Rainfall prediction method based on dynamic graph convolution network and expansion time sequence convolution

Abstract

The invention relates to the technical field of rainfall prediction, and discloses a rainfall prediction method based on a dynamic graph convolution network and expansion time sequence convolution. The method comprises the following steps: S1, data loading and preprocessing; loading and preprocessing the meteorological data, and constructing a three-dimensional data matrix containing multi-dimensional meteorological features, time dimensions and space dimensions; s2, constructing a dynamic graph convolutional network; and modeling the spatial dynamic dependency relationship of the three-dimensional data matrix through a dynamic graph convolutional network, wherein the dynamic graph convolutional network captures dynamic spatial association among different geographic nodes through a dynamically generated adjacent matrix. By adopting the technical scheme of the dynamic graph convolutional network, an adjacent matrix is dynamically generated through a multi-head attention mechanism, and the dynamic spatial dependency relationship between meteorological stations is captured in real time. The technical effect of accurately reflecting the interaction characteristic of the geographic node along with the time change is achieved. Compared with the technical scheme in the prior art that the static adjacent matrix cannot reflect the dynamic relation of the nodes, the problem that the spatial relation is rigid during modeling is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of rainfall prediction, and in particular to a rainfall prediction method based on a dynamic graph convolutional network and dilated temporal convolution. Background Art

[0002] In meteorological forecasting tasks, especially rainfall forecasting tasks, accurate forecasting models are crucial. Traditional physical modeling methods, such as numerical weather forecast models, are usually based on fluid dynamics and thermodynamic equations, using meteorological data as boundary conditions and initial conditions for iterative calculations. However, the accuracy of such models depends on a large amount of computing resources, and it is difficult to effectively capture complex spatiotemporal dependencies in the short term. As a means of extreme weather warning, rainfall forecasting has strict requirements for real-time prediction of sudden and localized precipitation phenomena, and traditional physical models often cannot meet this demand.

[0003] In recent years, deep learning-based spatiotemporal modeling methods have gradually attracted attention in the field of meteorological data processing and prediction. Deep learning models, especially convolutional neural networks and recurrent neural networks, have performed well in processing spatiotemporal series data. Typical time series prediction models such as long short-term memory networks and gated recurrent units can capture temporal dependencies, but have great limitations in modeling spatial dependencies. In addition, convolutional neural networks are mainly used for image data analysis and are difficult to directly process the complex spatial dynamic relationships in meteorological data.

[0004] Most existing methods rely on fixed time windows to model time series, and these window lengths are usually preset and difficult to flexibly adjust to adapt to dependencies of different time scales. In addition, networks with recurrent structures such as LSTM often encounter gradient vanishing or exploding problems when processing long-span data, resulting in insufficient modeling capabilities for long-range time dependencies. This limitation is particularly prominent in meteorological systems, as changes in rainfall usually involve the combined effects of short-term and long-term factors. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a rainfall prediction method based on dynamic graph convolutional networks and dilated time series convolution, which solves the problems of insufficient spatiotemporal dependency modeling capabilities, limited long-term dependency feature extraction of time series, and serious accumulation of multi-step prediction errors in rainfall prediction technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a rainfall prediction method based on a dynamic graph convolutional network and dilated temporal convolution, comprising the following steps; S1. Data loading and preprocessing: Load and preprocess meteorological data to construct a three-dimensional data matrix containing multi-dimensional meteorological characteristics, time dimension and space dimension; S2. Dynamic graph convolutional network construction: modeling the spatial dynamic dependency of a three-dimensional data matrix through a dynamic graph convolutional network, wherein the dynamic graph convolutional network captures the dynamic spatial associations between different geographic nodes through a dynamically generated adjacency matrix; S3. Construction of dilated temporal convolutional network: Modeling the temporal dependency of spatially modeled data through dilated temporal convolutional network, wherein the dilated temporal convolutional network realizes the extraction of multi-scale temporal dependency features through dilated convolution; S4, model training and verification: Model training is performed based on the spatiotemporal features jointly extracted by the dynamic graph convolutional network and the dilated temporal convolutional network; S5. Model prediction and evaluation: Use the trained model to make multi-step rainfall predictions based on the input meteorological data.

