Space-time fusion multi-effect quantity prediction method and device for hydraulic structure and electronic equipment

Abstract

The invention discloses a hydraulic structure space-time fusion multi-effect quantity prediction method and device and electronic equipment, and the method comprises the steps: carrying out the grid division of a hydraulic structure according to the monitoring points of the hydraulic structure, carrying out the monitoring data collection of the monitoring points of each grid at a preset moment, and enabling the monitoring data to comprise environment quantity monitoring data and effect quantity monitoring data; performing normalization preprocessing, abnormal value elimination, missing value complementation processing and space-time processing on the acquired monitoring data to form a construction input data set of the monitoring data; inputting the constructed input data set into a GRU gating circulation unit network, and extracting a multi-measuring-point effect quantity time sequence characteristic under the driving of an environment quantity, namely a time sequence characteristic vector; inputting the constructed input data set into a CNN convolutional neural network, and extracting a spatial feature matrix of spatial distribution among multiple measurement points; and splicing the obtained time sequence feature vector and the spatial feature matrix, inputting a full connection layer through a convolutional neural network to carry out nonlinear fusion, and synchronously outputting the performance prediction values of the hydraulic structure at multiple measurement points.

Description

Technical Field

[0001] This invention relates to a method, apparatus, and electronic device for predicting spatiotemporal fusion multi-effect quantities of hydraulic structures. Background Technology

[0002] Safety monitoring and performance prediction of hydraulic structures are key research areas in reservoir hydraulic structure safety. Affected by dynamic changes in reservoir water level, rainfall, and other environmental factors, monitoring data for hydraulic structures exhibits highly nonlinear, time-varying, and uncertain characteristics, with spatiotemporal differentiation among different monitoring points. Traditional statistical models, with their linear assumptions, struggle to characterize complex nonlinear relationships, while deterministic models are limited by the complexity of parameter calibration and insufficient timeliness. The rapid development of big data, artificial intelligence, and other technologies and theories provides strong technical support for data-driven methods. Many scholars have developed performance prediction models for hydraulic structures using methods such as convolutional neural networks, long short-term memory neural networks, multi-output support vector machines, and graph convolution, demonstrating the feasibility of using these new technologies for performance prediction. However, existing data-driven methods often focus on single-monitoring-point modeling, neglecting the spatiotemporal synergistic effects of the effect field formed by multiple monitoring points, leading to limited prediction accuracy. Therefore, multi-monitoring-point joint modeling for performance prediction has become a current research hotspot. Summary of the Invention

[0003] The purpose of this invention is to propose a spatiotemporal fusion multi-effect prediction method, device and electronic equipment for hydraulic structures, so as to improve the accuracy and efficiency of performance prediction of hydraulic structures at multiple measurement points.

[0004] To achieve the above objectives, the solution of the present invention is as follows:

[0005] A method for predicting the behavior of multiple effects of hydraulic structures through spatiotemporal fusion, the method comprising the following steps:

[0006] Step 1: Divide the hydraulic structure into grids based on the monitoring points of the hydraulic structure, and collect monitoring data at the scheduled time for each grid monitoring point. The monitoring data includes environmental quantity monitoring data and effect quantity monitoring data.

[0007] Step 2: Perform normalization preprocessing, outlier removal and missing value completion on the acquired monitoring data, and perform spatiotemporal processing to form the input dataset for constructing the monitoring data.

[0008] Step 3: Input the constructed input dataset into the GRU gated recurrent unit network to extract the time-series features of multi-point effect quantities driven by environmental quantities, which is the time-series feature vector;

[0009] Step 4: Input the constructed input dataset into the CNN convolutional neural network to extract the spatial feature matrix of the spatial distribution among multiple measurement points;

[0010] Step 5: Concatenate the obtained temporal feature vector with the spatial feature matrix, input it into a fully connected layer of a convolutional neural network for nonlinear fusion, and simultaneously output the predicted values ​​of the hydraulic structure performance at multiple measurement points.

[0011] The scheme further includes: the environmental monitoring data includes reservoir water level and rainfall, and the effect monitoring data includes dam deformation and seepage.

[0012] The solution further includes: In step 2, the spatiotemporal processing is as follows: For each monitoring point of the grid of the hydraulic structure, a two-dimensional data matrix is ​​formed at each time moment. Multiple two-dimensional data matrices at different times are stacked in chronological order to form a three-dimensional data matrix. The three-dimensional matrices of different sections are further combined into a four-dimensional data matrix to form a constructed input dataset with a unified time interval and total monitoring amount, and retaining spatiotemporal structural information.

