Typhoon weather forecasting method, medium and equipment

Abstract

The invention discloses a typhoon weather forecast method, medium and equipment, and relates to the technical field of typhoon weather forecast, and the method comprises the steps: obtaining typhoon data of time sequences at different time intervals; constructing a Typhoon path prediction model based on Transform, and inputting the Typhoon data of the time sequences at different time intervals into the Typhoon path prediction model to obtain a Typhoon path; typhoon longitude and dimension data on the predicted typhoon path and air pressure at the time corresponding to the predicted typhoon path are obtained, and the typhoon moving distance is calculated through the typhoon longitude and dimension data on the predicted typhoon path; and establishing a geographically weighted regression model, and inputting the typhoon moving distance and the air pressure at the time corresponding to the predicted typhoon path into the geographically weighted regression model to obtain the typhoon wind speed. The invention provides a more accurate method for forecasting the typhoon weather.

Description

Technical Field

[0001] The present invention relates to the technical field of typhoon weather forecasting, and in particular to a typhoon weather forecasting method, medium and equipment. Background Art

[0002] Typhoon, a powerful and deep "tropical weather system", can bring abundant fresh water resources, promote the improvement of fresh water supply, and transport heat. At the same time, it is also a very destructive natural disaster. Its occurrence is often accompanied by strong winds, heavy rains, storm surges and mudslides, causing serious economic losses and casualties. Accurately predicting the path of typhoons can not only reduce the direct impact of typhoon disasters, but also predict extreme weather such as heavy rains and strong winds, which is particularly important for deploying disaster prevention measures in advance.

[0003] Patent application CN202410781088.9 discloses a method for determining the position of a typhoon eye based on the tangent point trajectory of GNSS occultation data. First, the typhoon forecast information is read and the occultation profile data near the typhoon is screened; secondly, the typhoon eye is located in combination with the real-time meteorological data to determine whether the occultation profile passes through the typhoon eye; then, the occultation profile three-dimensional path is quadratically fitted to locate the typhoon eye with higher accuracy. However, this method is only suitable for determining the position of the typhoon eye, and does not study the specific path movement of the typhoon.

[0004] Patent application CN202110941926.0 discloses a method for predicting typhoon paths using a full-sequence LSTM network model. First, the data is preprocessed to convert the time series data of Erwin's longitude and latitude into a time series in units of tropical cyclones. Then, four sequence prediction models with different time steps are established. Different hidden layer nodes are set for each model. The model is optimized based on comparative analysis of the typhoon prediction effects under different parameters to make the prediction more accurate. However, the input parameters of the model only involve the characteristic parameters of the typhoon, namely the longitude and latitude and intensity of the typhoon, without considering the influence of external meteorological factors such as air pressure and temperature.

[0005] Patent application CN202410069593.0 discloses a typhoon wind speed prediction method and system based on the Deep-Pred framework. First, the environmental field of typhoon wind speed influencing factors is constructed, and the data set is normalized and processed by sliding window sliding to train the model. The model based on the Deep-Pred framework can more accurately extract the spatiotemporal characteristics of multidimensional typhoon information, and effectively predict the changing trend of typhoon intensity and maximum wind speed.

[0006] Patent application CN202210476184.3 discloses a typhoon wind speed intelligent prediction method based on the FEN-ConvLSTM model. First, a feature extraction network FEN is built using convolution and fusion, and the FEN network is used to extract the feature information of typhoon wind speed influencing factors, which is then integrated with the typhoon wind speed data into a new data set. Finally, a FEN-ConvLSTM model is built for training and testing, and the model can be used to predict wind speed.

