Improved ensemble empirical mode decomposition and LSTM model series connection wind speed prediction method

Abstract

The invention discloses a wind speed prediction method based on improved ensemble empirical mode decomposition and LSTM model series connection, and the method comprises the steps: adding grid point historical meteorological data into adaptive noise through singular spectrum analysis, carrying out the ensemble empirical mode decomposition, and carrying out the processing of an intrinsic mode function component and a residual error obtained through the decomposition; performing standardization processing on the topographic data, and extracting topographic spatial features through different spatial feature extraction schemes; and constructing the fused features into time sequence format data, inputting the time sequence format data into an LSTM model for prediction, obtaining a predicted intrinsic mode function component and a residual error, adding the intrinsic mode function component and the residual error to generate a short-time wind speed prediction result, comparing the short-time wind speed prediction result with the real-time wind speed data to calculate a model loss function, and training the LSTM model by adopting back propagation iteration to obtain a short-time wind speed prediction result. And performing rolling prediction to obtain a future wind speed prediction result. The prediction result of the method is highly consistent with the observation value in numerical value, and the method also shows high accuracy in trend prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of wind speed prediction, and in particular to a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series. Background Art

[0002] With the frequent occurrence of global climate change and extreme weather events, accurate wind speed forecasting is of great significance in many fields such as energy, agriculture, and shipping. Wind speed forecasting can not only provide reliable data support for wind power generation, but also provide a scientific basis for disaster prevention and mitigation. However, wind speed has a high degree of spatiotemporal complexity and nonlinear characteristics. Traditional forecasting methods are difficult to effectively capture these characteristics, which affects the prediction accuracy.

[0003] At present, wind speed prediction methods are mainly divided into two categories: physical models and data-driven models. The physical model is based on the principles of atmospheric dynamics and thermodynamics and uses numerical weather forecast models for prediction, but its computational cost is high, it is sensitive to initial conditions, and its prediction accuracy is limited. The data-driven model analyzes historical wind speed data and related meteorological variables and uses time series analysis, neural networks, support vector machines and other methods for prediction. Among them, the long short-term memory network (LSTM), as a special recursive neural network, is widely used in the field of wind speed prediction because of its advantages in capturing long-term dependencies. However, the traditional LSTM model still has shortcomings when dealing with wind speed sequences with highly non-stationary and multi-scale characteristics.

[0004] Empirical mode decomposition (EMD) is an adaptive decomposition method for dealing with nonlinear and non-stationary time series, which can decompose complex time series into several intrinsic mode functions (IMFs). Complete ensemble empirical mode decomposition (CEEMDAN) further improves EMD and overcomes the problem of modal aliasing by introducing white noise. However, adding adaptive noise based on the spectrum characteristics after introducing singular spectrum analysis SSA is more in line with the spectrum characteristics and complexity of wind speed series than adding white noise. Therefore, the method combining SSA-Enhanced CEEMDAN decomposition and LSTM model can better deal with the non-stationary characteristics of wind speed series and improve prediction accuracy and stability. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a wind speed prediction method and system that improves the series connection of ensemble empirical mode decomposition and LSTM model to solve the problem that the existing wind speed prediction model has unsatisfactory effect.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a wind speed prediction method of improving the series connection of ensemble empirical mode decomposition and LSTM model, comprising: obtaining grid historical meteorological data and terrain data of a target area to be predicted within a set time window; performing ensemble empirical mode decomposition on the grid historical meteorological data after adding adaptive noise through singular spectrum analysis, and normalizing the intrinsic mode function components and residuals obtained by the decomposition; standardizing the terrain data, and extracting terrain spatial features through different spatial feature extraction schemes; fusing the normalized intrinsic mode function components, residuals and terrain spatial features, constructing the fused features into time series format data and inputting them into the LSTM model for prediction, so as to obtain predicted intrinsic mode function components and residuals; generating a short-term wind speed prediction result by adding the predicted intrinsic mode function components and residuals, and comparing the short-term wind speed prediction result with the actual wind speed data to calculate the model loss function, iteratively training the LSTM model using back propagation, and performing rolling prediction to obtain future wind speed prediction results.

[0009] As a preferred solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series according to the present invention, wherein: the acquisition of grid historical meteorological data and terrain data of the target area to be predicted within the set time window includes:

[0010] The grid historical meteorological data of the target area to be predicted is obtained by superimposing the grid historical surface pressure data, the grid historical wind speed data and the grid historical temperature data according to the time step, wherein the grid historical wind speed data includes u-direction wind speed data and v-direction wind speed data;

[0011] The terrain data is obtained by superimposing historical altitude data, historical slope data and historical slope aspect data with a 1km resolution of the current regional grid points within a set time window according to the time step.

