Short-time wind speed prediction method and device based on mountainous area mixed strong wind classification and storage medium

Abstract

The application provides a short-time wind speed prediction method and device based on mountainous area mixed strong wind classification and a storage medium, relates to the technical field of short-time wind speed prediction, and comprises the following steps: performing mixed strong wind classification on measured meteorological data of a target area, obtaining periodic thermal driving wind, strong wind cooling and non-special type strong wind; performing short-time wind speed prediction of the non-special type strong wind by using SHO-VMD-SOA-KELM which is constructed based on variational mode decomposition optimized by a spot hyena algorithm and a kernel extreme learning machine optimized by a seagull optimization algorithm; performing short-time wind speed prediction of the periodic thermal driving wind based on probability prediction by using a mixed Gaussian model based on wind speed grading by using Bayes-Bi-LSTM-ECFRM based on environmental variable comparison and error correction; performing short-time wind speed prediction of the strong wind cooling by using a model in which feature selection is performed by using a maximum information coefficient method and error obtained by performing regression on the features is preprocessed by using a completely self-adaptive noise ensemble empirical mode decomposition algorithm. By using the prediction method provided in the application, the short-time wind speed of the mountainous area mixed strong wind can be quickly and accurately predicted.

Description

Technical Field

[0001] This application relates to the field of short-term wind speed prediction technology, and in particular to a short-term wind speed prediction method, device and storage medium based on the classification of mixed strong winds in mountainous areas. Background Technology

[0002] Due to the terrain, mountainous areas frequently experience winds of varying intensities and combinations. Strong winds in mountainous areas are primarily formed by the combined effects of orographic lifting, changes in pressure gradients, and the funneling effect. When airflow passes through valleys or narrow passages, it is accelerated by the terrain, creating localized strong winds, with wind speeds potentially significantly higher than in plains areas. For example, in narrow valleys, airflow accelerates due to limited space; this phenomenon is called the "funneling effect," similar to how water flows faster through a narrow pipe. Furthermore, the complex terrain of mountainous areas, such as steep slopes and winding mountain roads, further exacerbates wind instability, making strong winds frequent and difficult to predict.

[0003] Given the frequent and destructive nature of strong winds in mountainous areas, forecasting research is crucial for disaster prevention and mitigation. Accurate short-term wind speed forecasts can provide early warnings of the need for building reinforcement, preventing structural resonance or envelope failure. Simultaneously, they provide critical information for driving safety, helping to plan sheltered routes or suspend traffic in high-risk areas, reducing traffic accidents and secondary disasters (such as vehicle damage caused by rockfalls).

[0004] Therefore, there is a need for a short-term wind speed prediction method, device, and storage medium based on the classification of mixed strong winds in mountainous areas, in order to at least partially solve the above-mentioned technical problems. Summary of the Invention

[0005] In view of this, embodiments of this application provide a short-time wind speed prediction method, apparatus and storage medium based on the classification of mixed strong winds in mountainous areas, so as to at least solve one of the problems in the prior art.

[0006] In a first aspect, embodiments of this application provide a short-time wind speed prediction method based on the classification of mixed strong winds in mountainous areas, the prediction method comprising: Acquire measured meteorological data of the target area, and classify mixed strong winds based on the measured meteorological data to obtain periodic thermal-driven winds, strong wind cooling, and non-special types of strong winds; A hybrid model, SHO-VMD-SOA-KELM, is constructed based on Variational Mode Decomposition (VMD) optimized by the Spotted Hyena Algorithm (SHO) and Kernel Extreme Learning Machine (KELM) optimized by the Seagull Algorithm (SOA), for short-term wind speed prediction of non-special types of strong winds. The Bayes-Bi-LSTM-ECFRM model is based on environmental variable comparison and error correction, and a Gaussian mixture model is introduced to perform probabilistic prediction of short-term wind speed of periodic thermal-driven wind based on wind speed classification. A model using the Maximum Information Coefficient (MIC) method for feature selection and a fully adaptive noise ensemble empirical mode decomposition algorithm to preprocess the error obtained from feature regression is used for short-term wind speed prediction during strong winds and temperature drops.

[0007] Secondly, embodiments of this application also provide a short-time wind speed prediction device based on the classification of mixed strong winds in mountainous areas, the prediction device comprising: Memory is used to store executable instructions for a computer; A processor, used to execute computer-executable instructions stored in the memory, to predict the implementation of the above-mentioned technical solution.

[0008] Thirdly, embodiments of this application also provide a storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the prediction method of the above-described technical solution.

[0009] According to the prediction method of this application, after screening and classifying the mixed strong winds in mountainous areas, short-term wind speed prediction is performed using specially designed hybrid models or algorithms based on the obtained periodic thermal-driven winds, wind cooling, and different types of non-special winds. Compared with the general prediction methods of existing technologies, this method can achieve accurate and rapid short-term wind speed prediction.

[0010] Additional advantages, objectives, and features of this application will be set forth in part in the description which follows, and will in part become apparent to those skilled in the art upon review of the following description, or may be learned by practice of the application. The objectives and other advantages of this application can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0011] Those skilled in the art will understand that the purposes and advantages that can be achieved with this application are not limited to those specifically described above, and that the above and other purposes that this application can achieve will be more clearly understood from the following detailed description. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, do not constitute a limitation thereof. The components in the drawings are not drawn to scale but are merely for illustrating the principles of this application. For ease of illustration and description of certain parts of this application, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to this application. In the drawings: Figure 1 This is a flowchart of a prediction method according to an embodiment of this application; Figure 2 This is a flowchart of a prediction method according to another embodiment of this application; Figure 3 This is a schematic diagram of different types of strong winds in a prediction method according to an embodiment of this application; Figure 4 This is a schematic diagram of the decomposition results of wind speed data of non-special types of strong winds by SHO-VMD in a prediction method according to an embodiment of this application. Figure 5 This is a flowchart illustrating the KELM parameter optimization using SOA in a prediction method according to an embodiment of this application. Figure 6 This is a framework diagram of a prediction model for periodic heat-driven wind in a prediction method according to an embodiment of this application; Figure 7 This is a framework diagram of a wind-induced temperature drop prediction model in a prediction method according to an embodiment of this application; Figure 8 This is a schematic diagram of a prediction device according to an embodiment of this application; Figure 9 This is a schematic diagram of a prediction system according to an embodiment of this application. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.

[0014] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the structures and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0015] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0016] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0017] In the following description, embodiments of the present application will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0018] First, refer to Figure 1 and Figure 2 This application describes a short-time wind speed prediction method 100 based on the classification of mixed strong winds in mountainous areas, according to an embodiment of this application. For example... Figure 1 and Figure 2As shown, the prediction method 100 may include steps S110 to S140, as detailed below: In step S110, measured meteorological data of the target area is obtained, and mixed strong winds are classified based on the measured meteorological data to obtain periodic heat-driven winds, strong wind cooling and non-special types of strong winds.