[0007] In 2024, Yin Wenjing established a typhoon path prediction model based on recurrent neural networks. Using the processed typhoon path data set as samples, three recurrent neural networks were trained and the typhoon center position in the next 6, 12, and 24 hours was predicted. Experiments show that when the prediction time is 24 hours, the LSTM model has higher prediction accuracy and the error is much smaller than the GRU model and RNN model. In 2019, Xu Gaoyang established a typhoon path prediction model based on the gated unit network (GRU). Using the longitude and latitude and intensity information of the typhoon, the RNN, LSTM and GRU models were used to predict the typhoon's position information in the next 6 hours. Experiments show that GRU has the smallest mean absolute error on the test set and can effectively improve the accuracy of path prediction. In 2019, Chen Longbiao established a model based on path similarity to predict the typhoon path. Based on the key point similarity model, the typhoon path prediction can be realized by comprehensively solving the three indicators of the cumulative shortest distance between the current typhoon record point and the historical typhoon path point, the similarity of movement speed and the similarity of movement direction.

[0008] In 2023, Lan Bin established a typhoon wind speed prediction model based on deep learning. He obtained more accurate atmospheric temperature and humidity data through the ATMS microwave detector, and then preprocessed the satellite microwave data and extracted typhoon wind speed related features, and used the CNN neural network to build a wind speed estimation model. In 2023, Zhang Chunwei proposed a typhoon wind speed characteristics research model based on small and medium-scale models and neural networks. Experiments showed that the WRF-LES small and medium-scale model can effectively simulate the landing typhoon wind field. The extreme wind speed changes of the landing typhoon are divided into "single peak" and "double peak" categories. The BP neural network model is more effective in predicting wind speed. In 2019, Wang Meng constructed a maximum wind speed extreme prediction model for typhoons at landing based on fuzzy time series. The model takes into account the uncertainty and ambiguity of factors such as climate and regional environment, and is more objective and scientific when predicting wind speed, but the model can only make medium- and short-term predictions, and cannot make long-term predictions. In 2019, Li Chunxiang proposed a typhoon wind speed prediction model based on different signal decomposition methods and the least squares support vector machine (LSSVM) optimized by the propagation swarm algorithm (PSO). By comparison, the prediction accuracy of the VMD-PSO-LSSVM model is higher, but the signal decomposition method VMD involves more parameters, and the parameters still need to be adjusted for different signals, which is more complicated.

[0009] The above methods contribute to the prediction of typhoon paths and typhoon wind speeds, but these methods have certain defects. For typhoon path prediction, most of the focus is on predicting the location of the typhoon center, and the specific movement path of the typhoon has not yet been studied. Moreover, the influencing factors considered by the model are mainly based on the characteristic parameters of the typhoon itself, without considering the influence of external factors, such as meteorological factors - temperature, air pressure, etc. For the prediction of typhoon center wind force, neural network models are mostly used, which can effectively capture the spatiotemporal characteristics of the data, but the relationship between the typhoon center wind force and the typhoon movement distance has not yet been studied, and the wind speed is predicted through this relationship model. The typhoon center wind force is closely related to the typhoon movement distance, so it is also very important to construct their relationship model. Summary of the invention

[0010] The purpose of the present invention is to solve the problem that the typhoon meteorological forecast, the typhoon path forecast, does not involve the specific moving path of the typhoon, and the wind speed forecast does not consider external factors, and proposes a typhoon meteorological forecast method, comprising the following steps:

[0011] S1. Obtaining typhoon data of time series at different time intervals, the typhoon data including: typhoon longitude, latitude, typhoon level, wind speed, air pressure and temperature data;

[0012] S2. Build a typhoon path prediction model based on Transformer, input typhoon data of time series at different time intervals into the typhoon path prediction model, and obtain the typhoon path;

[0013] S3. Obtain the typhoon longitude and latitude data on the predicted typhoon path and the air pressure at the corresponding time of the predicted typhoon path, and use the typhoon longitude and latitude data on the predicted typhoon path to calculate the typhoon moving distance; establish a geographically weighted regression model, input the typhoon moving distance and the air pressure at the corresponding time of the predicted typhoon path into the geographically weighted regression model to obtain the typhoon wind speed.