[0007] Preferably, the step S1 specifically includes the following steps: S1.1. Load various meteorological characteristics including rainfall, precipitation rate, water vapor content, temperature, humidity, wind speed, terrain height, cloud cover, etc. from the meteorological dataset; S1.2, perform time alignment and spatial alignment on the loaded multi-dimensional meteorological characteristic data to construct a unified three-dimensional data matrix; S1.3, normalizing the characteristic dimensions of the three-dimensional data matrix, and adjusting the numerical range of each meteorological feature to zero mean and unit variance; S1.4. Generate input time window and prediction target window through sliding window method to form time series training samples and prediction targets.

[0008] Preferably, the step S2 specifically includes the following steps: S2.1, initialize the embedding vector for each geographic node, and calculate the static adjacency matrix between nodes based on the embedding vector; S2.2. Use the multi-head attention mechanism to dynamically adjust the adjacency matrix to capture the dynamic spatial associations between nodes in different time steps. S2.3. Perform convolution operations on node features based on the dynamic adjacency matrix to obtain the dynamic spatial features of geographic nodes.

[0009] Preferably, the multi-head attention mechanism specifically includes: Mapping the geographic node embedding vector into a query vector, a key vector, and a value vector; By calculating the weighted correlation between the query vector and the key vector, the dynamic correlation weight between the nodes at the time step is obtained; Generates a dynamic adjacency matrix based on dynamic relevance weights.

[0010] Preferably, the step S3 specifically includes the following steps: S3.1, convolution processing is performed on time series data through dilated convolution, which expands the temporal receptive field layer by layer and captures the dependent features of different time scales; S3.2, set different expansion rates to achieve multi-scale time-dependent modeling; S3.3. Filter irrelevant information through the gating mechanism and retain important time series features.

[0011] Preferably, the convolution kernel size and dilation rate set in each layer of the dilated convolution expand the temporal receptive field by increasing the dilation rate layer by layer, while extracting multi-scale temporal features within a fixed convolution kernel size range.

[0012] Preferably, the step S4 specifically includes the following steps: S4.1, using the time-windowed input data as training samples, the combination of mean absolute error and Huber loss function is used to measure the prediction error; S4.2, dynamically adjust the learning rate during model training through adaptive learning rate scheduling; S4.3. Use gradient clipping strategy to prevent gradient explosion during training.

[0013] Preferably, the gradient clipping limits the norm of the gradient to prevent excessive gradient updates, thereby improving the stability and convergence of the training process.

[0014] Preferably, the step S5 specifically includes the following steps: S5.1. Generate rainfall prediction results for test data based on the trained dynamic graph convolutional network and dilated temporal convolutional network models; S5.2, support multi-step prediction mode, recursively use the prediction result of the previous time step as the input of the next time step; S5.3. Use mean absolute error, root mean square error, and mean absolute percentage error to evaluate the prediction results.

[0015] Preferably, the joint feature extraction process of the dynamic graph convolutional network and the dilated temporal convolutional network directly transfers the features of each layer to the output layer through jump connections to enhance the feature expression ability and the convergence of model training.

[0016] The present invention provides a rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution, which has the following beneficial effects: 1. The present invention adopts the technical solution of dynamic graph convolutional network, dynamically generates adjacency matrix through multi-head attention mechanism, and captures the dynamic spatial dependency relationship between meteorological stations in real time. It achieves the technical effect of accurately reflecting the interactive characteristics of geographical nodes changing over time. Compared with the technical solution in the prior art where the static adjacency matrix cannot reflect the dynamic relationship of nodes, it solves the problem of rigid spatial relationship in modeling.