[0019] In step S120, a hybrid model SHO-VMD-SOA-KELM, which is constructed based on Variational Mode Decomposition (VMD) optimized by the Spotted Hyena Algorithm (SHO) and Kernel Extreme Learning Machine (KELM) optimized by the Seagull Algorithm (SOA), is used to predict short-term wind speeds for non-special types of strong winds.

[0020] In step S130, the Bayes-Bi-LSTM-ECFRM model is used for comparison of environmental variables and error correction, and a Gaussian mixture model is introduced to perform probabilistic prediction of short-term wind speed of periodic thermal-driven wind based on wind speed classification.

[0021] In step S140, the maximum information coefficient (MIC) method is used for feature selection, and the fully adaptive noise set empirical mode decomposition algorithm is used to preprocess the error obtained from feature regression to perform short-term wind speed prediction for strong winds and cooling.

[0022] According to the prediction method 100 of this application embodiment, strong winds in mountainous areas are first screened and classified. Then, based on the different types of periodic heat-driven winds, wind cooling, and non-special types of winds, a specially designed hybrid model SHO-VMD-SOA-KELM is used to predict short-term wind speed for non-special types of winds. For periodic heat-driven winds, a Bayes-Bi-LSTM-ECFRM model based on environmental variable comparison and error correction is used, and a Gaussian mixture model is introduced for short-term wind speed prediction. For wind cooling, a model that preprocesses the error obtained from feature regression using a fully adaptive noise ensemble empirical mode decomposition algorithm is used for short-term wind speed prediction. Compared with the general prediction methods of the prior art, this method can achieve accurate and rapid prediction of short-term wind speed.

[0023] The following will describe in detail the contents of the above steps of the prediction method 100 according to the embodiments of this application.

[0024] In the embodiments of this application, in step S110, measured meteorological data of the target area is obtained, and mixed strong winds are classified based on the measured meteorological data to obtain periodic heat-driven winds, strong wind cooling and non-special types of strong winds.

[0025] Specifically, corresponding measured meteorological data can be obtained by collecting meteorological monitoring data of wind fields in mountainous areas. Before classifying the measured meteorological data into mixed strong wind categories, the data can be preprocessed, such as by cleaning.

[0026] Data cleaning can be performed using a moving average window method to process the sampled data. For individual missing data, an arithmetic mean interpolation method is used to fill in the gaps in the sampled data. Low-quality data is inevitably collected during data transmission; data that deviates from the mean by more than three standard deviations will be considered outliers and discarded.

[0027] Then, the corresponding wind speed and temperature characteristic parameters are calculated, and strong winds are classified according to outlier principle, correlation principle, etc.

[0028] Specifically, the daily maximum 10-minute average wind speed is calculated based on measured meteorological data and used as a characteristic parameter for gale events. Measured meteorological data with a daily maximum 10-minute average wind speed greater than a set wind speed are selected as gale event data. For example, the daily maximum 10-minute average wind speed... >8 m / s Selected as a strong wind event.

[0029] Under the premise of a strong wind event, the Pearson correlation coefficient can be used as a characteristic parameter for periodic thermally driven winds. The p-value was used as the evaluation metric. Pearson correlation coefficients were selected. 0.4 and p-value 1% of high wind event data were used to avoid extreme correlations, as these are considered periodic heat-driven winds.

[0030] Under the premise of a strong wind event, the characteristic parameters for wind-induced temperature drop need to meet two conditions: First, a linear equation is generated using the 10-minute average temperature of the strong wind event data. Fitting, coefficients k The sign of the coefficient can express the overall pattern of temperature change on that day. First, it can serve as a standard for screening strong wind and cooling processes; second, it defines the period before the strong wind process ( ),middle( ),back( The daily average temperature before the strong wind and cooling process also needs to meet the following criteria: the daily average temperature before the strong wind process. Or the average daily temperature during a strong wind event. Strong wind events that simultaneously meet both of the above conditions are considered as wind-induced cooling events. Data on wind events excluding periodic heat-driven winds and wind-induced cooling events are considered non-special types of winds.

[0031] See Figure 3 It demonstrates periodic heat-driven winds, strong winds followed by cooling, and non-special types of strong winds.

[0032] In the embodiments of this application, in step S120, a hybrid model SHO-VMD-SOA-KELM is constructed based on Variational Mode Decomposition (VMD) optimized by the Spotted Hyena Algorithm (SHO) and Kernel Extreme Learning Machine (KELM) optimized by the Seagull Algorithm (SOA) to predict short-term wind speeds for non-special types of strong winds.

[0033] Step S120 may include the following steps: Step S121: The Spotted Hyena Algorithm (SHO) is used to optimize the parameters of Variational Mode Decomposition (VMD) and select the best parameters to construct the SHO-VMD model.

[0034] Step S122: The SHO-VMD model is used to decompose the wind speed data of non-special types of strong winds to obtain subsequences with different characteristic patterns. See [link to relevant documentation] Figure 4 This demonstrates the decomposition results of the SHO-VMD model on wind speed data that are not specific types of strong winds.

[0035] Step S123: The Seagull Optimization Algorithm (SOA) is used to optimize the parameters of the Kernel Extreme Learning Machine (KELM) and select the best parameters to construct the SOA-KELM model.

[0036] Step S124: Input each subsequence into the SOA-KELM model for prediction, and add the prediction results of each subsequence to obtain the final short-term wind speed prediction result.

[0037] Specifically, the hybrid model SHO-VMD-SOA-KELM consists of two parts: the first part is Variational Mode Decomposition (VMD) improved based on the Spotted Hyena Optimizer (SHO) algorithm, and the second part is Kernel Extreme Learning Machine (KELM) improved based on the Seagull Optimization Algorithm (SOA) algorithm. The first part is as follows: (1) VMD variational mode decomposition algorithm in, For the set of components of all K modes, To sample wind speed signals, The modal components obtained from the decomposition; K is the number of modes. for The corresponding frequency center; It follows the Dirac distribution; This is a convolution operation, where t is the sampling time. This indicates taking the partial derivative of the function. The imaginary unit *st* in mathematics represents the introduction of constraints.

[0038] (2) Lagrange Transform After the decomposition is completed, the VMD decomposition results obtained above will be used to further refine the decomposition process. , Perform a Lagrange transformation. As a Lagrange operator, it has the significance of maintaining the strictness of the constraints; As a penalty factor, the bandwidth of the output components is limited. Let denote a Lagrange multiplier, which is a function of t.

