Short-term wind speed prediction based on interval type-2 t-s fuzzy model with phase space reconstruction
By using phase space reconstruction and a type-2 TS fuzzy model, the problems of large errors and logical unanalyzability in short-term wind speed prediction are solved, achieving high-precision, real-time wind speed prediction, which is suitable for safety-critical systems in wind farms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TANGSHAN COLLEGE
- Filing Date
- 2026-03-12
- Publication Date
- 2026-06-19
AI Technical Summary
Existing short-term wind speed forecasting technologies suffer from large errors, long forecasting times, and unanalyzable logic, making it difficult to meet the real-time scheduling needs of wind farms.
A type 2 TS fuzzy model based on phase space reconstruction is adopted. Wind speed data is reconstructed through the phase space trajectory matrix. Irregular Gaussian membership function and linear function regular consequent are used. Particle swarm optimization and recursive least squares method are combined to optimize parameters to achieve wind speed prediction.
It improves the accuracy and robustness of wind speed prediction, provides an analysis of the physical meaning of wind speed changes, and meets the real-time scheduling needs of wind farms.
Smart Images
Figure CN122241072A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent prediction and advanced control technology for new energy power systems, and in particular to short-term wind speed prediction using a type 2 TS fuzzy model based on phase space reconstruction. Background Technology
[0002] With the global energy structure shifting towards renewable energy, wind power installed capacity has experienced exponential growth, exceeding 1,000 GW globally in 2023. Against this backdrop, the accuracy of wind speed forecasting has become a core factor affecting grid security and economic operation. According to the International Electrotechnical Commission (IEC) standard 61400-25, wind farms connected to the grid must provide short-term wind speed forecast data for the next 72 hours, and its accuracy directly determines power dispatch efficiency and wind curtailment rates.
[0003] Current wind speed forecasting technology faces three major technical bottlenecks: First, wind speed sequences exhibit typical non-stationary characteristics, influenced by atmospheric boundary layer turbulence, topographic flow, and thermal effects, resulting in a time-varying probability distribution. This causes traditional time series models (such as ARIMA and exponential smoothing) to experience a surge in prediction errors exceeding 35% in abrupt change scenarios. Second, wind speed systems are essentially low-dimensional chaotic dynamic systems, with measured Lyapunov exponent values ranging from 0.12 to 0.35. Small initial errors are amplified exponentially. While physical models based on numerical weather prediction (NWP) can capture weather-scale changes, they require supercomputer support, with single predictions taking over two hours, making it difficult to meet the 15-minute real-time scheduling requirements of wind farms. Finally, while existing data-driven models (such as LSTM and Transformer) can fit nonlinear relationships, their black-box nature makes their decision-making logic unanalyzable, leading European grid operators to explicitly exclude them from safety-critical systems. Summary of the Invention
[0004] Based on the above analysis, the present invention aims to provide a short-term wind speed prediction using an interval type 2 TS fuzzy model based on phase space reconstruction, in order to solve the problems of large error, long time, and unanalyzable logic in existing short-term wind speed predictions.
[0005] This invention provides a short-term wind speed prediction method based on a phase space reconstruction-based interval type 2 TS fuzzy model, comprising the following steps: Collect wind speed data up to the current moment and preprocess it to obtain a wind speed time series; The wind speed time series is reconstructed in phase space to obtain the phase space trajectory matrix; Input the phase space trajectory matrix into the trained interval type 2 TS fuzzy model to output short-term wind speed prediction values; In the interval type 2 TS fuzzy model, the antecedent is described by an irregular Gaussian membership function, which describes the fuzzy characteristics of the input variable. The left and right standard deviations of the irregular Gaussian membership function are adjustable.
[0006] Based on further improvements to the above method, the model training process specifically includes: Preprocessing: Using each column of the phase space trajectory matrix as input variables, the fuzzy characteristics of the input variables are described by the membership function of the interval type II fuzzy set, and the parameters of the membership function are adjustable; Rule base construction: Multiple fuzzy rules are formed based on the phase space trajectory matrix, and the consequent of each rule is a linear function; Fuzzy inference and type reduction: Based on the membership function, the activation interval of each rule is calculated by the fuzzy inference engine, and the fuzzy inference result is output by combining the rule consequent. The fuzzy inference result is then processed by the type reduction algorithm to obtain the predicted value. Model parameter optimization: Multiple optimization algorithms are used to optimize the antecedent and consequent parameters, and the iteration continues until the error between the model prediction and the true value converges, thus obtaining a well-trained model.