[0013] The solution further includes: the grid is divided into equal coordinate averages based on the distribution of monitoring points in hydraulic structures; during monitoring data collection, when no monitoring point data is found within the grid, the monitoring data for that point is assigned a value of 0; the location of the monitoring point is indexed by its relative coordinate position.

[0014] Based on the relative positions of the coordinates, the Pearson similarity coefficient is used to measure the correlation between time series data of effect quantities at different monitoring points, in order to determine the spatial relationship strength between effect quantities at each pair of monitoring points, thereby enhancing the extraction of spatial features.

[0015] The solution further includes: outlier removal is performed using the Raida criterion, and missing values ​​are filled in using quadratic interpolation.

[0016] The solution further includes: the GRU gated cyclic unit network contains update gates and reset gates, and its calculation formula is as follows:

[0017]

[0018] in:

[0019] The input information at time t;

[0020] To update the door; The hidden state at the current time step;

[0021] This is the candidate hidden state;

[0022] This is the hidden state from the previous time step;

[0023] To reset the door;

[0024] , , The weights and bias matrices are continuously updated for the gated recurrent unit network;

[0025] It is a sigmoid activation function that outputs a value in the range [0,1]. It updates the gate control to retain historical information and resets the gate control to forget the past state. The two work together to capture temporal dependencies.

[0026] The solution is further as follows: In step 4, the extraction of the spatial feature matrix of the spatial distribution among multiple measurement points is achieved by: using a convolution kernel to slide and extract local features, and then outputting the spatial feature matrix after dimensionality reduction by a pooling layer.

[0027] The solution further includes: the GRU gated cyclic unit network has no fewer than 2 layers and no fewer than 32 hidden units.

[0028] The solution further includes: the CNN convolutional neural network uses a 3×3 convolutional kernel with a stride of 1, and the network injection activation function is ReLU; the feature fusion layer has no less than 64 neurons, and the activation function is Sigmoid.

[0029] An apparatus and electronic device for implementing the aforementioned behavior prediction method, the apparatus comprising a data acquisition module, a temporal extraction module, a spatial extraction module, and a feature fusion module, wherein:

[0030] The data acquisition module is used to perform the work in steps 1 and 2;

[0031] The time-series extraction module is used to perform the work in step 3;

[0032] The spatial extraction module is used to perform the work in step 4;

[0033] The feature fusion module is used to perform the work in step 5;

[0034] The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1 to 9.

[0035] This invention divides hydraulic structures into grids and collects environmental monitoring data from multiple monitoring points at predetermined times. The monitoring points at each cross-section of the hydraulic structure form a two-dimensional data matrix at each time point, which is then stacked over time to form a three-dimensional data matrix. Multiple cross-sections further constitute a four-dimensional data matrix, providing structured input for subsequent spatiotemporal feature extraction. A gated recurrent unit (GRU) network is used to extract the temporal features of multi-measuring point effects driven by environmental quantities, outputting a temporal feature vector. A convolutional neural network (CNN) is used to mine the spatial distribution features between measuring points, obtaining a spatial feature matrix through convolution and pooling operations. The two types of features are concatenated and input into a fully connected layer for nonlinear fusion, achieving synchronous prediction of multi-measuring point effects. This method fully integrates spatiotemporal features, enabling high-precision synchronous prediction of multi-effects across the entire cross-section of hydraulic structures.

[0036] The invention will be further explained in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method of the present invention;

[0038] Figure 2 This is a schematic diagram of the grid division of monitoring points for hydraulic structures according to the present invention;

[0039] Figure 3 This is a schematic diagram illustrating the spatiotemporal correlation of different measurement points in this invention;

[0040] Figure 4 This is a schematic diagram of the structure of the device of the present invention;

[0041] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0042] Example 1:

[0043] A spatiotemporal fusion multi-effect performance prediction method for hydraulic structures is proposed to improve the accuracy and efficiency of performance prediction for hydraulic structures with multiple measurement points. The hydraulic structure can be a dam, for example... Figure 1 , Figure 2 and Figure 3 As shown, the method steps include:

[0044] Step 1: Divide the hydraulic structure into grids based on the monitoring points of the hydraulic structure, and collect monitoring data at predetermined times for the monitoring points of each grid. The monitoring data includes environmental quantity monitoring data, such as reservoir water level and rainfall, and effect quantity monitoring data, such as dam deformation and seepage.

[0045] Step 2: Perform normalization preprocessing, outlier removal and missing value completion on the acquired monitoring data, and perform spatiotemporal processing to form the input dataset for constructing the monitoring data.