[0039] (3) Cross-iteration calculation After the transformation, the variational modes need to be optimized using alternating direction multipliers to find the optimal solution to the variational problem. In the formula, Wiener filters for each modal component; To update parameters, This represents the number of iterations in this round. They are respectively The corresponding frequency domain form; The frequency center of each mode, Represents angular frequency. This represents the current estimate of the center angular frequency. Indicates the summation index. This represents the Fourier transform of the Lagrange multipliers at the nth iteration. Indicates the first Modality Fourier transform, This indicates that the k-th mode has passed through the th... n+ 1 The Fourier transform estimate updated after the next iteration.

[0040] (4) SHO algorithm To address the issue of parameter uncertainty in VMD, an improved algorithm, SHO-VMD, is proposed, combining the Spotted Hyena Algorithm (SHO) and VMD. The Spotted Hyena Algorithm (SHO) is a swarm intelligence optimization algorithm that simulates the predation behavior of wild spotted hyena packs. The above formula represents the fitness of a feasible solution. Fitness serves as an evaluation metric for feasible solutions and is also the basis for updating the hyena's position. Represents the original signal to be decomposed. This represents the sub-signal that has been decomposed.

[0041] Based on the positions of the prey and individual hyenas, a formula for determining the hyena's location is derived. The distance between the hyena and its prey; t 1 This represents the current iteration number; P(t1) represents the prey's position; P(t1) represents the current position of the hyena at time t1; B is the swing factor, used to adjust the hyena's position; E is the convergence factor; h is a control parameter that decreases linearly from its maximum value to 0 as the number of iterations increases. A random number uniformly distributed in the interval [0,1]. This indicates the preset maximum number of iterations.

[0042] When the convergence factor When the prey is attacked, the spotted hyena will attack; otherwise, it will stay away from it. Thus, the optimal cluster is determined. The first optimal hyena position was defined; This represents the location of other hyenas; N is the number of hyenas; A cluster representing N optimal solutions; This represents the position vector of the prey at time t1. This indicates the position of the individual hyena after the update at the next time step t1+1.

[0043] The second part is as follows: (5) Nuclear Extreme Learning Machine Based on the loss function consisting of the training error term and the regularization term of the output layer weight norm, the output layer weights are analytically calculated using Moore-Penrose (MP) generalized inverse matrix theory. The vector representing its input model, This represents the training objective; ideally, the model's error is 0; the ideal output would be... Indicates the connection of the first i The weights from each hidden layer node to the output layer Indicates the first i Each hidden layer node is for the input The activation output, Indicates the first j The bias term of each hidden layer node. Indicates connecting the input layer to the first... j The weight vector of each hidden layer node; The above equation can be expressed in matrix form as follows: ,in The input layer vector matrix; These are the connection weights between the input and output layers; This is the output layer vector matrix; regularization coefficients are introduced. To enhance network stability, the output weights and network output are shown below. in, For matrix transpose, The regularization coefficient is . identity matrix It is a feature mapping function. The Radial Basis Function (RBF) is chosen as the mapping function to map each sample point to an infinite-dimensional feature space, thereby making the originally linearly inseparable data linearly separable. The expression of the Gaussian kernel is shown below. Represents the kernel matrix. This represents the element in the i-th row and j-th column of the kernel matrix. This represents the specific mathematical expression of the Gaussian kernel function; The network output at this point is: As can be seen from the above formula, the regularization coefficient... and the hyperparameters of the kernel function It has played a crucial role in the network; see also Figure 5 To find the optimal parameters, the Seagull Optimization Algorithm (SOA), which has strong local search capabilities, is adopted.

[0044] (6) Seagull Optimization Algorithm SOA The Seagull Optimization Algorithm simulates the migration and attack behaviors of seagulls, corresponding to the global search and local search of the optimization algorithm, respectively. The mathematical model for migration is as follows. This indicates a new location where there is no conflict with other seagulls; This represents the current location of the seagull; This is the current iteration number; This represents the movement behavior of seagulls in a given search space; The initial value of the control factor is a constant that determines the intensity of the global exploration in the early stages of the algorithm. This indicates the preset maximum number of iterations.

[0045] After avoiding collisions with other seagulls, the seagulls will move to the optimal position, which can be expressed by the following formula. B1 represents the direction of the optimal position; B1 is a random number responsible for balancing the global and local searches. It is a random number between [0, 1]; It is the best place for seagulls to migrate. Indicates the number of iterations t At that time, the historical best position vector found in the population.

[0046] After migrating, seagulls will launch spiral attacks on their prey. This attack behavior represents a local search within the model, involving spiral-shaped movements during the attack. The attack behavior can be represented by the following formula. in, It is the radius of each spiral; [0, 2] Random angles within a certain range; and It is the correlation coefficient of the spiral shape; This is the candidate optimal solution (optimal position) that the seagull considers for attack.

[0047] In the embodiments of this application, in step S130, the Bayes-Bi-LSTM-ECFRM model based on environmental variable comparison and error correction is used, and a Gaussian mixture model is introduced to perform probabilistic prediction of short-term wind speed of periodic thermal-driven wind based on wind speed classification.

[0048] Step S130 may include the following steps: Step S131 involves performing single-step predictions of the wind speed and environmental variables of the periodic thermally driven wind to obtain a single-step prediction sequence. For example, KELM can be used to perform fast single-step predictions of the wind speed and environmental variables of the periodic thermally driven wind. The environmental variables here include temperature, temperature fluctuation rate (or temperature change rate), and wind speed fluctuation rate (or wind speed change rate). The predicted samples in the single-step prediction sequence that satisfy the characteristics of the periodic thermally driven wind are used as the test set, and historical periodic thermally driven wind samples are used as the training set. For details on how to determine whether the predicted samples of the single-step prediction sequence satisfy the characteristics of the periodic thermally driven wind, please refer to the relevant content in step S110, which will not be repeated here.

[0049] Step S132: Use the Bayes optimizer to optimize the key parameters of the bidirectional long short-term neural network Bi-LSTM, and perform multi-step prediction based on the single-step results (single-step prediction sequence) to obtain the predicted wind speed.

[0050] Step S133: Construct a joint probabilistic model JPM using temperature, temperature fluctuation rate, wind speed fluctuation rate, and wind speed from the training set, and form a feature library. Then, map the predicted features using the joint probabilistic model JPM to obtain the feature set of the input error correction model. The predicted features here include predicted wind speed, predicted temperature, predicted wind speed fluctuation rate, and predicted temperature fluctuation rate.

[0051] Step S134: Features are extracted using a convolutional neural network (CNN), and the correction amount is output using a bidirectional long short-term neural network (Bi-LSTM) to correct the error in the predicted wind speed.