[0017] 2. The present invention uses a dilated temporal convolutional network to perform multi-scale modeling of temporal dependency, and uses dilated convolution to expand the range of the temporal receptive field layer by layer. This achieves the effect of efficiently capturing complex dependency features over a long time span. The technical solution of the prior art recurrent neural network method with limited receptive field and high training cost is improved on this basis, solving the problem of low efficiency in processing long-term dependency features.

[0018] 3. The present invention effectively improves the stability and convergence speed of model training by combining adaptive learning rate scheduling and gradient clipping training optimization strategy. Compared with the fixed learning rate training method in the prior art, it solves the problem of model oscillation or training stagnation caused by excessive or small gradients, and ensures the robustness of large-scale meteorological data modeling.

[0019] 4. The present invention can generate accurate rainfall forecast results and comprehensively evaluate the performance of multi-time step forecasts by supporting recursive forecasting and multi-step parallel forecasting. Compared with the traditional solution that only supports single-step forecasting, it solves the problem of excessive error accumulation in long-term rainfall trend forecasting and provides more flexible forecasting capabilities for complex meteorological system analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please see attached Figure 1 , an embodiment of the present invention provides a rainfall prediction method based on a dynamic graph convolutional network and dilated temporal convolution, comprising the following steps; S1. Data loading and preprocessing: Load and preprocess meteorological data to construct a three-dimensional data matrix containing multi-dimensional meteorological characteristics, time dimension and space dimension; In this embodiment, loading of meteorological data includes: Rainfall; rainfall data for each geographical location; Precipitation rate; precipitation rate data for each geographical location; Water vapor content; data on the total water vapor content in the atmosphere; Temperature; Temperature data of each layer of the atmosphere; Humidity; relative humidity data of each layer of the atmosphere; Wind speed; atmospheric wind speed and direction at different altitudes; Terrain height; Terrain height data for each layer of the atmosphere; Cloud cover; Cloud cover in various layers of the atmosphere; Rainfall is a key parameter that directly characterizes precipitation intensity, while water vapor content reflects the potential for precipitation in a meteorological system. Humidity, temperature, and wind speed are important factors that affect air mass movement and cloud formation. Specifically, for meteorological features with inconsistent temporal distribution, interpolation methods are used to supplement the data of missing time steps to ensure that all feature data have observations at the same time step. Spatial alignment is achieved through interpolation and resampling to integrate data of different resolutions into a unified spatial grid; All read data are stacked into a three-dimensional array data with a shape of (time, longitude and latitude, parameters). This allows the model to input meteorological parameters at different time steps and different geographical locations as a whole, thereby capturing spatiotemporal dependencies and stacking the time series data of the file according to the parameter dimension to form X∈R N×M×F A three-dimensional matrix; in; T represents the number of time steps; M represents the number of geographical nodes; F represents the number of meteorological features; The generated three-dimensional matrix X is used as the input data matrix of the rainfall prediction model, which contains the multi-dimensional meteorological characteristic data of each node at each time step; For meteorological features, the StandardScaler standardization method is normalized by column (time dimension) to convert the data to a mean of 0 and a standard deviation of 1. The purpose is to make the numerical scales of different features similar, thereby improving the stability of model training; Normalization is done by adjusting the range of values ​​of each meteorological feature to zero mean and unit variance. Specifically, the normalization formula is: in; X is the original eigenvalue; μ represents the mean of the feature; σ represents the standard deviation of the feature; After normalization, each eigenvalue of the meteorological data matrix is ​​in the same numerical range, reducing the training instability caused by the numerical fluctuations between different eigenvalues; Through the sliding window method, the continuous time series data is divided into multiple input-output pairs, so that the model can learn the mapping relationship from historical data to future data; Set the window length to W, then take the data of W consecutive time steps as input, and select the data of the next H time steps as the prediction target; Specifically, the input time window can be expressed as; X input ={X t ,X t+1 ,…,X t+W-1} The corresponding prediction targets are; Y target ={X t+W ,X t+W+1 ,…,X t+W+H-1} in; t represents the starting position of the time step; W is the input window length; H is the prediction target window length; The specific values ​​of the window length W and the prediction step length H are determined according to the requirements of rainfall prediction. For short-term prediction, W = 6 and H = 2 (in hours), while for long-term prediction, W = 24 and H = 6 (in days); The feature data is stored in the form of (T, M, F), where the data at each time step t is represented as an M × F two-dimensional matrix containing the multidimensional feature values ​​of all geographic nodes; Through the above steps, the present invention realizes the complete transformation of meteorological data from original observations to a unified spatiotemporal format, providing a solid data foundation for the efficient training of subsequent dynamic graph convolutional networks and dilated temporal convolutional networks.