[0014] Furthermore, the sliding window method is used to convert the obtained time series typhoon data into a data set suitable for supervised learning. The length of the time series typhoon data is L, the length of the sliding window is a, and the sliding window moving step is 1, so L-a+1 data sequences are obtained. The data of the first a-1 moments of each data sequence are used as the input of the typhoon path prediction model, and the data of the a-th moment is used as the label. The supervised typhoon path prediction model predicts the offset of longitude and latitude at the a-th moment.

[0015] Furthermore, the typhoon's longitude and latitude data are used to calculate the typhoon's moving distance:

[0016]

[0017] Among them, d represents the typhoon moving distance, r represents the radius of the earth, (lon1, lat1) and (lon2, lat2) represent the longitude and latitude of two points respectively.

[0018] Furthermore, the Transformer's multi-head attention mechanism is expressed as:

[0019] MultiHeadAttn(Q,K,V)=Connect(head1,…head i ,…head H )W o

[0020] head i =Attention(Q i ,K i ,V i )

[0021]

[0022] in, denote query, key, and value vectors respectively, Q i ,K i ,V i denotes the i-th query, key, and value vectors, respectively. N and M denote the lengths of Q, K, and V, respectively. D k and D v Represents the data dimensions of K and V respectively, head i represents the i-th attention head, H represents the number of attention heads, and W o represents the weight, Represents the scaling factor.

[0023] Furthermore, the expression of the Transformer's feedforward neural network is:

[0024] FFN(H ′ )=ReLU(H ′ W 1 +b 1 )W 2 +b 2

[0025] Among them, FFN() represents the output of the feedforward neural network, H′ is the output of the previous layer, ReLU() represents the ReLU activation function, and W 1 and W 2 represents the weight, b 1 and b 2 Indicates offset top.

[0026] Furthermore, the expression of the Add&Normalize layer of Transformer is:

[0027] H=Norm(FFN(H ′ )+H ′ )

[0028] H ′ =Norm(MultiHeadAttn(X)+X)

[0029] Among them, Norm() represents the normalization function, MultiHeadAttn() represents the output of the multi-head attention mechanism, and X represents the input of the multi-head attention mechanism.

[0030] Furthermore, the geographically weighted regression model is expressed as follows:

[0031]

[0032] Among them, y i represents the wind speed value at the i-th sampling point, β0(u i ,v i ) represents the intercept term, p represents the total number of explanatory variables, (u i ,v i ) represents the coordinates of the i-th sampling point, β k (u i ,v i ) represents the regression analysis coefficient, x ik represents the kth explanatory variable of the ith sampling point, ε i is the regression residual.

[0033] Furthermore, the regression analysis coefficient of the geographically weighted regression model is estimated using the weighted least squares method, and the specific calculation formula is as follows:

[0034]

[0035] in, represents the estimate of the regression analysis coefficient, W(u i ,v i ) is the spatial weight diagonal matrix, X is the design matrix of the independent variables, and Y is the matrix of the dependent variable wind speed;

[0036] The specific expression of the spatial weight diagonal matrix is ​​as follows:

[0037] Where W ij is the weight from the jth sample point to the ith point to be estimated, d ij is the Euclidean distance between the i-th point to be estimated and the j-th sample point, and h is the bandwidth.

[0038] The present invention also proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned typhoon weather forecasting method is implemented.

[0039] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned typhoon weather forecasting method.

[0040] The beneficial effects brought by the technical solution provided by the present invention are:

[0041] The present invention obtains the time series data of typhoons. A Transformer model is constructed to predict the typhoon path based on the time series data of typhoons, and the specific movement path of the typhoon can be obtained; considering the spatial autocorrelation characteristics of the data, a geographically weighted regression model is constructed based on the typhoon movement distance and air pressure to predict the typhoon wind speed, which can effectively capture the temporal and spatial characteristics of the data. The present invention provides a more accurate method for typhoon weather forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of a typhoon weather forecasting method according to an embodiment of the present invention;

[0043] Figure 2 It is a block diagram of an electronic device in an exemplary embodiment of the present invention.