[0052] Step S135: Establish a Gaussian mixture model (GMM) based on wind speed levels by combining the error between the corrected wind speed and the measured wind speed with the historical prediction error. Use the expectation-maximization algorithm (EM) to estimate the mean / covariance / mixing ratio under each level, and output the prediction interval at a given confidence level.

[0053] Specifically, see Figure 6 The Bayes-Bi-LSTM-ECFRM model can include a wind speed prediction module based on a Bayesian optimizer, an error correction module based on a joint probability model, and an interval prediction module based on a Gaussian mixture model with wind speed classification.

[0054] (1) Wind speed prediction module based on Bayesian optimizer Bidirectional Long Short-Term Memory (Bi-LSTM) is an improved version of LSTM. Bi-LSTM consists of a forward LSTM and a backward LSTM. The Bi-LSTM model can capture temporal features from two different directions of information, improving upon the limitation of LSTM in learning temporal features from back to front.

[0055] To determine the undetermined parameters in the Bi-LSTM model, including the number of units, dropout rate, initial learning rate, and learning rate update period, a Bayesian optimization algorithm (Bayes for short) is used to optimize these parameters, as shown in the following formula. in It represents , indicating the observed set; x t Let it be the decision vector; Represents the observed value; Indicates the use of parameters x t The observed performance values ​​obtained after training and evaluating the model; Represents the error value; Represents the observed target data The possibility; represent f The prior distribution, represent f The posterior distribution of; Represents the parameters to be optimized. It is a normalization constant that ensures the sum of the probability distributions is 1.

[0056] (2) Error correction module based on joint probability model To discuss the actual distribution characteristics of wind speed, the Weibull extreme value distribution is used to fit the cumulative distribution function (CDF) curve of wind speed. Weibull extreme value distribution: in k 1 This represents the shape parameters. 1 represents the position parameter. This represents the scale parameter of the Weibull distribution.

[0057] When the Weibull distribution fits well, the regression coefficient Ri 2 The probability density function (PQ) reached 0.998, which can well describe the distribution characteristics of wind speed. By adjusting the parameters of the Weibull distribution, a joint probability model between wind and temperature was established, as shown in the following formula. A Convolutional Neural Network (CNN) is combined with a Bidirectional Long Short-Term Neural Network (Bi-LSTM) as an error correction module in the regression model. The CNN is used to extract features for the Bi-LSTM model to learn. After obtaining the features extracted by the CNN, they are input into the Bi-LSTM model for regression. The regression values ​​are used for error correction. The error between the corrected wind speed and the actual wind speed, along with historical prediction error values, are used to construct the corresponding Gaussian mixture model. An error correction module is constructed based on a joint probability model and a CNN-Bi-LSTM hybrid model.

[0058] Finally, different eigenvalues ​​were obtained through a joint probability model, including wind speed-temperature, wind speed-temperature volatility, and wind speed-wind speed volatility.

[0059] (3) Interval prediction module based on Gaussian mixture model with wind speed classification When using KELM to predict data sets, it was found that estimating the error using a single normal model overestimated the error range for low wind speeds and underestimated the error range for high wind speeds. In practical engineering, more attention is paid to the prediction results for high wind speeds. Therefore, Gaussian Mixture Models (GMMs) are constructed based on wind speed classifications to reflect the error values ​​at different wind speeds. GMM models are built upon the assumption that the population distribution includes... The probability model of individual distributions, the GMM model of wind speed-error, can estimate different prediction intervals based on the predicted wind speed value.

[0060] In the GMM model, different categories Different datasets correspond to different values ​​and mixing ratios; the Expectation Maximization Algorithm (EM) is used to estimate the mean and mixing ratio.

[0061] The adjusted wind speed classification is used to model a Gaussian mixture model, where the expression for the Gaussian mixture model is as follows: in, This represents the error value; This represents the predicted wind speed; The CDF curve represents the mixed Gaussian process under known wind speed conditions; Represents the mixing ratio; This represents the CDF curve of a two-dimensional Gaussian process under known wind speed conditions; n 1 This represents the number of parameters; Represents the covariance matrix; This represents the sample mean, including and , This indicates the number of Gaussian distributions, i.e., the number of preset subcategories.

[0062] In the embodiments of this application, in step S140, the maximum information coefficient (MIC) method is used for feature selection, and the model that preprocesses the error obtained from feature regression using the fully adaptive noise set empirical mode decomposition algorithm is used to predict short-term wind speed during strong winds and cooling.

[0063] Step S140 may include the following steps: Step S141: Perform wind-induced cooling time history identification, select wind-induced cooling samples that meet the characteristics of wind-induced cooling from the monitoring data, and obtain the corresponding wind speed time history and environmental variable sequence.

[0064] Step S142: The Kernel Extreme Learning Machine (KELM) is used to perform single-step prediction of the wind speed time history to obtain the predicted value. The error sequence is calculated from the measured value (corresponding measured wind speed) and the predicted value to establish a historical error database.

[0065] Step S143: Map the wind field parameters / environmental variables based on the Joint Probability Model (JPM) to obtain the mapped feature sequence. The wind field parameters include seven categories: temperature, rate of temperature change, rate of wind speed change, sine wind direction, cosine wind direction, sine angle of attack, and cosine angle of attack.

[0066] Step S144: Construct candidate feature sets of 7-dimensional measured features and 7-dimensional mapped features, and use the maximum information coefficient (MIC) method for feature selection to obtain the optimal feature subset and form a feature library.

[0067] Step S145: Using the feature library as input, construct a CNN-Bayes-Bi-LSTM error regression model, in which the Bayesian optimizer is used for parameter optimization, the convolutional neural network (CNN) is used for feature extraction, and the bidirectional long short-term neural network (Bi-LSTM) is used for time series modeling, outputting an error prediction sequence.

[0068] Step S146: The error prediction sequence is decomposed using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm to obtain multiple intrinsic mode function components (IMF components) and residues (RES). The intrinsic mode function is abbreviated as IMF.

[0069] Step S147: The intrinsic mode function components (IMF) and residual terms (RES) are predicted using kernel extreme learning machine (KELM), and the prediction results of each component are superimposed and reconstructed to obtain the prediction error value of the test set.

[0070] Step S148: Add the prediction error value back to the predicted value to obtain the final short-term wind speed prediction value for wind-induced cooling.

[0071] Specifically, see Figure 7 The model employs the Maximum Information Coefficient (MIC) method to select features from 7-dimensional measured features and 7-dimensional mapped features. The 7-dimensional feature set includes temperature, temperature change rate, wind speed change rate, wind direction sine, wind direction cosine, angle of attack sine, and angle of attack cosine. Furthermore, it utilizes fully adaptive noise set empirical mode decomposition to preprocess the errors obtained from feature regression. The model comprises three modules: an environmental variable feature mapping module, a feature selection module, and a mode decomposition-prediction module.