[0046] Step 3: Input the constructed input dataset into the GRU gated recurrent unit network to extract the time-series features of multi-point effect quantities driven by environmental quantities, which is the time-series feature vector;

[0047] Step 4: Input the constructed input dataset into the CNN convolutional neural network to extract the spatial feature matrix of the spatial distribution among multiple measurement points;

[0048] Step 5: Concatenate the obtained temporal feature vector with the spatial feature matrix, input it into a fully connected layer of a convolutional neural network for nonlinear fusion, and simultaneously output the predicted values ​​of the hydraulic structure performance at multiple measurement points.

[0049] The GRU gated recurrent unit network is an improved recurrent neural network. Through the update gate and reset gate mechanism, it can effectively capture long-term dependencies in time series and effectively solve the gradient vanishing problem in traditional recurrent neural networks. The update gate controls the influence of the previous state on the current state, and the reset gate controls the degree to which past information is ignored. The GRU gated recurrent unit network has no less than 2 layers and no less than 32 hidden units.

[0050] The CNN (Convolutional Neural Network) is a deep learning model specifically designed for processing data with a grid structure. Its architecture includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The convolutional layers extract local spatial features by sliding convolutional kernels across the input data. In this embodiment, the CNN uses 3×3 convolutional layers with a stride of 1. This configuration preserves the spatial resolution of the feature map while extracting the spatial relationships between adjacent monitoring points. The network injection activation function uses ReLU (Rectified Linear Unit), defined as f(x) = max(0,x), which effectively alleviates the gradient vanishing problem and accelerates training convergence. The feature fusion layer has at least 64 neurons, and the activation function is Sigmoid. The pooling layer reduces data dimensionality through downsampling while retaining key features. Max pooling or average pooling is commonly used; this embodiment employs max pooling, selecting the maximum value within a local region as the output. Through multiple convolutional and pooling operations, the CNN can progressively combine local features into a global spatial pattern, ultimately outputting a spatial feature matrix.

[0051] Preferred scheme: In step S2, the GRU network has 2 layers and no less than 32 hidden units; the CNN convolution kernel size is 3×3 with a stride of 1 and the activation function is ReLU; the feature fusion layer has no less than 64 neurons and the activation function is Sigmoid.

[0052] The grid division is based on the distribution of monitoring points on hydraulic structures, and the grid is as follows: Figure 2 As shown, dividing hydraulic structures into grids yields a data grid with a series of measurements. Each small square contains at most one measurement point; points without measurement points are assigned a value of 0. The location of measurement points is indexed by their relative coordinates. The measurement process of each point in the data grid is a typical time series. There are similarities and differences between multiple measurement points.

[0053] In step 2, the spatiotemporal processing is as follows: a two-dimensional data matrix is ​​formed for each monitoring point of the grid of the hydraulic structure at each time moment. The two-dimensional data matrices of multiple time moments are stacked in chronological order according to the time interval to form a three-dimensional data matrix. The three-dimensional matrices of different cross-sections are further combined into a four-dimensional data matrix, forming a constructed input dataset with a unified time interval and total monitoring amount, and retaining the spatiotemporal structure information.

[0054] The two-dimensional data matrix is ​​used to organize the monitoring data of all measuring points on the same cross section at the same time, maintaining the spatial structure of the measuring point data within the cross section; the three-dimensional data matrix reflects the pattern of monitoring data changing over time by adding a time dimension; the four-dimensional data matrix realizes a unified representation of multi-section data of the entire hydraulic structure by adding a cross section dimension; this data structure provides a foundation for GRU and CNN to extract temporal and spatial features respectively.

[0055] Wherein: the grid is arranged according to the distribution of monitoring points on hydraulic structures, such as... Figure 2 The equal coordinate average division shown means that when there is no monitoring point data in the grid during monitoring data collection, the monitoring data of that point is assigned a value of 0, and the monitoring point position is indexed by the relative coordinate position.

[0056] Based on the relative positions of the coordinates, the Pearson correlation coefficient is used to measure the correlation between time series data of effect quantities at different monitoring points, in order to determine the spatial relationship strength between effect quantities at each pair of monitoring points, thereby enhancing the extraction of spatial features.

[0057] The Pearson similarity coefficient is a statistic that measures the degree of linear correlation between two variables. Its value is between -1 and 1, with positive values ​​indicating positive correlation and negative values ​​indicating negative correlation. The larger the absolute value, the stronger the correlation. By calculating the spatial relationship strength between every two monitoring points in the predetermined monitoring points, spatial prior information is provided to CNN, enhancing the effectiveness of spatial feature extraction.

