the weighted low entropy prediction contribution .

[0116] At each time step , the wind speed data at this time step has been classified as high, medium or low entropy by S203. Therefore, when combining, only the output of the prediction model of the corresponding category at this time step is used. One implementation method is to , if it belongs to high entropy, the prediction of this point is , if it belongs to medium entropy, then , if it is low entropy, then Then, the final wind speed point forecast result It is weighted by adding the category-specific prediction value of the point to the corresponding category weight obtained by overall optimization (here the weight is for the entire component contribution, not point-by-point selection). A more reasonable approach is to make the final prediction Based on this time point Entropy category ,but Predictions taken directly from the model for that category , then the weight of S302 is used to evaluate the contribution of different models to the overall prediction or the integration in some integration strategy, rather than being directly used to generate the final prediction value for each point. Corrected understanding: The weight of S302 is applied to the combination. Assume that each model gives a prediction for all points (or gives a baseline prediction or zero prediction for non-dominant points). Then at each time point , wind speed point prediction results Calculated as: If at a certain point in time is clearly classified into a certain entropy category, such as high entropy, then the and may be 0 or a non-dominant prediction output by its respective model. More practically, at time point , if the point is classified as ,but Pick , and the weight It is used at a higher level (such as when deciding on a model combination strategy or evaluating the overall performance of a model). According to the description of the original text "aligning and superimposing the three types of weighted results by time step", the first explanation is adopted, that is, each model has a (possibly dominant or non-dominant) prediction value for all time points, or more precisely, each time point corresponds to only one entropy category, so the prediction value at that time point only comes from the model corresponding to that entropy category, and multiplying it by the global weight of the model is not very reasonable. The correct understanding should be that for each time point , first determine its entropy category , and then use the corresponding model Prediction The combination of S303 is more likely to be: using the points belonging to the high entropy interval , points belonging to the medium entropy interval use , points belonging to the low entropy range use . And optimize the weight It is used in S302 to evaluate the overall performance of the combination model, or these weights represent the "trust" or "contribution" under different entropy states and are used to adjust the final combination.

[0117] Here we strictly follow the literal meaning of "aligning and superimposing the three types of weighted results by time step": and the high entropy weight coefficient in the optimized weight Perform scalar multiplication and simultaneously calculate the entropy prediction results With corresponding weight The product of and low entropy prediction results With corresponding weight The weighted results of these three types of time series are summed up at each time step. Align and sum them to get the final wind speed point prediction result , generate a sequence of wind speed point forecast results. Here we assume , , For all time points There is a defined prediction sequence, even if The point does not belong to the entropy category dominated by the model, and the model still gives a prediction (possibly of lower quality).

[0118] See also Figure 5 , the steps for obtaining the interval prediction result set are as follows:

[0119] S401: Obtain prediction error sequences corresponding to multiple entropy classification intervals of the training set, divide the error data by interval, calculate the standard deviation of the error values in each entropy classification interval, and sort the standard deviations in ascending order according to the entropy classification labels to generate a dynamic error benchmark set;

[0120] Obtain the training set after the prediction process described in S303 (i.e., first perform entropy classification on the training set data, then apply the corresponding BiGRU, LSTM, SVR model for prediction, and finally use the optimized weights After weighted combination), the final prediction sequence of the training set is obtained . Calculate the corresponding forecast error sequence ,in is the true wind speed value in the training set.

[0121] According to the entropy classification label (low, medium, high) obtained in S103 of the training set data, the prediction error sequence According to the entropy classification interval to which the corresponding time point belongs, three independent error data subsets are formed: , , Among them, the time point The data segment belongs to low entropy, time point The data segment belongs to medium entropy, time point The data segment belongs to high entropy.

[0122] For all error values within each entropy classification interval, calculate its standard deviation. Error standard deviation of low entropy interval ,in is the mean of the error in the low entropy interval, is the number of error points in the low entropy interval. The standard deviation of the error in the medium entropy interval is . Error standard deviation of high entropy interval Calculated: low entropy interval error standard deviation m / s, standard deviation of the median entropy interval error m / s, high entropy interval error standard deviation m / s. These calculated standard deviations are stored in association with their respective entropy classification labels (low, medium, and high) to form a dynamic error benchmark set. This benchmark set reflects the typical level of predictive uncertainty of the model at different complexity (entropy states).