[0072] (1) Environmental variable feature mapping module Kernel Extreme Learning Machine (KELM) was used to make one-step predictions of historical wind cooling timelines, and the error values ​​were compared with historical measured values ​​to construct a historical error database. The existing wind field parameters were mapped using a Joint Probability Model (JPM) to obtain the mapping values ​​of wind speed for different environmental variables.

[0073] Regarding the process of mapping wind speed using environmental variables, the first step is to use an extreme value model to fit the unnormalized wind speed time series. The extreme value model uses the GEV distribution, and the formula is as follows. in and The parameters are to be determined. v The observed wind speed value, a This represents the lower limit of the possible values ​​for wind speed. b This represents the upper limit of the possible wind speed values.

[0074] (2) Feature selection module Feature mapping yields different features, but it also increases the dimensionality of the feature set, significantly increasing prediction time and compromising predictive timeliness. Therefore, it is necessary to reduce the dimensionality of the high-dimensional feature set. The MIC (Maximum Information Coefficient) method is used to evaluate the features suitable for regression, as shown in the following formula. in, a*b This represents the spatial division of intervals, and the grid division method is not certain. B As a variable, it restricts a*b The upper limit is taken as the power of 0.6 of the data volume. express X and Y The mutual information represents the closeness of the relationship between the two parties. The magnitude of the value is positively correlated with the degree of intimacy; Represents the feature to be measured. The target wind speed vector represents the regression; it is trained using a convolutional neural network (CNN) and a bidirectional long short-term neural network (Bi-LSTM), with parameters selected using the Bayes optimization algorithm. Its input is the selected features, and its output is the error value predicted by KELM.

[0075] (3) Modal decomposition-prediction module The error obtained from the regression of the above model is smaller than the wind speed value. Furthermore, using traditional Empirical Mode Decomposition (EMD) to decompose the error values ​​can lead to mode aliasing. Therefore, Fully Adaptive Noise Set Empirical Mode Decomposition (CEEMDAN) is used to preprocess the error obtained from the training set to improve prediction accuracy. The formula illustrates the process of adding noise to the original residual signal, where The error sequence to be decomposed is... These are the Gaussian white noise weighting coefficients. For the first The Gaussian white noise signal generated by the second processing. This represents the target signal obtained from the training set that needs to be decomposed.

[0076] in This represents the first modal component obtained from the CEEMDAN decomposition. Indicates the first i The first intrinsic mode function obtained after performing standard EMD decomposition on the sequence to be decomposed; This represents the residual signal after the first decomposition.

[0077] Based on the above description, the prediction method 100 according to the embodiments of this application can predict short-term wind speed for different types of mountain winds, such as periodic heat-driven winds, strong wind cooling, and non-special types of strong winds, by using a specially designed hybrid model SHO-VMD-SOA-KELM, or a Bayes-Bi-LSTM-ECFRM model based on environmental variable comparison and error correction, or a model that uses a fully adaptive noise ensemble empirical mode decomposition algorithm to preprocess the error obtained from feature regression. This can achieve accurate and rapid prediction of short-term wind speed.

[0078] refer to Figure 8 This application also provides a prediction apparatus 200 for implementing the prediction method 100 according to the embodiments of this application. The prediction apparatus 200 includes a processor 210 and a memory 220. The prediction apparatus 200 may include one or more processors 210 and one or more memories 220. The memory 220 stores an executable program that is run by the processor 210. When the executable program is run by the processor 210, it causes the processor 210 to perform the prediction method 100 described above according to the embodiments of this application.

[0079] The processor 210 may be a central processing unit (CPU) or other processing units with data processing capabilities and / or instruction execution capabilities.

[0080] The memory 220 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 210 may execute the program instructions to implement the client functions (implemented by the processor) in the embodiments of this application described herein, and / or other desired functions. Various applications and various data may also be stored in the computer-readable storage medium, such as various data used and / or generated by the applications.

[0081] The prediction device 200 may also include input and output devices, which are interconnected via a bus system and / or other forms of connection mechanisms. It should be noted that... Figure 8 The components and structure of the prediction device 200 shown are merely exemplary and not limiting; the prediction device 200 may also have other components and structures as needed.

[0082] The input device can be a device used by a user to input commands, and can include one or more of a keyboard, mouse, microphone, and touchscreen. Furthermore, the input device can also be any interface for receiving information.

[0083] The output device can output various information (e.g., images or sounds) to the outside (e.g., a user), and may include one or more of a display, speaker, etc. Furthermore, the output device can also be any other device with output functionality.

[0084] For example, the example prediction device 200 for implementing the prediction method 100 according to the embodiments of this application can be applied to terminal devices (such as mobile phones), tablet computers, laptop computers, ultra-mobile personal computers (UMPCs), handheld computers, netbooks, personal digital assistants (PDAs), wearable devices (such as smartwatches, smart glasses, or smart helmets), augmented reality (AR) devices, virtual reality (VR) devices, smart home devices, in-vehicle computers, and other electronic devices. The embodiments of this application do not impose any limitations on this.

[0085] Those skilled in the art can understand the specific operation of the prediction device 200 for implementing the prediction method 100 according to the embodiments of this application in conjunction with the content described above. For the sake of brevity, the specific details will not be repeated here, but only some main operations of the processor 210 will be described.

[0086] In one embodiment of this application, when the executable program is run by the processor 210, the processor 210 performs the following steps: Measured meteorological data of the target area are acquired, and mixed strong winds are classified based on the measured meteorological data to obtain periodic thermal-driven winds, strong winds with cooling, and non-special types of strong winds. A hybrid model SHO-VMD-SOA-KELM is constructed based on Variational Mode Decomposition (VMD) optimized by the Spotted Hyena Algorithm (SHO) and Kernel Extreme Learning Machine (KELM) optimized by the Seagull Algorithm (SOA) to predict short-term wind speeds of non-special types of strong winds. A Bayes-Bi-LSTM-ECFRM model is used based on environmental variable comparison and error correction, and a Gaussian mixture model is introduced to predict the short-term wind speed of periodic thermal-driven winds based on wind speed classification. A model using the Maximum Information Coefficient (MIC) method for feature selection and a fully adaptive noise ensemble empirical mode decomposition algorithm to preprocess the errors obtained from feature regression is used to predict the short-term wind speed of strong winds with cooling.

[0087] The above exemplarily illustrates a prediction method 100 according to an embodiment of this application. The following, in conjunction with... Figure 9 The following describes a prediction system 300 provided in another aspect of the embodiments of this application.