[0012] As a preferred solution of the wind speed prediction method of improving ensemble empirical mode decomposition and LSTM model in series according to the present invention, the step of ensemble empirical mode decomposition includes:

[0013] The grid historical meteorological data of the target area to be predicted is standardized and then embedded into a high-dimensional space to form a trajectory matrix, the trajectory matrix is ​​subjected to singular value decomposition, the first k largest singular values ​​and their corresponding left singular vectors and right singular vectors are selected, and the trend component, periodic component and noise component are reconstructed;

[0014] Performing fast Fourier transform analysis on the trend component, periodic component and noise component to obtain the spectrum characteristics of the components, and identifying the main frequency components, dynamically adjusting the intensity and frequency of the noise according to the spectrum characteristics, using frequency domain information to generate Gaussian white noise with frequency characteristics similar to those of the original data, and adding the generated adaptive noise to the original time series data to form a noise enhancement sequence;

[0015] Find all the maximum and minimum points in the noise enhancement sequence, fit the maximum and minimum points respectively by interpolation to obtain an upper envelope and a lower envelope, calculate the mean of the upper envelope and the lower envelope as the local mean of the current sequence, and subtract the local mean from the noise enhancement sequence to obtain a new sequence until the new sequence meets the intrinsic mode function conditions, that is, zero mean and monotonicity;

[0016] The sequence that meets the eigenmode function conditions is taken as an eigenmode function component and separated from the original sequence. The above process is repeated until all the eigenmode function components in the original sequence are separated. Each eigenmode function component and residual obtained by decomposition are normalized and used as the input of the subsequent wind speed prediction model.

[0017] As a preferred solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series according to the present invention, wherein: the extraction of terrain spatial features by different spatial feature extraction schemes includes:

[0018] The first spatial feature extraction scheme is to use a 3x3 convolution kernel to perform convolution processing on the terrain data, with a step size of 2 and a padding of 1, and downsample the feature map after the convolution processing through a 2x2 maximum pooling layer, and then use a 5x5 convolution kernel to perform a convolution operation on the downsampled feature map, with a step size of 2 and a padding of 2, to expand the feature receptive field, and downsample the feature map again through a 2x2 maximum pooling layer, and finally use a 7x7 convolution kernel to perform convolution processing, with a step size of 2 and a padding of 3; the convolved feature map is converted into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM1;

[0019] The second spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2 and a padding of 1, downsample the feature map after the convolution through a 2x2 maximum pooling layer, and then use a 5x5 convolution kernel to convolve the downsampled feature map, with a step size of 2 and a padding of 2; convert the convolved feature map into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM2;

[0020] The third spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2 and a padding of 1, and convert the convolved feature map into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM3.

[0021] As a preferred solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series according to the present invention, the fusion of the normalized intrinsic mode function components, residuals and the terrain spatial features includes:

[0022] The normalized intrinsic mode function components, residuals and the terrain spatial features are respectively used to construct feature matrices and concatenated into a unified feature matrix X = [X IMF ,X res ,X DEM1 ,X DEM2 ,X DEM3 ], where X IMF represents the characteristic matrix of the normalized intrinsic mode function components, X res Represents the characteristic matrix of the normalized residual, X DEM1 , X DEM2 , X DEM3 They represent the feature matrices of terrain spatial features extracted by different spatial feature extraction schemes;

[0023] Perform a linear transformation on the concatenated features to generate queries, keys, and values, calculate the dot product of the query and key, divide by the scaling factor to obtain an unnormalized attention score matrix, normalize the unnormalized attention score matrix using a Softmax function to obtain attention weights, and use the attention weights to perform a weighted summation on the values ​​to obtain a self-attention output matrix;

[0024] The self-attention output matrix is ​​fused and concatenated with the original features to obtain enhanced fused features.

[0025] As a preferred solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series according to the present invention, wherein: the step of constructing the fused features into time series format data and inputting it into the LSTM model for prediction includes:

[0026] The fused features are constructed into time series format data, and the rolling window technology is used to process the time series format data. The time window size is set to 7. In each iteration, the LSTM model receives the time format data of the previous 6 hours, and predicts the intrinsic mode function components and residuals of the 7th hour through forward propagation.

[0027] As a preferred solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series according to the present invention, the acquisition of the future wind speed prediction result includes:

[0028] Generate a short-term wind speed prediction result by adding the predicted intrinsic mode function component and the residual, compare the short-term wind speed prediction result with the actual wind speed data, and calculate the mean square error between the predicted value and the true value;

[0029] The gradient of the weights of each layer of the LSTM model is calculated by back propagation according to the calculated mean square error, and the gradient is propagated layer by layer through the network layer to update the weight parameters of each layer. During the rolling training process, the parameters of the LSTM model are continuously updated using a sliding time window. Each iteration uses the data of the past 6 hours for training to predict the wind speed in the next hour;

[0030] The new predicted value is compared with the true value to calculate the loss, and the model weights are updated through backpropagation. The future wind speed forecast results are obtained through rolling prediction.

[0031] In a second aspect, the present invention provides a wind speed prediction system of improving ensemble empirical mode decomposition and LSTM model in series, comprising:

[0032] A data acquisition unit, used to obtain grid historical meteorological data and terrain data of a target area to be predicted within a set time window;

[0033] An ensemble empirical mode decomposition unit is used to perform ensemble empirical mode decomposition on the grid point historical meteorological data after adding adaptive noise through singular spectrum analysis, and normalize the intrinsic mode function components and residuals obtained by the decomposition;

[0034] A terrain spatial feature extraction unit, used for standardizing the terrain data and extracting terrain spatial features through different spatial feature extraction schemes;

[0035] A feature fusion and prediction unit, used to fuse the normalized intrinsic mode function components, residuals and the terrain spatial features, construct the fused features into time series format data and input them into the LSTM model for prediction, so as to obtain predicted intrinsic mode function components and residuals;

[0036] The prediction result acquisition unit is used to generate a short-term wind speed prediction result by adding the predicted intrinsic mode function component and the residual, and compare the short-term wind speed prediction result with the actual wind speed data to calculate the model loss function, use back propagation to iteratively train the LSTM model, and obtain the future wind speed prediction result by rolling prediction.