[0123] Table 2: Dynamic error benchmark set

[0124] ;

[0125] As shown in Table 2, the table lists the error standard deviations calculated based on the final prediction error of the training set in different entropy classification intervals. These values will serve as the benchmark for adjusting the prediction width of subsequent intervals.

[0126] S402: Call the dynamic error benchmark set, align the wind speed point prediction results with the test set real-time errors by time step, and calculate the absolute value of the difference between the error value of each time step and the corresponding entropy classification benchmark value based on the normal distribution setting of the error distribution model. Use the maximum likelihood estimation method to update the mean and variance parameters of the distribution model to generate the error distribution parameters;

[0127] The initial mean of the normal distribution is , the variance is ;

[0128] Call the dynamic error benchmark set generated in step S401 (i.e., the standard deviation of each entropy classification interval m / s, m / s, m / s), and the wind speed point prediction results for the test set generated by S303 At the same time, obtain the real-time (or actual) wind speed observation value corresponding to the test set To calculate the real-time prediction error of the test set Align the point prediction results of the test set with the real-time error by time step.

[0129] For the error distribution model, it is set to obey the normal distribution, and a set of distribution parameters is maintained for each entropy category (low, medium, and high). Initially, the error mean of each category can be set m / s, error variance The square of the standard deviation of the corresponding category error obtained by S401 can be used, that is, , , .

[0130] When processing the test set data, for each time step First, the time step is determined by the component mapping model of S203 Entropy category (Low, Medium, or High). Then, collect all The real-time error of the test set (in Belongs to category The maximum likelihood estimation (MLE) method is used to update the category. Mean of the normal distribution model of the error and variance Parameter. Specific calculation: For low entropy categories, the updated mean , the updated variance ,in is the number of low entropy points in the test set. The same calculation is performed for the medium entropy and high entropy categories to generate the updated error distribution parameters for each entropy classification interval. .

[0131] S403: Based on the error distribution parameters, according to the proportion of the error value of the current time step exceeding the benchmark value, the confidence interval width is linearly expanded or reduced according to the proportional coefficient, the mean of the distribution model is used as the center of the interval, and the upper and lower bounds are calculated in combination with the adjusted width to generate a set of interval prediction results.

[0132] Based on the error distribution parameters (i.e., mean) of each entropy classification interval updated by calculation in step S402 and standard deviation ), and the error benchmark standard deviation of the corresponding category obtained in S401 (i.e. the values in Table 2).

[0133] At a certain point in time for the test set When generating interval predictions, first determine the entropy category to which the point belongs The center of the point prediction interval is set to the point prediction value obtained in S303 Add the updated error mean of this category Perform bias correction: .

[0134] Basic prediction interval half-width By the corresponding confidence level (e.g. 95%) normal distribution quantile (For 95%, ) and the updated error standard deviation of the category The product of is determined by: .

[0135] Then, according to the current time step Some characteristics of the recent error (such as the mean absolute value of or uncertainty indication of the predicted value of the point itself) and the standard deviation of the baseline error of the entropy classification to which the point belongs The interval width is adjusted by comparing the values of Used for adjustments, for example, if recent errors are significantly larger than the benchmark, ,on the contrary .definition ,in is an adjustment coefficient, , is the actual prediction error at the previous time step, and the denominator From Table 2. , then the interval will be expanded, otherwise it may be reduced (it is necessary to ensure is positive). A specific adjusted half-width is calculated as , that is, the adjustment factor is limited to between 0.5 and 2.0 to prevent overscaling.

[0136] Finally, at this point in time The upper and lower bounds of the prediction interval are: Lower limit This process is repeated for all time points in the test set to generate a complete set of interval forecast results.