[0088] Reference Figure 9 This document describes an example prediction system 300 for implementing the prediction method of the embodiments of this application. The prediction system 300 may include a mixed strong wind classification module 310, a non-special type strong wind prediction module 320, a periodic heat-driven wind prediction module 330, and a strong wind cooling prediction module 340. Wherein: The mixed strong wind classification module 310 is used to: acquire measured meteorological data of the target area, perform mixed strong wind classification based on the measured meteorological data, and obtain periodic heat-driven wind, strong wind cooling and non-special type strong wind.

[0089] The non-special type gale prediction module 320 is used to predict short-term wind speeds for non-special type gales by constructing a hybrid model SHO-VMD-SOA-KELM based on variational mode decomposition (VMD) optimized by the spotted hyena algorithm (SHO) and kernel extreme learning machine (KELM) optimized by the seagull optimization algorithm (SOA).

[0090] The periodic thermal-driven wind prediction module 330 is used to: use the Bayes-Bi-LSTM-ECFRM model based on environmental variable comparison and error correction, and introduce a Gaussian mixture model to perform probabilistic prediction of short-term wind speed of periodic thermal-driven wind based on wind speed classification.

[0091] The strong wind and cooling prediction module 340 is used to: use a model that employs the maximum information coefficient (MIC) method for feature selection and a fully adaptive noise set empirical mode decomposition algorithm to preprocess the error obtained from feature regression, and then perform short-term wind speed prediction for strong wind and cooling.

[0092] The prediction system 300 proposed in this application embodiment can quickly and accurately predict the short-term wind speed of mixed strong winds in mountainous areas.

[0093] Furthermore, according to embodiments of this application, this application also provides a storage medium on which a computer program is stored. When the computer program is run by a processor, it is used to execute corresponding steps of the prediction method 100 of this application. The storage medium may, for example, include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0094] Furthermore, according to embodiments of this application, this application also provides a computer program product, including computer instructions that, when executed by a processor, implement the steps of the prediction method 100 of embodiments of this application.

[0095] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0096] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0098] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0099] It should be noted that the above embodiments are illustrative of this application and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0100] The above description is merely a specific embodiment or illustration of the embodiments of this application. The scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A short-time wind speed prediction method based on the classification of mixed strong winds in mountainous areas, characterized in that, The prediction method includes: Acquire measured meteorological data of the target area, and classify mixed strong winds based on the measured meteorological data to obtain periodic thermal-driven winds, strong wind cooling, and non-special types of strong winds; A hybrid model, SHO-VMD-SOA-KELM, is constructed based on Variational Mode Decomposition (VMD) optimized by the Spotted Hyena Algorithm (SHO) and Kernel Extreme Learning Machine (KELM) optimized by the Seagull Algorithm (SOA), for short-term wind speed prediction of non-special types of strong winds. The Bayes-Bi-LSTM-ECFRM model is based on environmental variable comparison and error correction, and a Gaussian mixture model is introduced to perform probabilistic prediction of short-term wind speed of periodic thermal-driven wind based on wind speed classification. A model using the Maximum Information Coefficient (MIC) method for feature selection and a fully adaptive noise ensemble empirical mode decomposition algorithm to preprocess the error obtained from feature regression is used for short-term wind speed prediction during strong winds and temperature drops.

2. The prediction method according to claim 1, characterized in that, Based on the measured meteorological data, a mixed strong wind classification was performed to obtain periodic thermal-driven winds, strong wind cooling, and non-special types of strong winds, specifically including: The daily maximum 10-minute average wind speed is calculated based on measured meteorological data. Measured meteorological data with daily maximum 10-minute average wind speeds that are greater than the set wind speed are selected as gale event data. Screening Pearson correlation coefficients 0.4 and p-value 1% of strong wind events are considered as periodic thermal-driven winds; Linear equations were performed using 10-minute average temperatures for the gale event data. Fit, when coefficients And the average daily temperature before the strong wind event Or the average daily temperature during a strong wind event. The corresponding strong wind event data serves as the basis for determining wind-induced temperature drop. Data on wind events excluding periodic heat-driven winds and wind-induced cooling events are considered non-special types of winds.

3. The prediction method according to claim 1, characterized in that, A hybrid model, SHO-VMD-SOA-KELM, built based on Variational Mode Decomposition (VMD) optimized by the Spotted Hyena Algorithm (SHO) and Kernel Extreme Learning Machine (KELM) optimized by the Seagull Algorithm (SOA), is used for short-term wind speed prediction of non-special types of strong winds. Specifically, it refers to: The Spotted Hyena Algorithm (SHO) is used to optimize the parameters of Variational Mode Decomposition (VMD) and select the optimal parameters to construct the SHO-VMD model. The SHO-VMD model was used to decompose wind speed data of non-special types of strong winds to obtain subsequences with different characteristic patterns. The Seagull Optimization Algorithm (SOA) is used to optimize the parameters of the Kernel Extreme Learning Machine (KELM) and select the optimal parameters to construct the SOA-KELM model. Each subsequence is input into the SOA-KELM model for prediction, and the prediction results of each subsequence are summed to obtain the final short-term wind speed prediction result.