[0007] Based on the above method, a further improvement is made to map the one-dimensional wind speed sequence x(t) into an m-dimensional phase space trajectory matrix using the delayed coordinate method:
[0008] Where τ is the delay time and m is the embedding dimension. P Number of phase points , N Let x(t) be the length of the one-dimensional wind speed sequence; the parameter τ is determined by the mutual information method; m is determined by the spurious nearest neighbor method.
[0009] Based on further improvements to the above method, the parameters t Determined using the mutual information method, specifically: First, the sequence x ( t ) divided into K Divide the data into several equally spaced intervals and count the frequency of occurrence of each interval to obtain the probability density function. p ( x i Similarly, calculate the delay sequence. probability density p ( x j ), and joint probability density p ( x i , x j ); then based on p ( x i ),p ( x j ), p ( x i , x j )Calculate sequence x ( t ) and delayed sequence Mutual information function ,Pick At the first minimum point t As a delay time t .
[0010] Based on a further improvement to the above method, the parameter m is determined using the spurious nearest neighbor method, specifically: based on the determined delay time. t Construct m-dimensional phase space vectors , where i = 1, 2, ..., P, for each vector Find the closest point in phase space. The distance is denoted as d. m (i) Increase the dimension to m+1 and construct an m+1 dimensional phase space vector. and ,calculate and Distance d in m+1 dimensional space m+1 (i), if If the threshold T is exceeded, then the nearest neighbor is determined. For false nearest neighbors, gradually increase m to the preset maximum value, calculate the proportion of false nearest neighbors in the total number of vectors under the current dimension, and repeat the above process when the proportion is less than the preset value. The current m is the optimal embedding dimension.
[0011] Based on a further improvement of the above method, the irregular Gaussian membership function is: m ( x )=[ lower ( x ), m upper ( x )],in and These are the lower and upper membership functions, respectively. For when time , For when time Adjustable parameters and Control, the formula is as follows:
[0012] in, For input variables, These are the irregular Gaussian membership functions. , , , These are the adjustable parameters of the irregular Gaussian membership function. δ is a user-defined constant.
[0013] Based on a further improvement of the above method, the consequent is a linear function. ,in For the first l Output of the rules , For consequent parameters, For the input variables, k = 0, 1, 2, ..., m, .
[0014] Based on a further improvement of the above method, the rule expression is:
[0015] in x 1 ,x 2 ,…, x m It is an input variable. A l1 ，A l2 ，…，A lm It is an interval type 2 fuzzy set. , It is an output variable. It is a consequent parameter.
[0016] Based on the further improvement of the above method, the antecedent and consequent parameters are optimized using multiple optimization algorithms. The initial values of the antecedent and consequent parameters are random numbers, and they are optimized by particle swarm optimization algorithm and recursive least squares method, respectively.
[0017] Based on a further improvement of the above method, the activation interval of each rule is calculated using a fuzzy inference engine based on the membership function, and the fuzzy inference result is output in combination with the rule consequent. The fuzzy inference result is then processed by a reduction algorithm to obtain the predicted value, including: calculating the lower and upper limits of the activation of each rule using a product inference engine, and each rule... The output is The Karnik-Mendel algorithm is used based on the mean of the upper and lower limits of activation and Calculate the initial centroid c, and then calculate the output. Sort, find k such that Based on activation upper and lower limits and output Calculate the lower limit centroid and upper limit centroid If the difference between the upper and lower limit centroids and the initial centroid is less than the preset minimum value, then stop the iteration; otherwise, update c to... Repeat the above steps to output the interval. To obtain the final predicted value .
[0018] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. By capturing the dynamic uncertainty of wind speed through phase space reconstruction and processing the uncertainty of the rule antecedent through interval type 2 fuzzy set, the wind speed prediction is more accurate, more robust and more reliable in complex fluctuation scenarios. 2. An asymmetric Gaussian membership function with adjustable left and right standard deviations is used, which is more suitable for wind speed data distribution than the conventional Gaussian membership function, thus making wind speed prediction more accurate. 3. The consequent of the interval type 2 TS fuzzy rule is a linear function, which can provide an analysis of the physical meaning of wind speed changes.
[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description
[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.