[0044] Figure 3 It is a schematic diagram of the iterative process of training a typhoon path prediction model based on Transformer in an embodiment of the present invention;

[0045] Figure 4 It is a path matching graph of Typhoon Bebejia verified by using the Transformer-based typhoon path prediction model DTW in an embodiment of the present invention;

[0046] Figure 5 This is the LSTM DTW verification path matching graph of Typhoon Bebejia;

[0047] Figure 6 It is a schematic diagram of the path of Typhoon Bebejia predicted by the typhoon path prediction model based on Transformer in an embodiment of the present invention;

[0048] Figure 7 This is a schematic diagram of the path of Typhoon Bebejia predicted by LSTM;

[0049] Figure 8 It is the actual wind speed of Typhoon Bebejia (2024.09.10~2024.09.19);

[0050] Fig. 9 It is the wind speed of Typhoon Bebejia predicted by the method of the embodiment of the present invention (2024.09.10~2024.09.19). DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0052] Embodiment 1: A flow chart of a typhoon weather forecasting method according to an embodiment of the present invention is as follows Figure 1 , specifically including the following steps:

[0053] S1. Obtain typhoon data of time series at different time intervals, the typhoon data including: typhoon longitude, latitude, typhoon level, wind speed, air pressure and temperature data.

[0054] The above data were normalized using the maximization-minimization normalization method.

[0055]

[0056] Among them, x′ represents the standardized data, x represents the above-mentioned typhoon data, min(x) represents the minimum value of the data x, and max(x) represents the maximum value of the data x.

[0057] In order to enable the model to understand the time series rules of the data, according to the characteristics of time series data, the sliding window method is used to convert the obtained time series typhoon data into a data set suitable for supervised learning. The length of the time series typhoon data is L, the length of the sliding window is a, and the sliding window moving step is 1. L-a+1 data sequences are obtained. The data of the first a-1 moments of each data sequence are used as the input of the typhoon path prediction model, and the data at the ath moment is used as the label. The supervised typhoon path prediction model predicts the offset of longitude and latitude at the ath moment.

[0058] In the embodiment of the present invention, the sliding window size is set to 11, that is, the time series data of the next 1 step is predicted by using the time series of the past 10 steps. The time step of the sliding window movement is 1, and another sample data can be obtained by moving the sliding window backward. Among them, the input feature data are longitude, latitude, typhoon level, wind speed, air pressure and temperature, and the output feature data are the offset of the longitude at the next moment and the offset of the latitude at the next moment.

[0059] S2. Construct a typhoon path prediction model based on Transformer, which is mainly divided into two parts: encoder and decoder. The encoder includes absolute position encoding, multi-head self-attention mechanism and feedforward network. The decoder includes two linear layers to fuse the output results of the encoder. Input the typhoon data of time series with different time intervals into the typhoon path prediction model to obtain the typhoon path.

[0060] The measurement interval of the original typhoon data is 1 hour. The time series with an interval of 1 hour is obtained from it to predict the offset of the typhoon longitude and latitude in the next hour; the time series data sampled every 2 hours in the original typhoon data is used to predict the offset of the typhoon longitude and latitude 2 hours later; the time series data sampled every n hours in the original typhoon data is used to predict the offset of the typhoon longitude and latitude n hours later. The predicted offsets of the typhoon longitude and latitude in 1, 2, ..., n hours are connected to obtain the typhoon path.

[0061] The absolute position encoding of Transformer is expressed as:

[0062]

[0063] Among them, PE(t) i represents the encoding of the i-th position, ω i represents the frequency of the i-th position, t represents the index of the absolute position, i%2=1 represents that i is an odd number, and i%2=0 represents that i is an even number.