4. The prediction method according to claim 3, characterized in that, The prediction method includes: VMD variational mode decomposition algorithm in, For the set of components of all K modes, To sample wind speed signals, The modal components obtained from the decomposition; K is the number of modes. for The corresponding frequency center; It follows the Dirac distribution; This is a convolution operation, where t is the sampling time. This indicates taking the partial derivative of the function. In mathematics, the imaginary unit *st* represents the introduction of constraints. After the decomposition is completed, the VMD decomposition results obtained above will be used to further refine the decomposition process. , Perform a Lagrange transformation. As a Lagrange operator, it has the significance of maintaining the strictness of the constraints; As a penalty factor, the bandwidth of the output components is limited. Let represent the Lagrange multipliers, which are functions of t; After the transformation, the variational modes need to be optimized using alternating direction multipliers to find the optimal solution to the variational problem. In the formula, Wiener filters for each modal component; To update parameters, This represents the number of iterations in this round. They are respectively The corresponding frequency domain form; The frequency center of each mode, Represents angular frequency. This represents the current estimate of the center angular frequency. Indicates the summation index. This represents the Fourier transform of the Lagrange multipliers at the nth iteration. Indicates the first Modality Fourier transform, This indicates that the k-th mode has passed through the th... n+ 1 The Fourier transform estimate updated after the next iteration; To address the issue of parameter uncertainty in VMD, an improved algorithm, SHO-VMD, is proposed, combining the Spotted Hyena Algorithm (SHO) and VMD. The Spotted Hyena Algorithm (SHO) is a swarm intelligence optimization algorithm that simulates the predation behavior of wild spotted hyena packs. The above formula represents the fitness of a feasible solution. Fitness serves as an evaluation metric for feasible solutions and is also the basis for updating the hyena's position. Represents the original signal to be decomposed. Indicates the sub-signal that has been decomposed; Based on the positions of the prey and individual hyenas, a formula for determining the hyena's location is derived. The distance between the hyena and its prey; t 1 This represents the current iteration number; P(t1) represents the prey's position; P(t1) represents the current position of the hyena at time t1; B is the swing factor, used to adjust the hyena's position; E is the convergence factor; h is a control parameter that decreases linearly from its maximum value to 0 as the number of iterations increases. A random number uniformly distributed in the interval [0,1]. Indicates the preset maximum number of iterations; When the convergence factor When the prey is attacked, the spotted hyena will attack; otherwise, it will stay away from it. Thus, the optimal cluster is determined. The first optimal hyena position was defined; This represents the location of other hyenas; N is the number of hyenas; A cluster representing N optimal solutions; This represents the position vector of the prey at time t1. This indicates the position of the individual hyena after the update at the next time step t1+1; Based on the loss function consisting of the training error term and the regularization term of the output layer weight norm, the output layer weights are analytically calculated using Moore-Penrose generalized inverse matrix theory. The vector representing its input model, This represents the training objective; ideally, the model's error is 0; the ideal output would be... Indicates the connection of the first i The weights from each hidden layer node to the output layer Indicates the first i Each hidden layer node is for the input The activation output, Indicates the first j The bias term of each hidden layer node. Indicates connecting the input layer to the first... j The weight vector of each hidden layer node; The above equation can be expressed in matrix form as follows: ,in The input layer vector matrix; These are the connection weights between the input and output layers; This is the output layer vector matrix; regularization coefficients are introduced. To enhance network stability, the output weights and network output are shown below. For matrix transpose, The regularization coefficient is . identity matrix It is a feature mapping function. The Gaussian kernel function (RBF) is chosen as the mapping function to map each sample point to an infinite-dimensional feature space, thereby making the originally linearly inseparable data linearly separable. The expression of the Gaussian kernel is shown below. Represents the kernel matrix. Represents the kernel matrix of the first... i line, number j Column elements, This represents the specific mathematical expression of the Gaussian kernel function; The network output at this point is: As can be seen from the above formula, the regularization coefficient... and the hyperparameters of the kernel function It plays a crucial role in the network; in order to solve for the optimal parameters, the Seagull Optimization Algorithm (SOA), which has strong local search capabilities, is adopted; The Seagull Optimization Algorithm simulates the migration and attack behaviors of seagulls, corresponding to the global search and local search of the optimization algorithm, respectively. The mathematical model for migration is as follows. This indicates a new location where there is no conflict with other seagulls; This represents the current location of the seagull; This is the current iteration number; This represents the movement behavior of seagulls in a given search space; The initial value of the control factor is a constant that determines the intensity of the global exploration in the early stages of the algorithm. Indicates the preset maximum number of iterations; After avoiding collisions with other seagulls, the seagulls will move to the optimal position, which can be expressed by the following formula. B1 represents the direction of the optimal position; B1 is a random number responsible for balancing the global and local searches. It is a random number between [0, 1]; It is the best place for seagulls to migrate. Indicates the number of iterations t At that time, the historical best position vector found in the population; After migrating, seagulls will launch spiral attacks on their prey. This attack behavior represents a local search within the model. During the attack, they move in a spiral shape, and their attack behavior can be represented by the following formula. in, It is the radius of each spiral; [0, 2] Random angles within a certain range; and It is the correlation coefficient of the spiral shape; This is the candidate optimal solution that the seagulls consider attacking.

5. The prediction method according to claim 1, characterized in that, The Bayes-Bi-LSTM-ECFRM model, based on environmental variable comparison and error correction, and incorporating a Gaussian mixture model, is used for probabilistic prediction of short-term wind speeds in periodic thermally driven winds based on wind speed classification. Specifically, this includes: A single-step prediction sequence is obtained by making single-step predictions of wind speed and environmental variables of periodic thermally driven wind. Predicted samples that meet the characteristics of periodic thermally driven wind in the single-step prediction sequence are used as the test set, and historical periodic thermally driven wind samples are used as the training set. The key parameters of the bidirectional long short-time neural network Bi-LSTM are optimized using the Bayes optimizer, and the predicted wind speed is obtained by multi-step prediction based on the single-step results. The joint probability model JPM is constructed using temperature, temperature fluctuation rate, wind speed fluctuation rate and wind speed in the training set and a feature library is formed. Then, the predicted features are mapped using the joint probability model JPM to obtain the feature set of the input error correction model. The predicted wind speed is corrected by using a convolutional neural network (CNN) to extract features and a bidirectional long short-term neural network (Bi-LSTM) to regress and output correction values. By combining the error between the corrected wind speed and the measured wind speed with the historical prediction error, a Gaussian mixture model (GMM) is established according to wind speed levels. The expectation-maximization algorithm (EM) is used to estimate the mean, covariance, and mixing ratio under each level, and the prediction interval at a given confidence level is output accordingly.