[0021] Figure 1 This is a flowchart of the short-term wind speed prediction using the interval type 2 TS fuzzy model based on phase space reconstruction in Embodiment 1 of the present invention. Figure 2 This is a graph showing actual data from a wind farm in Embodiment 1 of the present invention; Figure 3 The wind speed time series prediction results and error graph are shown in Embodiment 1 of the present invention. Detailed Implementation
[0022] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.
[0023] A specific embodiment of the present invention discloses a short-term wind speed prediction method based on a phase space reconstruction interval type 2 TS fuzzy model, such as... Figure 1 As shown, it includes the following steps: S1: Collect wind speed data up to the current moment and preprocess it to obtain a wind speed time series.
[0024] It should be noted that preprocessing of wind speed data includes data cleaning and noise reduction; data cleaning includes using linear interpolation to fill in missing values, the formula being... Outliers are identified using the 3σ criterion and the moving average of adjacent time points is used. Replacement; for denoising wind speed data, wavelet thresholding is used. First, the wind speed sequence is decomposed into wavelets (e.g., using db4 wavelets, decomposed into 3 levels), and then soft thresholding is applied to high-frequency coefficients (threshold). l = s 2log N , s The standard deviation of noise. N (where is the sequence length), and then the denoised sequence is obtained through wavelet reconstruction.
[0025] S2: Reconstruct the phase space of the wind speed time series to obtain the phase space trajectory matrix.
[0026] Specifically, the delayed coordinate method is used to map the one-dimensional wind speed sequence x(t) into an m-dimensional phase space trajectory matrix:
[0027] Where τ is the delay time and m is the embedding dimension. P Number of phase points , N Let x(t) be the length of the one-dimensional wind speed sequence; the parameter τ is determined by the mutual information method; m is determined by the spurious nearest neighbor method.
[0028] Specifically, the parameters t Determined using the mutual information method, specifically: First, the sequence x ( t ) divided into K Divide the data into several equally spaced intervals and count the frequency of occurrence of each interval to obtain the probability density function. p ( a Similarly, the delayed sequence Divided into K Calculate the probability density of equidistant intervals. p ( b ), and joint probability density p ( a , b ); then based on p (a ), p ( b ), p ( a , b) Calculate sequence x ( t ) and delayed sequence Mutual information function ,Pick At the first minimum point t As a delay time t .
[0029] It should be noted that, a It is the original wind speed sequence x ( t Discrete values of ) b It's a delay. t Post-sequence x ( t + t Discrete values of ) p ( a )express x ( t )= a The probability, p ( b ) represents a delayed sequence = b The probability, joint probability density p ( a , b ) Statistical analysis of the original sequence using histograms a And the delayed sequence is taken b The frequency proportion is obtained from the joint probability density. p ( a , b )beg p ( a ), p ( b ):
[0030] The formula for the mutual information function is as follows:
[0031] Specifically, the parameter m is determined using the spurious nearest neighbor method, as follows: S11: Based on the determined delay time t Construct m-dimensional phase space vectors
[0032] Where m = 2, 3, ..., m max, i=1,2,…,P , N Given the length of the one-dimensional wind speed sequence x(t), for each vector Find the closest point in phase space. Its distance is denoted as
[0033] S12: Construct an m+1 dimensional phase space and find the corresponding... and Two points with the same subscript and ,calculate and distance ,like If the threshold T is reached, then the nearest neighbor in m dimensions is determined. For false nearest neighbors, obtain the number of false nearest neighbors in the m-dimensional phase space, and calculate the proportion of false nearest neighbors in the m-dimensional space in the total number of vectors; S13: Let m = m + 1, repeat steps S11 and S12 to obtain the proportion of false nearest neighbors in the total number of vectors under the current dimension. If the proportion is less than the preset value, then the current m is the optimal embedding dimension.
[0034] It should be noted that the total number of phase space vectors It can be seen that when m = m + 1, Therefore, it is possible for j>P. m+1 In this situation, at this time The point corresponding to dimension m+1 If it does not exist, then directly determine the m-dimensional dimension. It is a false nearest neighbor.
[0035] It should be noted that, The threshold T indicates that as the dimension increases, the distance between the same pair of neighboring points increases significantly. The core of phase space reconstruction is to embed a one-dimensional time series into a sufficiently high multi-dimensional space so that the dynamic characteristics of the system are fully "unfolded". When the embedding dimension m is insufficient, the phase space cannot fully unfold all the degrees of freedom of the system, causing points that are originally far apart in the "real physical space" to be incorrectly "compressed" very close in the low-dimensional projection (similar to projecting a three-dimensional sphere onto a two-dimensional plane, where two distant points on the sphere may be very close in the projection). When the dimension increases to m+1, the phase space "unfolds" more degrees of freedom, and the originally compressed distance will "restore" the true value. If two points that are very close in the low dimension suddenly have a significantly increased distance in the high dimension, it means that their "proximity" in the low dimension is an illusion caused by insufficient dimension, rather than a true dynamic proximity. These points are "false nearest neighbors".