[0037] In a third aspect, the present invention provides a computing device, comprising:

[0038] Memory and processor;

[0039] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method are implemented.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the method when executed by a processor.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] ① The present invention considers the actual influence of multiple factors in the process of wind speed prediction, introduces terrain data and grid meteorological data (wind speed, temperature, air pressure, etc.), and uses the improved CEEMDAN decomposition module to finely decompose these data, so as to better extract the characteristics of each factor, and finally improve the accuracy of prediction;

[0043] ② The present invention adopts a multi-stage decomposition and normalization processing strategy. The original time series data is first decomposed by SSA, and then combined with adaptive noise generation and EMD decomposition to make the obtained intrinsic mode function components more accurate and stable. Through this improved CEEMDAN decomposition method, the noise in the data is effectively reduced, and the sensitivity of the prediction model to different frequency components is improved;

[0044] ③ The present invention uses three different convolutional pooling schemes to process terrain data, extracts terrain features of different scales (DEM1, DEM2, DEM3), and splices these features into a unified feature tensor, which is input into the model, so that the model can understand and utilize terrain information more comprehensively, thereby improving the accuracy of wind speed prediction;

[0045] ④ The present invention integrates the features of terrain data and meteorological data through the attention mechanism, fully utilizes the spatial and temporal characteristics of terrain and meteorological elements, and achieves more accurate wind speed prediction. The final model shows good generalization ability and stability in practical applications, which significantly improves the reliability and practicality of wind speed prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0047] Figure 1 This is an overall flow chart of a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0048] Figure 2 An LSTM model structure diagram of a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0049] Figure 3 A schematic flow chart of a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0050] Figure 4 An improved CEEMDAN module structure diagram of a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0051] Figure 5 A structural diagram of a terrain DEM spatial feature extraction module of a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0052] Figure 6 A diagram showing the fusion process of DEM features, IMF and residual features of a wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0053] Figure 7 A comparison chart of the wind speed predicted at 1:00 on January 1, 2023 and the actual wind speed according to the wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0054] Figure 8 A comparison chart of the wind speed predicted at 2:00 on January 1, 2023 and the actual wind speed according to the wind speed prediction method combining improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention;

[0055] Fig. 9 A comparison chart of the forecast results for the next 72 hours in the plateau area of ​​the wind speed forecasting method combining the improved ensemble empirical mode decomposition and the LSTM model in series according to an embodiment of the present invention and the ERA5 product;

[0056] Fig.10 A comparison chart of the forecast results of the future 72 hours in a hilly area of ​​the wind speed forecasting method combining the improved ensemble empirical mode decomposition and the LSTM model in series according to an embodiment of the present invention and the ERA5 product;

[0057] Fig.11This is a comparison chart of the forecast results for the next 72 hours in the basin area of ​​the wind speed forecasting method combining the improved ensemble empirical mode decomposition and LSTM model in series according to an embodiment of the present invention and the ERA5 product. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0059] Example 1

[0060] Reference Figure 1 to Figure 6 As an embodiment of the present invention, a wind speed prediction method of improving ensemble empirical mode decomposition and LSTM model in series is provided. The input data uses ERA5 grid data of horizontal wind speed, vertical wind speed, surface air pressure, temperature and other elements in the past 4 years in the study area. The time window is set to 7. Each iteration uses the historical wind speed data of the previous 6 hours to predict the wind speed of the 7th hour. Then, the loss is calculated with the ERA5 real-time data to back-propagate and update the model weight. After the model training is completed, the future wind speed prediction is performed. Specifically, Figure 3 The following steps are shown:

[0061] S100: Acquire historical grid meteorological data and terrain data of a target area to be predicted within a set time window;

[0062] In the embodiment of the present application, the grid historical meteorological data of the target area to be predicted is obtained by superimposing the grid historical surface pressure data, the grid historical wind speed data and the grid historical temperature data according to the time step, wherein the grid historical wind speed data includes u-direction wind speed data and v-direction wind speed data;

[0063] In an embodiment of the present application, the terrain data is obtained by superimposing the historical altitude data, historical slope data and historical slope aspect data of the current regional grid with a resolution of 1 km within a set time window according to the time step.

[0064] S200: The grid historical meteorological data is subjected to ensemble empirical mode decomposition after adding adaptive noise through singular spectrum analysis, and the intrinsic mode function components and residuals obtained by decomposition are normalized. The detailed improved CEEMDAN module structure is shown in Figure 4 As shown;

[0065] In the embodiment of the present application, the grid historical meteorological data of the target area to be predicted is standardized and then embedded in a high-dimensional space to form a trajectory matrix. The trajectory matrix is ​​subjected to singular value decomposition, and the formula is X=U∑VT, where U is a left singular vector matrix, ∑ is a singular value diagonal matrix, and V is a right singular vector matrix. The singular values ​​are sorted by size, and the first k largest singular values ​​and their corresponding left singular vectors and right singular vectors are selected according to the contribution rate of the eigenvalues, and reconstructed to obtain trend components (low frequency), periodic components (medium frequency) and noise components (high frequency). Each component represents structural changes on different time scales, such as long-term trends, seasonal changes and high-frequency noise. The trend component selects the component corresponding to the largest singular value, the periodic component selects the component corresponding to the singular value in the middle range, and the noise component selects the component corresponding to the smaller singular value. These components can be expressed as:

[0066] X trend =U1Σ1V1 T ,X periodic =U2Σ2V2 T ,X noise =U3Σ3V3 T

[0067] Furthermore, the frequency spectrum characteristics of the trend component, periodic component and noise component are analyzed by fast Fourier transform, and the FFT transformation results are as follows:

[0068] F k =FFT(X k )

[0069] Among them, F k Is ingredient X k Spectrum diagram of

[0070] Furthermore, the main frequency components are identified, the intensity and frequency of the noise are dynamically adjusted according to the spectrum characteristics, and the frequency domain information is used to generate Gaussian white noise with similar frequency characteristics to the original data. The generated adaptive noise is added to the original time series data to form a noise enhancement sequence, which is expressed as:

[0071] X enhanced =X+αN

[0072] Furthermore, all the maximum and minimum points in the noise-enhanced sequence are found, and the upper and lower envelopes are obtained by fitting the maximum and minimum points respectively through interpolation. Then, the mean of the upper and lower envelopes is calculated as the local mean of the current sequence, and the noise-enhanced sequence is subtracted from the local mean to obtain a new sequence, until the new sequence meets the intrinsic mode function conditions, that is, zero mean and monotonicity;

[0073] Furthermore, the sequence that satisfies the intrinsic mode function condition is taken as an intrinsic mode function component and separated from the original sequence. The above process is repeated until all the intrinsic mode function components in the original sequence are separated. The EMD process can be expressed as:

[0074]

[0075] Among them, IMFi represents the i-th intrinsic mode function, and residual represents the residual;

[0076] Furthermore, each intrinsic mode function component and residual obtained by decomposition is normalized to ensure that the intrinsic mode functions (IMFs) and residuals of different frequencies have the same scale in the subsequent model processing to avoid unstable model training and inaccurate prediction due to dimensional differences; the original IMF component is set to IMF i , the residual is residual, the normalized IMF component and residual are expressed as IMFi' and residual' respectively, and the normalized formula is as follows:

[0077]

[0078] Among them, min(IMFi) and max(IMFi) represent the minimum and maximum values ​​of IMFi', respectively, and min(residual) and max(residual) represent the minimum and maximum values ​​of the residual, respectively.

[0079] S300: standardizing the terrain data and extracting terrain spatial features through different spatial feature extraction schemes;

[0080] It should be noted that in this embodiment, the SRTMDEMUTM 90M resolution digital elevation data product of the study area (103°36' to 109°35' east longitude, 24°37' to 29°13' north latitude) is obtained from the geospatial data cloud through advanced search selection and resampled to 1 km resolution through ArcGIS tools. The size of the DEM data after resampling is (1,600,600), and the DEM spatial features are extracted by applying three different convolution pooling schemes, where the formula for the convolution operation is:

[0081]

[0082] Where X is the input data matrix with a size of H×W, W is the convolution kernel with a size of M×N, b is the bias term, i, j are the position indexes of the output matrix Y, and Y is the output matrix with a size of H'×W'. H' and W' can be calculated by the following formula:

[0083]

[0084] Further, such as Figure 5 As shown, different spatial feature extraction schemes are configured as follows:

[0085] The first spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2, a padding of 1, an input size of (1, 600, 600), and an output size of (64, 300, 300). The convolution-processed feature map is downsampled through a 2x2 maximum pooling layer, and the output size is (64, 150, 150). Then, a 5x5 convolution kernel is used to convolve the downsampled feature map, with a step size of 2 and a padding of 2. The field of view is expanded, and the output size is (64, 75, 75). The feature map is downsampled again through a 2x2 maximum pooling layer, and the output size is (64, 37, 37). Finally, a 7x7 convolution kernel is used for convolution processing, with a step size of 2 and a padding of 3, and the output size is (64, 18, 18). The feature map after convolution is converted into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM1, and the final output size is (64, 24, 28).

[0086] The second spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2, a padding of 1, an input size of (1, 600, 600), and an output size of (32, 300, 300). The convolution-processed feature map is downsampled through a 2x2 maximum pooling layer, and the output size is (32, 150, 150). Then, a 5x5 convolution kernel is used to convolve the downsampled feature map, with a step size of 2, a padding of 2, and an output size of (32, 75, 75). The convolution-processed feature map is converted into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM2, and the final output size is (32, 24, 28).

[0087] The third spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2, a padding of 1, an input size of (1, 600, 600), and an output size of (16, 300, 300). The feature map after convolution is converted into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM3, and the final output size is (16, 24, 28);

[0088] It should be noted that through the three different convolution pooling configurations of the terrain DEM spatial feature extraction module, terrain features of different scales can be extracted respectively, and the depth and details of feature extraction can be flexibly adjusted. These features will be used as input for feature fusion together with the IMF components and residuals for the subsequent wind speed prediction model.