6. The prediction method according to claim 5, characterized in that, The Bayes-Bi-LSTM-ECFRM model includes a wind speed prediction module based on a Bayesian optimizer, an error correction module based on a joint probability model, and an interval prediction module based on a Gaussian mixture model with wind speed classification. The Bi-LSTM (Bi-LSTM) neural network consists of a forward LSTM and a backward LSTM. The Bi-LSTM model can capture temporal features from two different directions of information, which improves the ability of LSTM to learn temporal features from back to front. To determine the undetermined parameters in the Bi-LSTM model, including the number of units, dropout rate, initial learning rate, and learning rate update period, a Bayesian optimizer is used to optimize these parameters, as shown in the following formula. in It represents , indicating the observed set; x t Let it be the decision vector; Represents the observed value; Indicates the use of parameters x t The observed performance values ​​obtained after training and evaluating the model; Represents the error value; Represents the observed target data The possibility; represent f The prior distribution, represent f The posterior distribution of; Represents the parameters to be optimized. It is a normalization constant that ensures the sum of the probability distributions is 1; To discuss the actual distribution characteristics of wind speed, the Weibull extreme value distribution is used to fit the cumulative distribution function (CDF) curve of wind speed. Weibull extreme value distribution in k 1 This represents the shape parameters. 1 represents the position parameter. This represents the scale parameter of the Weibull distribution; When the regression coefficient Ri of the Weibull distribution represents the goodness of fit... 2 The probability density function (PQ) reached 0.998, which is sufficient to describe the distribution characteristics of wind speed. By adjusting the parameters of the Weibull distribution, a joint probability model between wind and temperature was established, as shown in the following formula. A convolutional neural network (CNN) and a bidirectional long short-term neural network (Bi-LSTM) are combined as an error correction module in a regression model. The CNN is used to extract features for the Bi-LSTM model to learn. After obtaining the features extracted by the CNN, they are input into the Bi-LSTM model for regression. The regression values ​​are used for error correction. The error between the corrected wind speed and the actual wind speed, along with the historical prediction error values, are used to construct the corresponding Gaussian mixture model. The error correction module is based on a joint probability model and a CNN-Bi-LSTM hybrid model. Different eigenvalues ​​were further obtained through the joint probability model, including wind speed-temperature, wind speed-temperature volatility, and wind speed-wind speed volatility. When using KELM to predict data sets, it was found that estimating the error using a single normal model overestimated the error range for low wind speeds and underestimated the error range for high wind speeds. In practical engineering, more attention is paid to the prediction results for high wind speeds. Therefore, a Gaussian Mixture Model (GMM) is constructed based on wind speed classification to reflect the error values ​​at different wind speeds. The GMM model is built upon the assumption that the population distribution includes... The probability model of individual distribution, the GMM model of wind speed-error estimates different prediction intervals based on the predicted wind speed value; In the GMM model, different categories Different datasets correspond to different values ​​and mixing ratios; the Expectation-Maximization (EM) algorithm is used to estimate the mean and mixing ratio. The adjusted wind speed classification is used to model a Gaussian mixture model, where the expression for the Gaussian mixture model is as follows: in, This represents the error value; This represents the predicted wind speed; The CDF curve represents the mixed Gaussian process under known wind speed conditions; Represents the mixing ratio; This represents the CDF curve of a two-dimensional Gaussian process under known wind speed conditions; n 1 This represents the number of parameters; Represents the covariance matrix; This represents the sample mean, including and , This indicates the number of Gaussian distributions, i.e., the number of preset subcategories.

7. The prediction method according to claim 1, characterized in that, A model employing the Maximum Information Coefficient (MIC) method for feature selection and using a fully adaptive noise ensemble empirical mode decomposition algorithm to preprocess the errors obtained from feature regression is used for short-term wind speed prediction during periods of strong wind and temperature drop. Specifically, this includes: The wind-induced cooling time history is identified, and wind-induced cooling samples that meet the characteristics of wind-induced cooling are selected from the monitoring data to obtain the corresponding wind speed time history and environmental variable sequence. The wind speed time history was predicted step by step using the KELM nuclear extreme learning machine to obtain the predicted value, and the error sequence was calculated from the measured value and the predicted value to establish a historical error database. Based on the joint probability model JPM, wind field parameters / environmental variables are mapped to obtain a mapping feature sequence; Construct candidate feature sets of 7-dimensional measured features and 7-dimensional mapped features, and use the maximum information coefficient (MIC) method for feature selection to obtain the optimal feature subset and form a feature library; Using the feature library as input, a CNN-Bayes-Bi-LSTM error regression model is constructed, in which the Bayesian optimizer is used for parameter optimization, the convolutional neural network (CNN) is used for feature extraction, and the bidirectional long short-term neural network (Bi-LSTM) is used for time series modeling, outputting an error prediction sequence. The error prediction sequence is decomposed by the fully adaptive noise set empirical mode decomposition algorithm CEEMDAN to obtain multiple intrinsic mode function components (IMF) and residual terms (RES). Kernel Limit Learning Machine (KELM) is used to predict each intrinsic mode function (IMF) component and residual term (RES) separately. The prediction results of each component are then superimposed and reconstructed to obtain the prediction error value of the test set. The prediction error value is added back to the predicted value to obtain the final short-term wind speed prediction value for the wind-induced cooling.

8. The prediction method according to claim 7, characterized in that, The 7-dimensional feature set includes temperature, temperature change rate, wind speed change rate, wind direction sine, wind direction cosine, angle of attack sine, and angle of attack cosine. The model, which uses fully adaptive noise set empirical mode decomposition to preprocess the error obtained from feature regression, has three modules: environmental variable feature mapping module, feature selection module, and mode decomposition-prediction module. Kernel Extreme Learning Machine (KELM) is used to predict historical wind cooling timelines, and the error values ​​are compared with historical measured values ​​to construct a historical error database. The Joint Probability Model (JPM) is used to map existing wind field parameters to obtain the mapping values ​​of wind speed for different environmental variables. Regarding the process of mapping wind speed using environmental variables, the first step is to fit the unnormalized wind speed time series using an extreme value model. The extreme value model employs the GEV distribution, and the formula is as follows. in and The parameters are to be determined. v The observed wind speed value, a This represents the lower limit of the possible values ​​for wind speed. b This represents the upper limit of the possible values ​​for wind speed; Feature mapping yields different features, but it also increases the dimensionality of the feature set, leading to increased prediction time and a loss of predictive timeliness. Therefore, it is necessary to reduce the dimensionality of the high-dimensional feature set. The MIC (Maximum Information Coefficient) method is used to evaluate the features available for regression, as shown in the following formula. in, a*b This represents the spatial division of intervals, and the grid division method is not certain. B As a variable, it restricts a*b The upper limit is taken as the power of 0.6 of the data volume. express X and Y The mutual information represents the closeness of the relationship between the two parties. The magnitude of the value is positively correlated with the degree of intimacy; Represents the feature to be measured. The target wind speed vector represents the regression; it is trained using a convolutional neural network (CNN) and a bidirectional long short-term neural network (Bi-LSTM), with parameters selected using the Bayes optimization algorithm. Its input is the selected features, and its output is the error value predicted by KELM. The error obtained from the regression of the above model is smaller than the wind speed value. Furthermore, using traditional Empirical Mode Decomposition (EMD) to decompose the error values ​​can lead to mode aliasing. Therefore, Fully Adaptive Noise Set Empirical Mode Decomposition (CEEMDAN) is used to preprocess the error obtained from the training set to improve prediction accuracy. The formula illustrates the process of adding noise to the original residual signal, where Let be the error sequence to be decomposed. These are the Gaussian white noise weighting coefficients. For the first The Gaussian white noise signal generated by the second processing. This represents the target signal that needs to be decomposed, obtained from the training set. in This represents the first modal component obtained from the CEEMDAN decomposition. Indicates the first i The first intrinsic mode function obtained after performing standard EMD decomposition on the sequence to be decomposed; This represents the residual signal after the first decomposition.

9. A short-time wind speed prediction device based on the classification of mixed strong winds in mountainous areas, characterized in that, The prediction device includes: Memory is used to store executable instructions for a computer; A processor, when executing computer-executable instructions stored in the memory, implements the prediction method of any one of claims 1 to 8.

10. A storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the prediction method according to any one of claims 1 to 8.

Images