[0137] A wind speed prediction system based on entropy clustering, the wind speed prediction system based on entropy clustering is used to execute the above-mentioned wind speed prediction method based on entropy clustering, the system comprising:

[0138] The decomposition and clustering module is used to decompose the wind speed sequence of the training set into a subsequence set through the permutation entropy algorithm, divide the subsequences into subsequences using a sliding window, calculate the probability distribution of wind speed values within the window, generate a dynamic classification threshold based on the normalized entropy function, divide the subsequences into a high entropy component set, a medium entropy component set, and a low entropy component set, integrate them into an entropy classification result set, and pass the entropy classification result set to the component mapping module;

[0139] The medium entropy component set is the set of subsequences whose normalized entropy values are in the middle 30% interval of the dynamic classification threshold;

[0140] The component mapping module is used to obtain the original wind speed sequence of the training set and the entropy classification result set, generate a component mapping model through GRU network training, input the wind speed sequence of the test set into the model to output the entropy component set of the test set, and pass the entropy component set of the test set to the weight prediction module;

[0141] The weight prediction module is used to call the entropy component set of the test set, input the high entropy component set into the bidirectional gated recurrent unit to generate a high entropy prediction value, input the medium entropy component set into the long short-term memory model to generate a medium entropy prediction value, and input the low entropy component set into the support vector regression model to generate a low entropy prediction value. The initial weight is assigned based on the entropy value ratio, the weight parameter is optimized using the chaotic mapping algorithm, and the wind speed point prediction results are generated by superposition. The wind speed point prediction results are then passed to the interval correction module;

[0142] The interval correction module is used to establish a dynamic error benchmark set based on the prediction error standard deviation of the entropy classification interval of the training set, call the wind speed point prediction results and the real-time error of the test set, calculate the relative difference between the real-time error and the benchmark value through the error distribution model, adjust the confidence interval width, and output the interval prediction result set.

[0143] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A wind speed prediction method based on entropy clustering, characterized in that: The following steps are involved: S1: Perform multi-scale decomposition of the wind speed sequence in the training set using the permutation entropy algorithm to generate a set of subsequences. Perform sliding window probability distribution calculation on the multiple subsequences, set a dynamic classification threshold based on the normalized entropy value, and divide the subsequence set into an entropy classification result set. S2: Inputting the original wind speed sequence of the training set and the entropy classification result set into the GRU network to perform gated state parameter training, generating a component mapping model, inputting the original wind speed sequence of the test set into the component mapping model to output the entropy component set of the test set; S3: Inputting the high entropy component in the entropy component set of the test set into a bidirectional gated recurrent unit to generate a high entropy prediction result, inputting the medium entropy component into a long short-term memory model to generate a medium entropy prediction result, and inputting the low entropy component into a support vector regression model to generate a low entropy prediction result, allocating initial weights based on the entropy value ratio, iteratively optimizing the weight parameters using a chaotic mapping algorithm, and performing weighted superposition on the three types of prediction results to generate a wind speed point prediction result; The entropy classification result set specifically includes a high entropy subsequence category, a medium entropy subsequence category, and a low entropy subsequence category. The test set entropy component set includes a high entropy component, a medium entropy component, and a low entropy component. The wind speed point prediction result specifically refers to a high entropy prediction result, a medium entropy prediction result, a low entropy prediction result, and an optimized weight parameter superposition value. The normalized entropy value is obtained by linear mapping formula Calculate, where Represents the normalized value of the permutation entropy of the current window, represents the permutation entropy value of the i-th sliding window, Represents the maximum value of the permutation entropy value in all sliding windows, Represents the minimum value of the permutation entropy value in all sliding windows. The dynamic classification threshold is the coordinate value corresponding to the extreme point of the second-order derivative of the entropy distribution density curve after mapping. The extreme point represents the location where the density change is most significant and serves as the classification boundary for category division. The dynamic classification threshold is the coordinate value corresponding to the extreme point of the second-order derivative of the entropy distribution density curve after mapping; The GRU network includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is equal to the time step of the wind speed sequence. The hidden layer has two layers, each containing 128 neurons. The number of neurons in the output layer is 32. The chaotic mapping algorithm adopts the Logistic mapping function Generate a random sequence where is the chaotic sequence value of the nth iteration, is the chaotic sequence value of the n+1th iteration, and the fitness function adopts the mean square error form, that is, ,in, represents the mean square error of the prediction results, is the sample size, is the jth true wind speed value, For the j-th predicted wind speed value, the gradient descent method is used to update the weight perturbation amplitude. When optimizing the weight, the initial weight is distributed according to the entropy ratio. Finally, through multiple iterative optimizations, the converged optimal weight parameters are obtained.