[0064] The multi-head self-attention mechanism is to extract the attention weights in different modes. The multi-head attention mechanism of Transformer is expressed as:

[0065] MultiHeadAttn(Q,K,V)=Connect(head1,…head i ,…head H )W o

[0066] head i =Attention(Q i ,K i ,V i )

[0067]

[0068] in, denote query, key, and value vectors respectively, Q i ,K i ,V i denotes the i-th query, key, and value vectors, respectively. N and M denote the lengths of Q, K, and V, respectively. Dk and D v Represents the data dimensions of K and V respectively, head i represents the i-th attention head, H represents the number of attention heads, and W o represents the weight, Represents the scaling factor.

[0069] The feedforward neural network is a fully connected network. The expression of the Transformer feedforward neural network is:

[0070] FFN(H ′ )=ReLU(H ′ W 1 +b 1 )W 2 +b 2

[0071] Among them, FFN() represents the output of the feedforward neural network, H′ is the output of the previous layer, ReLU() represents the ReLU activation function, and W 1 and W 2 represents the weight, b 1 and b 2 Indicates offset top.

[0072] Using residual blocks to connect the multi-head self-attention mechanism and the feedforward neural network, the Add&Normalize layer expression of Transformer is:

[0073] H=Norm(FFN(H ′ )+H ′ )

[0074] H ′ =Norm(MultiHeadAttn(X)+X)

[0075] Among them, Norm() represents the normalization function, MultiHeadAttn() represents the output of the multi-head attention mechanism, and X represents the input of the multi-head attention mechanism.

[0076] S3. Obtain the longitude and latitude data of the typhoon on the predicted typhoon path, as well as the air pressure at the corresponding time of the predicted typhoon path. For example, the air pressure at the location of the typhoon path 2 hours later can be predicted using the air pressure predicted by the local meteorological bureau.

[0077] The longitude and latitude data of the typhoon on the predicted typhoon path are used to calculate the typhoon moving distance. The calculation formula is:

[0078]

[0079] Among them, d represents the typhoon moving distance, r represents the radius of the earth, (lon1, lat1) and (lon2, lat2) represent the longitude and latitude of two points respectively.

[0080] Establish a geographically weighted regression model, which is a spatial analysis technology that has evolved from the traditional linear regression model. It can solve the problems of heterogeneity and spatial non-stationarity in spatial data. Input the typhoon moving distance and the air pressure at the corresponding time of the predicted typhoon path into the geographically weighted regression model to obtain the typhoon wind speed.

[0081] The geographically weighted regression model is expressed as follows:

[0082]

[0083] Among them, y i represents the wind speed value at the i-th sampling point, β0(u i ,v i ) represents the intercept term, p represents the total number of explanatory variables, (u i ,v i ) represents the coordinates of the i-th sampling point, β k (u i ,v i ) represents the regression analysis coefficient, x ik represents the kth explanatory variable of the ith sampling point, ε i is the regression residual.

[0084] Regression analysis coefficient of the geographically weighted regression model: estimated using the weighted least squares method, the specific calculation formula is as follows:

[0085]

[0086] in, represents the estimate of the regression analysis coefficient, W(u i ,v i ) is the spatial weight diagonal matrix, X is the design matrix of the independent variables, and Y is the matrix of the dependent variable wind speed.

[0087] The specific expression of the spatial weight diagonal matrix is ​​as follows:

[0088] Where W ij is the weight from the jth sample point to the ith point to be estimated, d ij is the Euclidean distance between the i-th point to be estimated and the j-th sample point, h is the bandwidth, and the bandwidth can be determined using the AIC information criterion.

[0089] Since air pressure and wind speed are highly correlated, when establishing the relationship model between wind speed and typhoon moving distance, the air pressure variable is considered to be introduced as a control variable. That is, the input variables of the wind speed prediction model are typhoon moving distance and air pressure, and the output variable is wind speed.

[0090] The above-mentioned predicted typhoon path and typhoon wind speed station are the elements of wind weather forecast.

[0091] Embodiment 2: In an exemplary embodiment, a computer-readable storage medium is included, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned typhoon weather forecasting method is implemented.