[0023] S2. Dynamic graph convolutional network construction: modeling the spatial dynamic dependency of a three-dimensional data matrix through a dynamic graph convolutional network, wherein the dynamic graph convolutional network captures the dynamic spatial associations between different geographic nodes through a dynamically generated adjacency matrix; In this embodiment, a method of using a dynamic graph convolutional network is proposed in step S2 to accurately model spatial dynamic dependencies by dynamically generating an adjacency matrix, thereby providing more accurate input features for subsequent temporal dependency modeling; In this embodiment, the dynamic adjacency matrix is ​​generated by node embedding, and the geographic node is initially represented as an embedding vector h of fixed dimension. iNode embeddings can be randomly initialized or assigned by geographic features. These embedding vectors are used to calculate the static adjacency matrix A static ,Specifically, the initial relationship weights of nodes i and j are given by the formula; in; h i and h j is the embedding vector of nodes i and j; represents the inner product operation; The multi-head attention mechanism achieves dynamic feature modeling by mapping node embeddings into query, key, and value vectors. The formula is as follows: Q=W q ·h i ,K=W k ·h j ,V=W v ·h j Among them: W q , W k , W v is a mapping matrix, used to generate query, key and value representations respectively; Specifically, the dynamic relevance weight A dynamic,ij It is calculated by weighting the query vector and the key vector; Where: d k is a scaling factor equal to the dimension of the embedding vector, used to prevent gradient explosion; The dynamic adjacency matrix can be adjusted dynamically at the time step to reflect the changes in meteorological conditions at the current time step; In the dynamic graph convolutional network, the convolution operation is performed through the adjacency matrix A dynamic Aggregate node features, the formula is as follows; H (l+1) =σ(A dynamic ·H (l) ·W (l) ) in; H (l) is the feature matrix of the l-th layer nodes; W (l) is the weight matrix; σ is a nonlinear activation function. Through convolution operations, the dynamic features of meteorological nodes can be propagated and integrated layer by layer in the spatial dimension. The dimension of the node feature matrix H is M×F, where M is the number of nodes and F is the number of features. After multiple layers of convolution, the feature vector of the node is strengthened into a high-dimensional feature containing dynamic information of the surrounding nodes; The skip connection transfers the low-level features directly to the high-level ones. Specifically, the output features of the convolution will be added to the features of the previous layer through the residual connection. The formula is: H (l+1) =σ(A dynamic ·H (l) ·W (l) )+H (l) It can enhance the ability of feature expression, alleviate the problem of gradient disappearance, and improve the stability of training; To ensure the consistency of the eigenvalue range, the adjacency matrix is ​​normalized before convolution. The normalization formula is: Where; D is the adjacency matrix A dynamic The degree matrix is ​​used to balance the feature propagation of nodes; Through the above steps, the dynamic graph convolutional network realizes the dynamic spatial modeling of meteorological data. The characteristics of each node not only reflect its own information, but also capture the dynamic relationship with its related nodes. S3. Construction of dilated temporal convolutional network: Modeling the temporal dependency of spatially modeled data through dilated temporal convolutional network, wherein the dilated temporal convolutional network realizes the extraction of multi-scale temporal dependency features through dilated convolution; In this embodiment, the problem of time dimension feature extraction is solved by dilating the temporal convolutional network, combining dilated convolution and multi-scale modeling, laying a key foundation for the final rainfall prediction. By introducing the dilation rate d between the convolution kernels, the time step range covered by the receptive field is increased; Specifically, the output of the dilated convolution can be expressed as; in; y[i] is the output of the i-th time step; x[i] is the input time series; w[k] is the weight of the convolution kernel; K is the convolution kernel size; d is the expansion rate; By adjusting the dilation rate d, the receptive field can be flexibly expanded to capture features with a longer time span; The dilated temporal convolutional network consists of multiple convolutional layers, each using a different dilation rate. The first convolutional layer is responsible for extracting short-term temporal dependency features, and subsequent layers gradually capture features over longer time ranges; Specifically, the design formula for multi-layer convolution is: in; H (l) is the input feature of the lth layer; W (l) is the corresponding convolution kernel weight; The expansion rate is d l Convolution operation; σ is the activation function; Through multi-layer superposition, features at different time scales are integrated into the same network; On the basis of dilated convolution, the network introduces a gating unit to select important temporal features. The gating mechanism consists of two parts: one part generates the filter and the other part generates the gating signal. The specific formula is as follows; z[i]=tanh(f(x[i]))·sigmoid(g(x[i])) in; f(x[i]) and g(x[i]) are the filter signal and gate signal generated by the convolution operation, respectively; Tanh and sigmoid are activation functions; The final output z[i] is the filtered time feature, which can avoid the interference of irrelevant information on the prediction results; By directly connecting the input of each convolution layer to the output, the underlying features are fully utilized in the network. The specific formula is: H (l+1) =F(H (l) )+H (l) Among them, F(H (l) ) is the output of the current convolutional layer, H (l) is the input feature of the previous layer; Layer normalization is used to standardize the output of each convolution layer to make the numerical range more stable. Layer normalization is done through the following formula; Among them, μ and σ are the mean and standard deviation of the output features respectively. The standardized features can accelerate the training process and improve the convergence of the network. By expanding the temporal convolutional network, the present invention realizes multi-scale modeling capabilities from short-term to long-term in the time dimension, providing comprehensive time feature input for subsequent rainfall forecasts. Combined with dynamic spatial features, the present invention can provide efficient and accurate rainfall forecasts under complex meteorological conditions.