[0036] For example, given a simplified wind speed time series {x(t)}=[1.2,1.8,2.3,2.0,1.6,1.4], τ=1, sequence length N=6, threshold T=1.5, and a preset value of 5% for the proportion of false nearest neighbors, the process of determining m using the false nearest neighbor method is as follows: Initialize the embedding dimension m=2, and construct a 2D phase space vector. Calculate the nearest neighbor and distance for each vector.
[0037] Constructing 3D phase space vectors Calculate the 3D distance between 2D neighbor pairs and identify false nearest neighbors.
[0038] Determined by threshold, no False nearest neighbor, no The spurious nearest neighbor, due to It does not exist, therefore yes , False nearest neighbor, yes Since the false nearest neighbors are found, when m=2, the proportion of false nearest neighbors is 3 / 5=60%>5%. Similarly, when m=3, the proportion of false nearest neighbors is 33.3% > 5%; when m=4, the proportion of false nearest neighbors is 0% < 5%, which satisfies the condition that the proportion of false nearest neighbors is less than the preset value. Therefore, the optimal embedding dimension is determined to be 4.
[0039] S3: Input the phase space trajectory matrix into the trained interval type 2 TS fuzzy model and output the short-term wind speed prediction value; In the interval type 2 TS fuzzy model, the antecedent is described by an irregular Gaussian membership function, which describes the fuzzy characteristics of the input variable. The left and right standard deviations of the irregular Gaussian membership function are adjustable.
[0040] Specifically, the model training process includes: Preprocessing: Using each column of the phase space trajectory matrix as input variables, the fuzzy characteristics of the input variables are described by the membership function of the interval type II fuzzy set, and the parameters of the membership function are adjustable; Rule base construction: Multiple fuzzy rules are formed based on the phase space trajectory matrix, and the consequent of each rule is a linear function; Fuzzy inference and type reduction: Based on the membership function, the activation interval of each rule is calculated by the fuzzy inference engine, and the fuzzy inference result is output by combining the rule consequent. The fuzzy inference result is then processed by the type reduction algorithm to obtain the predicted value. Model parameter optimization: Multiple optimization algorithms are used to optimize the antecedent and consequent parameters, and the iteration continues until the error between the model prediction and the true value converges, thus obtaining a well-trained model.
[0041] Specifically, the irregular Gaussian membership function is: m ( x )=[ lower ( x ), upper ( x )],in and These are the lower and upper membership functions, respectively. For when time , For when time Adjustable parameters and Control, the formula is as follows:
[0042] in, For input variables, These are the irregular Gaussian membership functions. , , , These are the adjustable parameters of the irregular Gaussian membership function. δ is a user-defined constant.
[0043] Specifically, the consequent is a linear function. ,in For the first l Output of the rules ， For consequent parameters, For the input variables, k = 0, 1, 2, ..., m, .
[0044] Specifically, the rule expression is:
[0045] in x 1 ,x 2 ,…, x m It is an input variable. A l1 ，Al2 ，…，A lm It is an interval type 2 fuzzy set. , It is an output variable. It is a consequent parameter.
[0046] Specifically, the antecedent and consequent parameters are optimized using multiple optimization algorithms. The initial values of the antecedent and consequent parameters are random numbers, and they are optimized using particle swarm optimization algorithm and recursive least squares method, respectively.
[0047] It should be noted that the initial values of the antecedent and consequent parameters are random numbers, and are updated using the particle swarm optimization algorithm and the recursive least squares method, respectively. The objective function of the particle swarm optimization algorithm is the root mean square error (RMSE) between the model's predicted value and the actual wind speed value. Each particle represents a set of antecedent parameters. First, the particle swarm is initialized, then the objective function of each particle is calculated, and the position and velocity of the particles are updated. This process is iterated until the objective function converges, and the optimal antecedent parameters are output. The consequent parameters are updated using the recursive least squares method for each rule. Calculate the activation range of each sample in the training set. The weighted recursive least squares method is used to calculate the mean activation value. As weights, the objective function is ,in This is the actual wind speed value. For the first The predicted values are used to recursively update the parameters, iterating until convergence, and then the optimal consequent parameters are output.