2. The wind speed prediction method based on entropy clustering according to claim 1, characterized in that: The steps for obtaining the entropy classification result set are specifically as follows: S101: Obtain a training set wind speed sequence, call a permutation entropy algorithm to set embedding dimension parameters and delay parameters, construct a multidimensional phase space based on the embedding dimension parameters, perform time delay slicing on the sequence using the delay parameters, reconstruct phase space trajectories at multiple time scales, and decompose the original sequence based on similarity differences between the trajectories to generate multi-scale decomposition subsequences; S102: Based on the multi-scale decomposition subsequence, set the coverage length parameter and the moving step parameter of the sliding window, successively intercept the local data segments of the subsequence according to the step parameter, count the frequency of occurrence of differentiated wind speed values in each window, divide the frequency by the window length parameter, calculate the probability distribution of the wind speed values corresponding to the multiple windows, and generate a probability distribution matrix; The window length parameter is unified in seconds, and the moving step parameter is 1 second; S103: extracting the permutation entropy value of each window from the probability distribution matrix, locating the maximum and minimum values of the entropy value sequence, mapping the entropy value to the interval of 0-1, analyzing the distribution density curve of the mapped entropy value, calculating the second-order derivative of the curvature as the entropy value changes, locating the entropy value coordinates corresponding to the derivative extreme point, and setting the coordinate value as the classification threshold, dividing the high entropy and low entropy categories, and generating a dynamic classification threshold result; The second-order derivative is calculated by a numerical difference method, and the extreme point determination criterion is that the sign of the derivative changes from positive to negative or from negative to positive.

3. The wind speed prediction method based on entropy clustering according to claim 2, characterized in that: The steps for obtaining the test set entropy component set are specifically as follows: S201: Obtain the original wind speed sequence of the training set and the entropy classification result set, construct a GRU network input layer based on the time step characteristics of the wind speed sequence, and at the same time, normalize and initialize the input layer weights according to the multi-category probability distribution in the entropy classification result set, calculate the weight error of the forget gate and the input gate time step by time step using the gradient descent method, iteratively update the gate control unit parameters through chain derivation, and generate a gate control state parameter set; The normalization initialization adopts the Xavier method, and the learning rate of the gradient descent method is set to ; S202: calling the gate state parameter set, dividing the training set wind speed sequence into subsequence segments, and establishing a mapping index between the subsequence and the hidden layer state based on the entropy classification label, calculating the error gradient between the hidden layer state and the label through back propagation, updating the output gate weight and the fully connected layer parameters, and generating the component mapping model parameters; S203: Segment the test set wind speed sequence into subsequence segments based on the component mapping model parameters, input the GRU network to calculate the hidden layer state vector, extract the neuron activation value corresponding to the entropy classification label, splice the activation value sequence by time step, and generate a test set entropy component set.

4. The wind speed prediction method based on entropy clustering according to claim 3, characterized in that: The steps for obtaining the wind speed point prediction result are specifically as follows: S301: calling the test set entropy component set, inputting the high entropy component into the bidirectional gated recurrent unit based on the entropy classification label, expanding the time step forward and reverse hidden state transfer, calculating the output value of the multi-time step gated unit, inputting the medium entropy component into the long short-term memory model, filtering the time-dependent features according to the forget gate and input gate states, inputting the low entropy component into the support vector regression model, mapping the feature space using the radial basis kernel function and solving the regression hyperplane, and generating a component prediction result set; The radial basis kernel function bandwidth parameter ,The hyperplane solution adopts the sequential minimum optimization algorithm; S302: Based on the initial entropy value ratios of the high entropy, medium entropy, and low entropy components in the component prediction result set, the proportion of the high entropy component prediction result in the total entropy value is calculated, and the proportion is used as the initial weight distribution coefficient. At the same time, the Logistic mapping function in the chaotic mapping algorithm is called to generate a random sequence. The mean square error of the prediction result is used as the fitness function, and the perturbation amplitude of the weight parameter is iteratively updated to generate the optimized weight parameter. The disturbance amplitude update step size is ; S303: Call the component prediction result set and the optimized weight parameter, perform a scalar product of the high entropy prediction result and the high entropy weight coefficient in the optimized weight, synchronously calculate the product of the medium entropy and low entropy prediction results and the corresponding weights, align and superimpose the three types of weighted results according to the time step, and generate a wind speed point prediction result.