[0024] S4, model training and verification: Model training is performed based on the spatiotemporal features jointly extracted by the dynamic graph convolutional network and the dilated temporal convolutional network; In this embodiment, step S4 optimizes the parameters of the model through a carefully designed training and verification strategy to maximize the feature expression potential of the data while ensuring the stability and generalization ability of the model training. This step includes a variety of optimization measures, including the design of the loss function, learning rate scheduling, gradient clipping, and dynamic evaluation of the verification process; The loss function adopts the combination of mean absolute error and Huber loss. MAE can measure the overall error of the prediction well. Its formula is as follows; in; y i is the i-th true value; is the corresponding predicted value; N is the sample size; Huber loss is a smooth loss function suitable for noisy data. Its formula is: The Huber loss treats small errors similarly to mean square error, while it exhibits anti-interference ability similar to MAE for large errors; L total =α·L MAE +β·L Huber Among them, α and β are weight coefficients used to balance the impact of the two losses; A fixed learning rate may cause oscillation or premature convergence during training. To solve this problem, the present invention adopts the ReduceLROnPlateau strategy. When the validation set loss no longer decreases after several rounds of training, the learning rate will automatically decrease. The initial learning rate is set to lr0, and each adjustment is multiplied by the attenuation factor γ. The specific update rules are as follows; 1 t+1 =lr t ·γ,if validation loss does not improve Among them, lr t Represents the learning rate of the tth iteration, γ is usually set to 0.1 or 0.5; The gradient of the neural network may be too large at some levels, resulting in unstable model parameter updates. Gradient clipping limits the norm of the gradient to ensure that the amplitude of each update does not exceed the preset threshold. Specifically, the operation of gradient clipping can be expressed as; in; g is the original gradient; τ is the threshold for gradient clipping; ‖g‖2 is the second norm of the gradient; When the gradient norm exceeds the threshold, the gradient will be scaled to the range of τ; The data distribution of the validation set and the training set needs to be consistent to ensure the reliability of the evaluation results. After each round of training, the prediction error of the model is calculated using the validation set, and the validation loss value is recorded; The verification process includes the following steps: Input the data of the validation set to generate rainfall prediction results; Calculate the error between the predicted value and the true value, using the above loss function for measurement; Record the current validation loss value and compare it with the historical minimum value. If the validation loss is further reduced, update the optimal model parameters. In this embodiment, a stable and efficient model training process is implemented, which provides reliable network parameters for the subsequent prediction stage. Based on the actual needs of the rainfall prediction task, it is ensured that the model can fully adapt to the changes in the complex meteorological system.