[0048] Specifically, the activation interval of each rule is calculated using a fuzzy inference engine based on the membership function, and the fuzzy inference result is output in combination with the rule consequent. The fuzzy inference result is then processed by a reduction algorithm to obtain the predicted value, including: calculating the lower and upper limits of activation for each rule using a product inference engine, and each rule... The output is The Karnik-Mendel algorithm is used based on the mean of the upper and lower limits of activation and Calculate the initial centroid c, and then calculate the output. Sort, find k such that Based on activation upper and lower limits and output Calculate the lower limit centroid and upper limit centroid If the difference between the upper and lower limit centroids and the initial centroid is less than the preset minimum value, then stop the iteration; otherwise, update c to... Repeat the above steps to output the interval. To obtain the final predicted value .
[0049] It should be noted that the minimum activation level for each rule is [not specified]. The upper limit is ,in, and The output of each rule for the lower and upper membership functions is as follows: The initial centroid is the mean activation level. The lower limit centroid is The upper limit centroid is .
[0050] For example, using wind speed data from a Colorado wind farm from August 1st to August 15th, 2020 as experimental data, the aforementioned short-term wind speed prediction method based on the interval type 2 TS fuzzy model using phase space reconstruction was applied to predict wind speed. 4320 data points were collected at a 5-minute sampling interval. Figure 2 ; The delay time was determined using the mutual information method and the false nearest neighbor method, respectively. and embedding dimension The wind speed sequence was reconstructed into phase space, resulting in 3972 valid phase points. 3178 points were used for the training set, and the remaining points were allocated for testing. The test set was input into the trained model to obtain prediction results and prediction errors, as shown below. Figure 3 ; Different models were used to predict the same wind speed data, and the root mean square error and mean absolute error of the prediction results were calculated. The results are shown in the table below: Table 1 Error Analysis of Wind Speed Time Series Model
[0051] Wherein, RMSE represents the root mean square error, and MAE represents the mean absolute error. The TSFS-PSO model does not perform phase space reconstruction of the time series, but only uses the PSO algorithm to optimize the parameters in the type 1 TS fuzzy model. The TS-PAR model is a type 1 fuzzy model that does not use the particle swarm optimization algorithm and uses phase space reconstruction for data preprocessing. The TS-PAR-PSO model is a type 1 fuzzy model that uses the particle swarm optimization algorithm and uses phase space reconstruction for data preprocessing. IT2FS-PAR-PSO represents the model proposed in this invention. It can be seen that the model proposed in this invention has the smallest root mean square error and mean absolute error in wind speed prediction results, and the model prediction accuracy is the highest.
[0052] Compared with existing technologies, the short-term wind speed prediction based on the interval type 2 TS fuzzy model provided in this embodiment captures the dynamic uncertainty of wind speed through phase space reconstruction and processes the uncertainty of the rule antecedent through the interval type 2 fuzzy set; it adopts an asymmetric Gaussian membership function with adjustable left and right standard deviations, which is more suitable for wind speed data distribution than the conventional Gaussian membership function; the consequent of the interval type 2 TS fuzzy rule is a linear function, which can provide an analysis of the physical meaning of wind speed changes.
[0053] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0054] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A short-term wind speed prediction based on phase space reconstruction of interval type-2 T-S fuzzy model, characterized in that, Includes the following steps: Collect wind speed data up to the current moment and preprocess it to obtain a wind speed time series; The wind speed time series is reconstructed in phase space to obtain the phase space trajectory matrix; Input the phase space trajectory matrix into the trained interval type 2 TS fuzzy model to output short-term wind speed prediction values; In the interval type 2 TS fuzzy model, the antecedent is described by an irregular Gaussian membership function, which describes the fuzzy characteristics of the input variable. The left and right standard deviations of the irregular Gaussian membership function are adjustable.