[0089] S400: fusing the normalized intrinsic mode function components, residuals and terrain spatial features to obtain fused features;

[0090] In the embodiments of the present application, Figure 6 As shown in the figure, the normalized intrinsic mode function components, residuals and terrain spatial features are respectively used to construct feature matrices and spliced ​​into a unified feature matrix X = [X IMF ,X res ,X DEM1 ,X DEM2 ,X DEM3 ], where X IMF represents the characteristic matrix of the normalized intrinsic mode function components, X res Represents the characteristic matrix of the normalized residual, X DEM1 , X DEM2 , X DEM3 They represent the feature matrices of terrain spatial features extracted by different spatial feature extraction schemes;

[0091] In the embodiment of the present application, the concatenated features are linearly transformed to generate the query matrix Q, the key matrix K and the value matrix V. The calculation formula is as follows:

[0092] Q=W Q X+b Q ,K=W K X+b K ,V=W V X+b V

[0093] Among them, W Q , W K , W V is the weight matrix of linear transformation, b Q , b K , b V is the bias term, X is the input feature;

[0094] Furthermore, the dot product of the query and the key is calculated and divided by the scaling factor to obtain the unnormalized attention score matrix. The unnormalized attention score matrix is ​​normalized by the Softmax function to obtain the attention weights. The values ​​are weighted and summed using the attention weights to obtain the self-attention output matrix, which is expressed as:

[0095]

[0096] in, is the scaling factor, taking the dimensions of query and key;

[0097] Furthermore, the self-attention output matrix is ​​fused and concatenated with the original features to obtain the enhanced fused features Z = AV, which are input into the subsequent LSTM model.

[0098] S500: constructing the fused features into time series format data and inputting them into the LSTM model for prediction to obtain the predicted intrinsic mode function components and residuals;

[0099] In the embodiment of the present application, the fused features are constructed into time series format data, and the rolling window technology is used to process the time series format data. The time window size is set to 7. In each iteration, the LSTM model receives the time format data of the previous 6 hours, and predicts the intrinsic mode function components and residuals of the 7th hour through forward propagation;

[0100] Further, such as Figure 2 As shown, the composition of the LSTM model is:

[0101] ①LSTM1 uses 128 units, each unit contains 128 hidden units, and the activation function uses tanh to calculate new candidate memory states. The regression activation function uses σ. The specific formula is:

[0102] i t =σ(W i ·[h t-1 , x t ]+b i )

[0103]

[0104] o t =σ(W o ·[h t-1 , x t ]+b o )

[0105] h t =o t ⊙tanh(C t )

[0106] Where σ is the sigmoid function, W f , W i , W C , W O are the weight matrices of the forget gate, input gate, candidate memory state, and output gate, respectively, and b f , b i , b C , b o is the corresponding bias term, f t is the activation value of the forget gate, which is used to control the degree of forgetting of the memory state at the previous moment. tis the activation value of the input gate, which is used to control the update degree of the memory state by the current input. t is the memory state at the current moment, which is obtained by weighted combination of the memory state at the previous moment and the candidate memory state. t is the activation value of the output gate, which is used to control the influence of the memory state on the current hidden state. t It is the hidden state at the current moment, obtained by weighted combination of memory state and output gate.

[0107] ②LSTM2 uses the output of the first layer as input to further extract time series features. This layer has 128 units, each of which manages the flow of information through a complex gating mechanism (forget gate, input gate, output gate). The tanh activation function is used to capture the nonlinear features in the data and maintain the stability of the gradient. The regression activation function sigmoid is used to control whether the information of each time step should be passed to the next layer, ensuring that the model can retain key time series information and provide rich feature representation for subsequent predictions.

[0108] ③LSTM3 receives the output of the second layer and further refines the feature representation. By reducing the number of units to 64, the feature dimension is gradually compressed. A more compact and efficient feature representation is learned from a large amount of data to reduce the risk of overfitting. Tanh is also used as the activation function to ensure the ability of nonlinear processing, while the sigmoid regression activation function ensures the continuity of sequence information, so that the prediction of each time step is based on complete context information, which is conducive to capturing the long-term dependency of the time series.

[0109] ④LSTM4 uses 32 units to further deepen the advanced feature extraction of time series data. More abstract and key information is extracted from the existing feature representation. The application of the tanh activation function once again emphasizes the model's ability to handle nonlinear data, while the sigmoid regression activation function ensures the effective transmission of information, allowing the model to accurately capture dynamic changes in the data. Further extract advanced features.

[0110] ⑤LSTM5 reduces the number of units to 8, focusing on the most important predictive factors. The tanh activation function extracts the final prediction value from the highly abstract features, while the sigmoid regression activation function is used in the output layer, which is responsible for converting the output of the LSTM unit into the final prediction probability or value, especially for the prediction of the last time step, to ensure the accuracy and reliability of the model output. Through this layer, the model can generate accurate predictions of each intrinsic mode function (IMF) and residual within the next hour.