[0025] S5, model prediction and evaluation; use the trained model to perform multi-step prediction of rainfall on the input meteorological data; in this embodiment, the trained dynamic graph convolutional network and the dilated temporal convolutional network are used to perform multi-step prediction of rainfall on the meteorological data, and the performance of the model is comprehensively evaluated; The initial input window contains data from the past W time steps, which serves as the basis for prediction. Let the input time series be X input , containing multiple time steps, geographic nodes and features, in the form of Where W represents the number of time steps in the input window, M represents the number of geographic nodes, and F represents the number of meteorological features; Specifically, the input window is extracted from the meteorological observation data of the test set. The input window can be adjusted according to the different requirements of the prediction task. For short-term rainfall prediction, the value of W can be set to a smaller value; for long-term rainfall trend analysis, it is recommended to choose a larger W. The model generates rainfall forecast results for the next H time steps, assuming that the initial input window is X input The model generates the prediction value of the first time step through joint feature extraction of dynamic graph convolutional network and dilated temporal convolutional network The formula is as follows; Where f(·) represents the trained prediction model; The prediction of subsequent time steps can be done recursively, that is, the prediction result of the previous time step is used as input and the prediction range is gradually expanded; Specifically, the prediction formula for the tth time step is; Among them, X recursive is the input window, which contains the data of W-1 historical time steps and the t-1th predicted value In order to improve the efficiency and accuracy of prediction, a multi-step parallel prediction method can be combined, that is, by expanding the features of the input window; directly generating prediction results for H time steps, the formula is as follows; Error indicators include mean absolute error, root mean square error and mean absolute percentage error, among which the formula of MAE is; in; N represents the number of test samples; is the predicted value; Y i is the true value; As an alternative, RMSE is suitable for measuring the impact of outliers on model predictions, and its formula is; Specifically, MAPE is used to evaluate the relative proportion of forecast errors, and the formula is as follows; If the true value Y i In the case of zero values, the data can be smoothed to avoid zero division problems in calculations; The prediction results are saved as time series files in the form of npy files or csv files, which contain the predicted time step, geographical location and rainfall value for subsequent analysis; The model prediction and evaluation phase completes the entire process from data input to result output. The recursive prediction method combined with a variety of error evaluation methods provides comprehensive technical support for rainfall prediction. At the same time, multi-format storage and visualization further enhance the interpretability and practicality of the results.