[0092] Example 3: Please refer to Figure 2 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0093] Wherein, a computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned typhoon weather forecasting method.

[0094] In order to verify the effectiveness of the method of the present invention, the LSTM model and Transformer model commonly used in time series analysis are used for comparison. The designed network hyperparameters are shown in Table 1:

[0095] Table 1

[0096]

[0097]

[0098] The model is evaluated using MSE, RMSE, MAE and DTW. MSE is the square of the difference between the true value and the predicted value and then summed and averaged. The calculation formula is as follows:

[0099] RMSE is the root mean square error, and its calculation formula is as follows:

[0100] MAE is the squared absolute error, and its calculation formula is as follows:

[0101] In the formula, y i represent the predicted value and the true value respectively.

[0102] The DTW algorithm is used to measure the similarity of two time series of different lengths. There is a certain similarity between time series on the time axis, but due to the different number of sampling points on different time series, and the uncertainty of the correspondence between two similar time series, DTW can find the optimal match between the two sequences by adaptively warping the two sequences and performing dynamic time warping, thus achieving accurate spatiotemporal alignment between the two sequences. The specific algorithm formula of DTW is as follows:

[0103]

[0104] S= <s1,s2,…s n >

[0105] T= <t1,t2,…t n >

[0106] Where S and T are two time series, n and m are the number of sampling points of time series S and T respectively. Head(S) represents the first point s1 of the trajectory of S; Rest(S) represents the trajectory formed by removing s1 from the time series S. Similarly, Head(T) represents the first point t1 of the trajectory of T; Rest(T) represents the trajectory formed by removing t1 from the time series T.

[0107] The model error evaluation of the LSTM model and the typhoon path prediction model (Transformer) of the present invention is shown in Table 2. It can be seen from the table that the typhoon path prediction model of the present invention has the smallest error indicators. In this typhoon path prediction experiment, the prediction performance of the typhoon path prediction model of the present invention is higher than the prediction accuracy of the LSTM model. Moreover, when the training round is 20, the typhoon path prediction model of the present invention is close to convergence (see Appendix 2). Figure 3 ).

[0108] Table 2

[0109]

[0110] Schematic diagram of the iterative process of training the typhoon path prediction model based on Transformer in the embodiment of the present invention Figure 3 , the embodiment of the present invention verifies the path matching graph of Typhoon Bebejia based on the typhoon path prediction model DTW based on Transformer Figure 4 , LSTM DTW verification of Typhoon Bebejia’s path matching graph reference Figure 5 , the schematic diagram of the path of Typhoon Bebejia predicted by the typhoon path prediction model based on Transformer in the embodiment of the present invention is referenced Figure 6 , LSTM predicted typhoon Bebejia path diagram reference Figure 7From the above prediction results, it can be seen that the typhoon path prediction model based on Transformer of the present invention has better prediction effect than the LSTM model.

[0111] The regression results of the geographically weighted regression model are shown in Table 3. From Table 3, it can be seen that the goodness of fit value R2 of the geographically weighted regression model is 0.854, which shows that the model fitting effect is good. In addition, the regression coefficient of the typhoon moving distance is positive and negative, that is, the variable has spatial non-stationarity, which indicates that when other conditions remain unchanged, the intensity and direction of the impact of the typhoon moving distance on the typhoon center wind speed will change with the evolution of space.

[0112] Table 3

[0113]

[0114] Typhoon Bebejia's actual wind speed (2024.09.10~2024.09.19) Figure 8 The wind speed of Typhoon Bebejia (2024.09.10~2024.09.19) predicted by the method of the embodiment of the present invention is referenced to Fig. 9 .from Figure 8 and Fig. 9 It can be seen that the prediction effect of the present invention is relatively good, and the overall change trends of the predicted value and the true value are roughly the same.