[0111] S600: generating a short-term wind speed prediction result by adding the predicted intrinsic mode function component and the residual;

[0112] In the embodiment of the present application, the output of the LSTM model is obtained, including each intrinsic mode function (IMF) and residual value for the next hour, and the IMF and residual are added to obtain a complete short-term wind speed prediction result, and the formula is:

[0113]

[0114] S700: Compare the short-term wind speed forecast result with the actual wind speed data to calculate the model loss function, use back propagation to iteratively train the LSTM model, and perform rolling forecasting to obtain the future wind speed forecast result;

[0115] In the embodiment of the present application, the short-term wind speed prediction result is compared with the actual wind speed data, and the mean square error between the predicted value and the true value is calculated, and the formula is:

[0116]

[0117] Furthermore, the gradient of the weights of each layer of the LSTM model is calculated by back propagation based on the calculated mean square error, and the gradient is propagated through the network layer by layer to update the weight parameters of each layer. The update formula is:

[0118]

[0119] Among them, θ represents the weight parameter of the model, and η represents the learning rate;

[0120] Furthermore, in the rolling training process, the parameters of the LSTM model are continuously updated using a sliding time window. Each iteration uses the data from the past 6 hours for training, predicts the wind speed in the next hour, and superimposes the meteorological data at every 6 time points. For example, the three meteorological variables from 1 to 6 o'clock are superimposed. The dimension of each time point is (3, 24, 28), and the dimension of the 6 time points is (18, 24, 28). The ERA5 reanalysis dataset is used as the grid point real data and the predicted data to calculate the loss for training the model.

[0121] Furthermore, the new predicted value is compared with the true value to calculate the loss, and the model weights are updated through backpropagation, and the future wind speed forecast results are obtained through rolling prediction.

[0122] From the above, it can be seen that ① the present invention takes into account the actual influence of multiple factors in the process of wind speed prediction, introduces terrain data and grid meteorological data (wind speed, temperature, air pressure, etc.), and uses the improved CEEMDAN decomposition module to finely decompose these data, so as to better extract the characteristics of each factor, and finally improve the accuracy of prediction; ② the present invention adopts a multi-stage decomposition and normalization processing strategy, first decomposes the original time series data by SSA, and then combines adaptive noise generation and EMD decomposition to make the obtained intrinsic mode function components more accurate and stable. Through this improved CEEMDAN decomposition method, the noise in the data is effectively reduced, and the prediction model is improved. Sensitivity of different frequency components; ③ The present invention uses three different convolutional pooling schemes to process terrain data, respectively extracts terrain features of different scales (DEM1, DEM2, DEM3), and splices these features into a unified feature tensor, which is input into the model, so that the model can understand and utilize terrain information more comprehensively, thereby improving the accuracy of wind speed prediction; ④ The present invention fuses the features of terrain data and meteorological data through the attention mechanism, fully utilizes the spatial and temporal characteristics of terrain and meteorological elements, and realizes more accurate wind speed prediction. The final model shows good generalization ability and stability in practical applications, which significantly improves the reliability and practicability of wind speed prediction.

[0123] The above is a schematic scheme of a wind speed prediction method of an improved ensemble empirical mode decomposition and LSTM model in series in this embodiment. It should be noted that the technical solution of the wind speed prediction system of the improved ensemble empirical mode decomposition and LSTM model in series and the technical solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series belong to the same concept, and the details of the technical solution of the wind speed prediction system of the improved ensemble empirical mode decomposition and LSTM model in series in this embodiment that are not described in detail can all be referred to the description of the technical solution of the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series.

[0124] This embodiment also provides a wind speed prediction system that improves the series connection of ensemble empirical mode decomposition and LSTM model, including:

[0125] A data acquisition unit, used to obtain grid historical meteorological data and terrain data of a target area to be predicted within a set time window;

[0126] The ensemble empirical mode decomposition unit is used to perform ensemble empirical mode decomposition after adding adaptive noise to the grid historical meteorological data through singular spectrum analysis, and normalize the intrinsic mode function components and residuals obtained by the decomposition;

[0127] A terrain spatial feature extraction unit is used to perform standardization processing on terrain data and extract terrain spatial features through different spatial feature extraction schemes;

[0128] The feature fusion and prediction unit is used to fuse the normalized intrinsic mode function components, residuals and terrain spatial features, construct the fused features into time series format data and input them into the LSTM model for prediction, so as to obtain the predicted intrinsic mode function components and residuals;

[0129] The prediction result acquisition unit is used to generate a short-term wind speed prediction result by adding the predicted intrinsic mode function component and the residual, and compare the short-term wind speed prediction result with the actual wind speed data to calculate the model loss function, use back propagation to iteratively train the LSTM model, and obtain the future wind speed prediction result by rolling prediction.

[0130] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0131] This embodiment also provides a computing device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a wind speed prediction method of an improved set empirical mode decomposition and a LSTM model in series is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the housing of the computer device, or an external keyboard, touch pad or mouse, etc.

[0132] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0133] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0134] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0135] Example 2

[0136] Reference Figure 7 to Figure 11 Based on the previous embodiment, this embodiment provides an application example of a wind speed prediction method that improves the ensemble empirical mode decomposition and LSTM model in series, in order to verify and illustrate the technical effects used in this method.

[0137] In this embodiment, the range of 103°36' to 109°35' east longitude and 24°37' to 29°13' north latitude is selected as the research area. The area includes the entire Guizhou Province, and the data includes 24 latitude grid points and 28 longitude grid points. The grid historical meteorological data of the target area to be predicted from 2019 to 2022 is obtained from the ERA5 daily meteorological observation data, with a time step of 1 hour. The meteorological elements in the historical meteorological data include 2m temperature, surface air pressure, u wind speed, and v wind speed.