2. The short-term wind speed prediction based on phase space reconstruction of interval type-2 T-S fuzzy model according to claim 1, characterized in that, The model training process specifically includes: Preprocessing: Using each column of the phase space trajectory matrix as input variables, the fuzzy characteristics of the input variables are described by the membership function of the interval type II fuzzy set, and the parameters of the membership function are adjustable; Rule base construction: Multiple fuzzy rules are formed based on the phase space trajectory matrix, and the consequent of each rule is a linear function; Fuzzy inference and type reduction: Based on the membership function, the activation interval of each rule is calculated by the fuzzy inference engine, and the fuzzy inference result is output by combining the rule consequent. The fuzzy inference result is then processed by the type reduction algorithm to obtain the predicted value. Model parameter optimization: Multiple optimization algorithms are used to optimize the antecedent and consequent parameters, and the iteration continues until the error between the model prediction and the true value converges, thus obtaining a well-trained model.
3. The short-term wind speed prediction based on phase space reconstruction interval type-2 T-S fuzzy model according to claim 1, characterized in that, The one-dimensional wind speed sequence x(t) is mapped to an m-dimensional phase space trajectory matrix using the delayed coordinate method: where τ is the delay time, m is the embedding dimension, P is the number of phase points , N is the length of the one-dimensional wind speed series x(t); the parameter τ is determined by the mutual information method; m is determined by the false nearest neighbor method.
4. The short-term wind speed prediction using the interval type 2 TS fuzzy model based on phase space reconstruction according to claim 1, characterized in that, The parameters τ Determined by the mutual information method, specifically: First, the sequence x ( t ) divided into K Divide the data into several equally spaced intervals and count the frequency of occurrence of each interval to obtain the probability density function. p ( x i Similarly, calculate the delay sequence. probability density p ( x j ), and joint probability density p ( x i , x j ); then based on p ( x i ), p ( x j ), p ( x i , x j )Calculate sequence x ( t ) and delayed sequence Mutual information function ,Pick At the first minimum point τ As a delay time τ .
5. The short-term wind speed prediction using the interval type 2 TS fuzzy model based on phase space reconstruction according to claim 1, characterized in that, The parameter m is determined by a false nearest neighbor method, specifically: based on the determined delay time τ Constructing an m-dimensional phase space vector Where i=1, 2, …, P, for each vector Finding the nearest point in the phase space The distance is recorded as d m (i), increasing the dimension to m+1, constructing an m+1-dimensional phase space vector And Calculating And The distance d in the m+1-dimensional space m+1 (i), if > threshold T, then determine that the nearest neighbor point is a false nearest neighbor point, gradually increasing m to a preset maximum value, repeating the above process, calculating the proportion of the false nearest neighbor points in the total number of vectors under the current dimension, when the proportion is less than a preset value, the current m is the optimal embedding dimension.
6. The short-term wind speed prediction using the interval type 2 TS fuzzy model based on phase space reconstruction according to claim 1, characterized in that, The irregular Gaussian membership function is: μ ( x )=[ μlower ( x ), μupper ( x )],in and These are the lower and upper membership functions, respectively. For when time , For when time Adjustable parameters and Control, the formula is as follows: in, For input variables, These are the irregular Gaussian membership functions. , , , These are the adjustable parameters of the irregular Gaussian membership function. δ is a user-defined constant.
7. The model training process according to claim 2, characterized in that, The consequent is a linear function. ,in For the first l Output of the rules , For consequent parameters, For the input variables, k = 0, 1, 2, ..., m, .
8. The model training process according to claim 2, characterized in that, The rule expression is: in x 1 ,x 2 ,…, x m It is an input variable. A l1 ，A l2 ，…，A lm It is an interval type 2 fuzzy set. , It is an output variable. It is a consequent parameter.
9. The model training process according to claim 2, characterized in that, The parameters of the antecedent and consequent are optimized using a variety of optimization algorithms. The initial values of the antecedent and consequent parameters are random numbers, and the optimization is performed by particle swarm optimization algorithm and recursive least squares method, respectively.
10. The model training process according to claim 2, characterized in that, The activation interval of each rule is calculated using a fuzzy inference engine based on the membership function. The fuzzy inference result is then output after combining the rule consequents. This fuzzy inference result is processed by a reduction algorithm to obtain the predicted value, including: calculating the lower and upper limits of activation for each rule using a product inference engine; and for each rule... The output is The Karnik-Mendel algorithm is used based on the mean of the upper and lower limits of activation and Calculate the initial centroid c, and then calculate the output. Sort, find k such that Based on activation upper and lower limits and output Calculate the lower limit centroid and upper limit centroid If the difference between the upper and lower limit centroids and the initial centroid is less than the preset minimum value, then stop the iteration; otherwise, update c to... Repeat the above steps to output the interval. To obtain the final predicted value .