[0026] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution, characterized in that: The steps include: S1. Data loading and preprocessing: Load and preprocess meteorological data to construct a three-dimensional data matrix containing multi-dimensional meteorological characteristics, time dimension and space dimension; S2. Dynamic graph convolutional network construction; modeling the spatial dynamic dependency relationship of the three-dimensional data matrix through the dynamic graph convolutional network, wherein the dynamic graph convolutional network captures the dynamic spatial association between different geographic nodes through a dynamically generated adjacency matrix; S3. Construction of dilated temporal convolutional network: Modeling the temporal dependency of spatially modeled data through dilated temporal convolutional network, wherein the dilated temporal convolutional network realizes the extraction of multi-scale temporal dependency features through dilated convolution; S4, model training and verification: Model training is performed based on the spatiotemporal features jointly extracted by the dynamic graph convolutional network and the dilated temporal convolutional network; S5. Model prediction and evaluation: Use the trained model to make multi-step rainfall predictions based on the input meteorological data.

2. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 1 is characterized in that: In step S1, The steps include: S1.

1. Load multiple meteorological characteristics including rainfall, precipitation rate, water vapor content, temperature, humidity, wind speed, and cloud cover from the meteorological data set; S1.2, perform time alignment and spatial alignment on the loaded multi-dimensional meteorological characteristic data to construct a unified three-dimensional data matrix; S1.3, normalizing the characteristic dimensions of the three-dimensional data matrix, and adjusting the numerical range of each meteorological feature to zero mean and unit variance; S1.

4. Generate input time window and prediction target window through sliding window method to form time series training samples and prediction targets.

3. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 1, characterized in that: The step S2 specifically includes the following steps: S2.1, initialize the embedding vector for each geographic node, and calculate the static adjacency matrix between nodes based on the embedding vector; S2.2, using the multi-head attention mechanism to dynamically adjust the adjacency matrix to capture the dynamic spatial associations between nodes at different time steps; S2.

3. Perform convolution operation on node features based on the dynamic adjacency matrix to obtain dynamic spatial features of geographic nodes.

4. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 3 is characterized in that: The multi-head attention mechanism specifically includes: Mapping the geographic node embedding vector into a query vector, a key vector, and a value vector; By calculating the weighted correlation between the query vector and the key vector, the dynamic correlation weight between the nodes at the time step is obtained; Generates a dynamic adjacency matrix based on dynamic relevance weights.

5. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 1, characterized in that: The step S3 specifically includes the following steps: S3.1, convolution processing is performed on time series data through dilated convolution, which expands the temporal receptive field layer by layer and captures the dependent features of different time scales; S3.2, set different expansion rates to achieve multi-scale time-dependent modeling; S3.

3. Filter irrelevant information through the gating mechanism and retain important time series features.

6. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 5, characterized in that: The dilated convolution sets the convolution kernel size and dilation rate in each layer, expands the temporal receptive field by increasing the dilation rate layer by layer, and extracts multi-scale temporal features within a fixed convolution kernel size range.

7. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 1, characterized in that: In step S4, The steps include: S4.1, using the time-windowed input data as training samples, the combination of mean absolute error and Huber loss function is used to measure the prediction error; S4.2, dynamically adjust the learning rate during model training through adaptive learning rate scheduling; S4.

3. Use gradient clipping strategy to prevent gradient explosion during training.

8. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 7, characterized in that: The gradient clipping limits the norm of the gradient to prevent excessive gradient updates and improve the stability and convergence of the training process.

9. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 1, characterized in that: The step S5 specifically includes the following steps: S5.

1. Generate rainfall prediction results for test data based on the trained dynamic graph convolutional network and dilated temporal convolutional network models; S5.2, support multi-step prediction mode, recursively use the prediction result of the previous time step as the input of the next time step; S5.

3. Use mean absolute error, root mean square error, and mean absolute percentage error to evaluate the prediction results.

10. The rainfall prediction method based on dynamic graph convolutional network and dilated temporal convolution according to claim 1, characterized in that: The joint feature extraction process of the dynamic graph convolutional network and the dilated temporal convolutional network directly transfers the features of each layer to the output layer through skip connections to enhance the feature expression ability and the convergence of model training.

Images