[0138] Furthermore, in order to demonstrate the beneficial effects of the present invention and verify the good effect of SSA-Enhanced CEEMDAN decomposition in wind speed prediction, an ablation experiment was carried out. The ensemble empirical mode decomposition (CEEMDAN) module, SSA-Enhanced CEEMDAN decomposition module, and SSA-Enhanced-CEEMDAN-DEM module were successively added to the wind speed prediction model to carry out ablation experiments, and the changes in the two evaluation indicators of root mean square error (RMSE) and correlation coefficient (CC) over 4 years were compared. The results are shown in Table 1.

[0139] Table 1: Ablation experiment results;

[0140]

[0141] In a specific embodiment of the present invention, the method of the present invention is used to predict the wind speed in Guizhou Province in the next 72 hours, as shown in the attached figure. Figure 7, Attachment Figure 8 As shown in the figure, it can be observed that the wind speed of the existing ERA5 reanalysis data is highly close to the wind speed prediction value of the improved CEEMDAN decomposition and LSTM model. This shows that the wind speed prediction method of the present invention can better fit the ERA5 actual wind speed in Guizhou Province, and shows an obvious similar trend in spatiotemporal distribution. These observations further verify the advantages of the wind speed prediction method of the present invention in wind speed forecasting and provide support for its reliability in practical applications.

[0142] In addition, in order to verify the feedback of SSA-Enhanced CEEMDAN decomposition and LSTM model on the terrain in predicting the wind speed in the next 72 hours, different terrain areas in the study area are analyzed. The terrain is high in the west and low in the east. The plateaus, hills and basins in the study area are analyzed as representatives. The prediction effects of 10m wind speed in three different terrain areas are as follows: Fig. 9 , 10 and Fig.11 The detailed results are shown in Table 2.

[0143] Table 2: MSE and CC of 10m wind speed in three different terrain areas;

[0144]

[0145] From the above, it can be seen that the method of the present invention has made many innovations in model construction and data processing: ① Combining SSA and CEEMDAN decomposition, the decomposition quality of meteorological data is effectively improved; ② Comprehensively considering terrain and meteorological elements, the input features of the prediction model are comprehensively enhanced; ③ The attention mechanism and LSTM network are used to improve the accuracy of wind speed prediction.

[0146] Compared with the existing technology, the present invention significantly improves the accuracy and reliability of wind speed prediction through comprehensive, multi-level data processing and innovative modeling methods, and has broad application prospects and practical value.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0148] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0149] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0150] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0151] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0152] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0153] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A wind speed prediction method based on improved ensemble empirical mode decomposition and LSTM model in series, characterized in that: include: Obtain historical grid meteorological data and terrain data of the target area to be predicted within a set time window; The grid historical meteorological data is subjected to ensemble empirical mode decomposition after adding adaptive noise through singular spectrum analysis, and the intrinsic mode function components and residuals obtained by the decomposition are normalized; The terrain data is standardized and the terrain spatial features are extracted through different spatial feature extraction schemes; The normalized intrinsic mode function components, residuals and the terrain spatial features are fused, and the fused features are constructed into time series format data and input into the LSTM model for prediction to obtain predicted intrinsic mode function components and residuals; A short-term wind speed prediction result is generated by adding the predicted intrinsic mode function component and the residual, and the short-term wind speed prediction result is compared with the actual wind speed data to calculate the model loss function. The LSTM model is iteratively trained using back propagation, and the future wind speed prediction result is obtained by rolling prediction.

2. The wind speed prediction method of improved ensemble empirical mode decomposition and LSTM model series connection as claimed in claim 1, characterized in that: The acquisition of grid historical meteorological data and terrain data of the target area to be predicted within the set time window includes: The grid historical meteorological data of the target area to be predicted is obtained by superimposing the grid historical surface pressure data, the grid historical wind speed data and the grid historical temperature data according to the time step, wherein the grid historical wind speed data includes u-direction wind speed data and v-direction wind speed data; The terrain data is obtained by superimposing historical altitude data, historical slope data and historical slope aspect data with a 1km resolution of the current regional grid points within a set time window according to the time step.

3. The wind speed prediction method of improved ensemble empirical mode decomposition and LSTM model series connection as claimed in claim 2, characterized in that: The steps of the ensemble empirical mode decomposition include: The grid historical meteorological data of the target area to be predicted is standardized and then embedded into a high-dimensional space to form a trajectory matrix, the trajectory matrix is ​​subjected to singular value decomposition, the first k largest singular values ​​and their corresponding left singular vectors and right singular vectors are selected, and the trend component, periodic component and noise component are reconstructed; Performing fast Fourier transform analysis on the trend component, periodic component and noise component to obtain the spectrum characteristics of the components, and identifying the main frequency components, dynamically adjusting the intensity and frequency of the noise according to the spectrum characteristics, using frequency domain information to generate Gaussian white noise with frequency characteristics similar to those of the original data, and adding the generated adaptive noise to the original time series data to form a noise enhancement sequence; Find all the maximum and minimum points in the noise enhancement sequence, fit the maximum and minimum points respectively by interpolation to obtain an upper envelope and a lower envelope, calculate the mean of the upper envelope and the lower envelope as the local mean of the current sequence, and subtract the local mean from the noise enhancement sequence to obtain a new sequence until the new sequence meets the intrinsic mode function conditions, that is, zero mean and monotonicity; The sequence that meets the eigenmode function conditions is taken as an eigenmode function component and separated from the original sequence. The above process is repeated until all the eigenmode function components in the original sequence are separated. Each eigenmode function component and residual obtained by decomposition are normalized and used as the input of the subsequent wind speed prediction model.