5. The wind speed prediction method based on entropy clustering according to claim 4, characterized in that: The method further comprises: S4: Establish a dynamic error benchmark set based on the error standard deviation of multiple entropy classification intervals in the training set, input the wind speed point prediction results and the real-time error of the test set into the error distribution model, adjust the confidence interval width according to the relative relationship between the error and the benchmark value, and output the interval prediction result set.

6. The wind speed prediction method based on entropy clustering according to claim 5, characterized in that: The interval prediction result set includes a dynamic error benchmark, real-time error distribution data, a confidence interval width adjustment parameter, and a probability interval range; The dynamic error benchmark set is an error standard deviation sequence arranged in ascending order by entropy classification labels, and the confidence interval width adjustment formula is: ,in, is the width of the current confidence interval, is the proportionality coefficient, is the wind speed prediction error at time t, is the error benchmark value of the kth entropy classification interval, and The difference between them reflects the uncertainty of the prediction. The confidence interval width is adjusted dynamically based on the difference, and the probability interval range is determined by the error distribution and the width parameter.

7. The wind speed prediction method based on entropy clustering according to claim 6, characterized in that: The steps for obtaining the interval prediction result set are specifically as follows: S401: Obtain prediction error sequences corresponding to multiple entropy classification intervals of the training set, divide the error data by interval, calculate the standard deviation of the error values in each entropy classification interval, and sort the standard deviations in ascending order according to the entropy classification labels to generate a dynamic error benchmark set; S402: calling the dynamic error benchmark set, aligning the wind speed point prediction result with the test set real-time error by time step, and calculating the absolute value of the difference between the error value of each time step and the corresponding entropy classification benchmark value based on the normal distribution setting of the error distribution model, and updating the mean and variance parameters of the distribution model using the maximum likelihood estimation method to generate error distribution parameters; The initial mean of the normal distribution is , the variance is ; S403: Based on the error distribution parameters, according to the proportion of the error value of the current time step exceeding the benchmark value, the confidence interval width is linearly expanded or reduced according to the proportional coefficient, the mean of the distribution model is used as the center of the interval, and the upper and lower bounds are calculated in combination with the adjusted width to generate an interval prediction result set.

8. A wind speed prediction system based on entropy clustering, characterized in that: The system is used to implement the wind speed prediction method based on entropy clustering according to any one of claims 1 to 7, and the system includes: The decomposition and clustering module is used to decompose the wind speed sequence of the training set into a subsequence set through the permutation entropy algorithm, divide the subsequences into subsequences using a sliding window, calculate the probability distribution of wind speed values within the window, generate a dynamic classification threshold based on the normalized entropy function, divide the subsequences into a high entropy component set, a medium entropy component set, and a low entropy component set, integrate them into an entropy classification result set, and pass the entropy classification result set to the component mapping module; The middle entropy component set is a set of subsequences whose normalized entropy values are in the middle 30% interval of the dynamic classification threshold; A component mapping module is used to obtain the original wind speed sequence of the training set and the entropy classification result set, generate a component mapping model through GRU network training, input the wind speed sequence of the test set into the model to output the entropy component set of the test set, and pass the entropy component set of the test set to the weight prediction module; A weight prediction module is used to call the test set entropy component set, input the high entropy component set into a bidirectional gated recurrent unit to generate a high entropy prediction value, input the medium entropy component set into a long short-term memory model to generate a medium entropy prediction value, input the low entropy component set into a support vector regression model to generate a low entropy prediction value, assign initial weights based on the entropy value ratio, optimize the weight parameters using a chaotic mapping algorithm, superimpose and generate wind speed point prediction results, and pass the wind speed point prediction results to an interval correction module; The interval correction module is used to establish a dynamic error benchmark set based on the prediction error standard deviation of the entropy classification interval of the training set, call the wind speed point prediction results and the real-time error of the test set, calculate the relative difference between the real-time error and the benchmark value through the error distribution model, adjust the confidence interval width, and output the interval prediction result set.

Images