[0058] For example, if we select two monitoring points A and B, and their effect size data over a past period are x and y respectively, the formula for calculating the Pearson correlation coefficient r is:

[0059]

[0060] In the formula, x i y i This represents the i-th value in the sequence; , The mean of the series is given. The correlation coefficient r ranges from -1 to 1, with a higher correlation as the absolute value approaches 1. The positive and negative signs of the correlation coefficient represent positive and negative correlations, respectively. Specifically, |r| ≥ 0.8 indicates a very strong correlation, 0.6 ≤ |r| < 0.8 indicates a strong correlation, and 0.4 ≤ |r| < 0.6 indicates a moderate correlation.

[0061] The outlier removal process employs the Raida criterion to remove outliers and uses quadratic interpolation to fill in missing values.

[0062] In this embodiment: the GRU gated cyclic unit network includes update gates and reset gates, and the calculation formula for the GRU calculation process is as follows:

[0063]

[0064] in: The input information at time t; To update the door; The hidden state at the current time step; This is the candidate hidden state; This is the hidden state from the previous time step; To reset the door; , , The weights and bias matrices are continuously updated for the gated recurrent unit network; It is a sigmoid activation function that outputs a value in the range [0,1]. The update gate controls the degree to which historical information is retained and the degree to which the hidden state of the previous time step affects the hidden state of the current time step. The closer its value is to 1, the more historical information is retained. The reset gate controls the degree to which past states are forgotten and the degree to which past information is ignored. The closer its value is to 0, the more past information is forgotten. The two work together to capture temporal dependencies.

[0065] In step 4, the extraction of the spatial feature matrix of the spatial distribution among multiple measurement points involves: extracting local features using a sliding convolutional kernel, followed by dimensionality reduction via a pooling layer to output the spatial feature matrix. This process utilizes a CNN convolutional neural network to uncover the spatial distribution patterns of multiple measurement points.

[0066] CNN is a deep learning architecture specifically designed for processing gridded data. Its basic components include convolutional layers, pooling layers, and fully connected layers.

[0067] Convolutional layers extract local spatial features by sliding convolutional kernels across the input data. In this application, a 3×3 convolutional kernel with a stride of 1 is used. This configuration can extract the spatial relationships between adjacent monitoring points while maintaining the spatial resolution of the feature map. The activation function used is ReLU (Rectified Linear Unit), defined as f(x) = max(0,x), which can effectively alleviate the gradient vanishing problem and accelerate training convergence.

[0068] Pooling layers reduce data dimensionality while preserving key features through downsampling operations. Commonly used methods include max pooling and average pooling; this application employs max pooling, selecting the maximum value within a local region as the output.

[0069] Through multi-layer convolution and pooling operations, CNNs can gradually combine local features into a global spatial pattern, and finally output a spatial feature matrix.

[0070] Example 2:

[0071] An apparatus and electronic device for implementing the behavior prediction method described in Embodiment 1, such as Figure 4 and Figure 5 As shown, the device includes a data acquisition module, a time-series extraction module, a spatial extraction module, and a feature fusion module, wherein:

[0072] The data acquisition module is used to perform the work of steps 1 and 2, to divide the hydraulic structure into grids to collect environmental quantities and multi-point effect quantities, and to perform normalization processing to construct the input dataset; each cross section will form a two-dimensional data matrix at each time, and multiple data matrices will be stacked according to the time interval to form a three-dimensional data matrix, and different cross sections will further form a four-dimensional data matrix.

[0073] The time series extraction module is used to implement the work in step 3. Through the gated recurrent unit (GRU) network, it extracts causal time series features based on the time-varying curves of environmental quantities and multiple measurement point effect quantities, and obtains a time series feature vector.

[0074] The spatial extraction module is used to implement the work in step 4. It mines the spatial distribution pattern of multiple measurement points through a convolutional neural network, extracts local features by sliding convolutional kernels, and outputs a spatial feature matrix after dimensionality reduction by a pooling layer.

[0075] The feature fusion module is used to implement the work in step 5, which concatenates the temporal feature vector with the spatial feature matrix, inputs it into the fully connected layer for nonlinear fusion, and synchronously outputs the predicted values ​​of the behavior of hydraulic structures at multiple measurement points.

[0076] The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in Embodiment 1.

[0077] In applications, electronic devices may include, but are not limited to, processors and memory. Figure 5 This is merely an example of an electronic device and does not constitute a limitation on the device. It may include more or fewer components than illustrated, or combinations of certain components, or different components, such as input / output devices, network access devices, etc. Input / output devices may include cameras, audio capture / playback devices, displays, etc. Network access devices may include network modules for wireless network communication with external devices.