4. The wind speed prediction method of improved ensemble empirical mode decomposition and LSTM model series connection as claimed in claim 2, characterized in that: The extraction of terrain spatial features by different spatial feature extraction schemes includes: The first spatial feature extraction scheme is to use a 3x3 convolution kernel to perform convolution processing on the terrain data, with a step size of 2 and a padding of 1, and downsample the feature map after the convolution processing through a 2x2 maximum pooling layer, and then use a 5x5 convolution kernel to perform a convolution operation on the downsampled feature map, with a step size of 2 and a padding of 2, to expand the feature receptive field, and downsample the feature map again through a 2x2 maximum pooling layer, and finally use a 7x7 convolution kernel to perform convolution processing, with a step size of 2 and a padding of 3; the convolved feature map is converted into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM1; The second spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2 and a padding of 1, downsample the feature map after the convolution through a 2x2 maximum pooling layer, and then use a 5x5 convolution kernel to convolve the downsampled feature map, with a step size of 2 and a padding of 2; convert the convolved feature map into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM2; The third spatial feature extraction scheme is to use a 3x3 convolution kernel to convolve the terrain data, with a step size of 2 and a padding of 1, and convert the convolved feature map into a fixed-length feature vector through an adaptive average pooling layer to obtain feature DEM3.

5. The wind speed prediction method of improved ensemble empirical mode decomposition and LSTM model in series as claimed in claim 3 or 4, characterized in that: The step of fusing the normalized intrinsic mode function components, the residuals and the terrain spatial features comprises: The normalized intrinsic mode function components, residuals and the terrain spatial features are respectively used to construct feature matrices and concatenated into a unified feature matrix X = [X IMF ,X res ,X DEM1 ,X DEM2 ,X DEM3 ], where X IMF represents the characteristic matrix of the normalized intrinsic mode function components, X res Represents the characteristic matrix of the normalized residual, X DEM1 , X DEM2 , X DEM3 They represent the feature matrices of terrain spatial features extracted by different spatial feature extraction schemes; Perform a linear transformation on the concatenated features to generate queries, keys, and values, calculate the dot product of the query and key, divide by the scaling factor to obtain an unnormalized attention score matrix, normalize the unnormalized attention score matrix using a Softmax function to obtain attention weights, and use the attention weights to perform a weighted summation on the values ​​to obtain a self-attention output matrix; The self-attention output matrix is ​​fused and concatenated with the original features to obtain enhanced fused features.

6. The wind speed prediction method of improved ensemble empirical mode decomposition and LSTM model series connection as claimed in claim 5, characterized in that: The step of constructing the fused features into time series format data and inputting it into the LSTM model for prediction includes: The fused features are constructed into time series format data, and the rolling window technology is used to process the time series format data. The time window size is set to 7. In each iteration, the LSTM model receives the time format data of the previous 6 hours, and predicts the intrinsic mode function components and residuals of the 7th hour through forward propagation.

7. The wind speed prediction method of improved ensemble empirical mode decomposition and LSTM model series connection as claimed in claim 6, characterized in that: The acquisition of the future wind speed prediction result includes: Generate a short-term wind speed prediction result by adding the predicted intrinsic mode function component and the residual, compare the short-term wind speed prediction result with the actual wind speed data, and calculate the mean square error between the predicted value and the true value; The gradient of the weights of each layer of the LSTM model is calculated by back propagation according to the calculated mean square error, and the gradient is propagated layer by layer through the network layer to update the weight parameters of each layer. During the rolling training process, the parameters of the LSTM model are continuously updated using a sliding time window. Each iteration uses the data of the past 6 hours for training to predict the wind speed in the next hour; The new predicted value is compared with the true value to calculate the loss, and the model weights are updated through backpropagation. The future wind speed forecast results are obtained through rolling prediction.

8. A system using the wind speed prediction method of the improved ensemble empirical mode decomposition and LSTM model in series as claimed in any one of claims 1 to 7, characterized in that: include: A data acquisition unit, used to obtain grid historical meteorological data and terrain data of a target area to be predicted within a set time window; An ensemble empirical mode decomposition unit is used to perform ensemble empirical mode decomposition on the grid point historical meteorological data after adding adaptive noise through singular spectrum analysis, and normalize the intrinsic mode function components and residuals obtained by the decomposition; A terrain spatial feature extraction unit, used for standardizing the terrain data and extracting terrain spatial features through different spatial feature extraction schemes; A feature fusion and prediction unit, used to fuse the normalized intrinsic mode function components, residuals and the terrain spatial features, construct the fused features into time series format data and input them into the LSTM model for prediction, so as to obtain predicted intrinsic mode function components and residuals; The prediction result acquisition unit is used to generate a short-term wind speed prediction result by adding the predicted intrinsic mode function component and the residual, and compare the short-term wind speed prediction result with the actual wind speed data to calculate the model loss function, use back propagation to iteratively train the LSTM model, and obtain the future wind speed prediction result by rolling prediction.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having computer program instructions stored thereon, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images