[0115] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A typhoon weather forecast method, characterized in that: The following steps are involved: S1. Obtaining typhoon data of time series at different time intervals, the typhoon data including: typhoon longitude, latitude, typhoon level, wind speed, air pressure and temperature data; S2. Build a typhoon path prediction model based on Transformer, input typhoon data of time series at different time intervals into the typhoon path prediction model, and obtain the typhoon path; S3. Obtain the typhoon longitude and latitude data on the predicted typhoon path and the air pressure at the corresponding time of the predicted typhoon path, and use the typhoon longitude and latitude data on the predicted typhoon path to calculate the typhoon moving distance; establish a geographically weighted regression model, input the typhoon moving distance and the air pressure at the corresponding time of the predicted typhoon path into the geographically weighted regression model to obtain the typhoon wind speed.

2. A typhoon weather forecast method according to claim 1, characterized in that: The sliding window method is used to convert the obtained time series typhoon data into a data set suitable for supervised learning. The length of the time series typhoon data is L, the length of the sliding window is a, and the sliding window moving step is 1, so L-a+1 data sequences are obtained. The data of the first a-1 moments of each data sequence are used as the input of the typhoon path prediction model, and the data of the a-th moment is used as the label. The supervised typhoon path prediction model predicts the offset of longitude and latitude at the a-th moment.

3. A typhoon weather forecasting method according to claim 1, characterized in that: The typhoon moving distance is calculated using the typhoon's longitude and latitude data as follows: Among them, d represents the typhoon moving distance, r represents the radius of the earth, (lon1, lat1) and (lon2, lat2) represent the longitude and latitude of two points respectively.

4. A typhoon weather forecasting method according to claim 1, characterized in that: The multi-head attention mechanism of Transformer is expressed as: MultiHeadAttn(Q,K,V)=Connect(head1,…head i ,…head H )W o head i =Attention(Q i ,K i ,V i ) in, denote query, key, and value vectors respectively, Q i ,K i ,V i denotes the i-th query, key, and value vectors, respectively. N and M denote the lengths of Q, K, and V, respectively. D k and D v Respectively represent the data dimensions of K and V, head i represents the i-th attention head, H represents the number of attention heads, and W o represents the weight, Represents the scaling factor.

5. A typhoon weather forecasting method according to claim 1, characterized in that: The expression of the Transformer feedforward neural network is: FFN(H ′ )=ReLU(H ′ W 1 +b 1 )W 2 +b 2 , Among them, FFN() represents the output of the feedforward neural network, H′ is the output of the previous layer, ReLU() represents the ReLU activation function, and W 1 and W 2 represents the weight, b 1 and b 2 Indicates offset top.

6. A typhoon weather forecasting method according to claim 5, characterized in that: The expression of the Add&Normalize layer of Transformer is: H=Norm(FFN(H ′ ) + H ′ ) H ′ =Norm(MultiHeadAttn(X)+X), Among them, Norm() represents the normalization function, MultiHeadAttn() represents the output of the multi-head attention mechanism, and X represents the input of the multi-head attention mechanism.

7. A typhoon weather forecasting method according to claim 1, characterized in that: The geographically weighted regression model is expressed as follows: Among them, y i represents the wind speed value at the i-th sampling point, β0(u i ,v i ) represents the intercept term, p represents the total number of explanatory variables, (u i ,v i ) represents the coordinates of the i-th sampling point, β k (u i ,v i ) represents the regression analysis coefficient, x ik represents the kth explanatory variable of the ith sampling point, ε i is the regression residual.

8. A typhoon weather forecasting method according to claim 7, characterized in that: Regression analysis coefficient of the geographically weighted regression model: estimated using the weighted least squares method, the specific calculation formula is as follows: in, represents the estimate of the regression analysis coefficient, W(u i ,v i ) is the spatial weight diagonal matrix, X is the design matrix of the independent variables, and Y is the matrix of the dependent variable wind speed; The specific expression of the spatial weight diagonal matrix is ​​as follows: Where W ij is the weight from the jth sample point to the ith point to be estimated, d ij is the Euclidean distance between the i-th point to be estimated and the j-th sample point, and h is the bandwidth.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 8.

Images