[0078] In applications, the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0079] In applications, memory can be an internal storage unit of an electronic device in some embodiments, such as a hard drive or RAM. In other embodiments, memory can be an external storage device of the electronic device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Memory can also include both internal and external storage units of the electronic device. Memory is used to store operating systems, applications, bootloaders, data, and other programs, such as program code for computer programs. Memory can also be used to temporarily store data that has been output or will be output.

[0080] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps in the above-described method embodiments.

[0081] This application implements all or part of the processes in the methods of the above embodiments, which can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can at least include: any entity or device capable of carrying the computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, such as a USB flash drive, a portable hard drive, a magnetic disk, or an optical disk.

[0082] Those skilled in the art will recognize that the device and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0083] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interface, or the device may be indirectly coupled or communicated, and may be electrical, mechanical, or other forms.

[0084] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

[0085] The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction and are not intended to limit the scope of protection of this application.

[0086] The above-mentioned spatiotemporal fusion multi-effect prediction method for hydraulic structures, through the GRU-CNN parallel architecture, simultaneously captures the environmental driving mechanism (temporal dimension) and spatial correlation law (spatial dimension), solving the problem of single-point model ignoring spatiotemporal synergistic effects; the fusion layer introduces a fully connected network to adaptively integrate multi-dimensional features; it supports synchronous prediction of multiple measurement points across the entire cross section, improving computational efficiency and providing real-time technical support for the performance evolution analysis of hydraulic structures.

Claims

1. A method for predicting the spatiotemporal fusion multi-effect quantity of a hydraulic structure, characterized in that, The method steps include: Step 1, grid division is performed on the hydraulic structure according to the monitoring points of the hydraulic structure, and the monitoring data of each grid at the scheduled time is collected, the monitoring data including environmental quantity monitoring data and effect quantity monitoring data; Step 2, the acquired monitoring data is normalized, abnormal values are removed, missing values are completed, and a constructed input data set of the monitoring data is formed through space-time processing; Step 3, the constructed input data set is input into a GRU gate recurrent unit network to extract time sequence characteristics of the effect quantity under the driving of the environmental quantity, i.e. a time sequence characteristic vector; Step 4, the constructed input data set is input into a CNN convolutional neural network to extract a spatial feature matrix of the spatial distribution among the multiple monitoring points; Step 5, the acquired time sequence characteristic vector and the spatial feature matrix are spliced, nonlinear fusion is performed through a convolutional neural network input full connection layer, and a hydraulic structure state prediction value of the multiple monitoring points is synchronously output.

2. The prediction method of claim 1, wherein, The environmental quantity monitoring data includes reservoir water level and rainfall, and the effect quantity monitoring data includes dam deformation and seepage.

3. The prediction method of claim 1, wherein, In step 2, the space-time processing is that a two-dimensional data matrix is formed for each section of the grid of the hydraulic structure at each time, a plurality of two-dimensional data matrices at different times are stacked in time sequence to form a three-dimensional data matrix, and the three-dimensional matrices of different sections are combined into a four-dimensional data matrix to form a constructed input data set with unified time interval and total monitoring amount, and retaining space-time structure information.

4. The prediction method of claim 1, wherein, The grid is divided by equal coordinate average according to the distribution of the monitoring points on the hydraulic structure, in the monitoring data collection, when there is no monitoring point data in the grid, the monitoring data of the point is assigned as 0, and the position of the monitoring point is indexed through coordinate relative position; Based on the coordinate relative position, the Pearson similarity coefficient is used to measure the correlation between the time sequence data of the effect quantities of different monitoring points to determine the spatial relationship strength between the effect quantities of each two monitoring points, which is used to enhance the extraction of spatial features.

5. The prediction method of claim 1, wherein, The abnormal value removal adopts the Ralda criterion to remove abnormal values, and the quadratic interpolation method is used to complete the missing values.

6. The prediction method of claim 1, wherein, The GRU gate recurrent unit network includes an update gate and a reset gate, and the calculation formula is: In step 4, the spatial feature matrix of the spatial distribution among the multiple monitoring points is obtained by sliding a convolution kernel to extract local features, and then outputting a spatial feature matrix after dimension reduction through a pooling layer. is the input information at time t; to update the door; is the hidden state for the current time step; is a candidate hidden state; is a hidden state of the previous time step; to reset the door; , , updating the weight and bias matrices for the network of gated recurrent units; is a sigmoid activation function that outputs a value in the range [0, 1], the update gate controls how much of the history information is preserved, the reset gate controls how much of the past state is forgotten, and the two work together to capture temporal dependencies.

7. The prediction method of claim 1, wherein, The number of layers of the GRU gate recurrent unit network is not less than 2, and the number of hidden units is not less than 32.

8. The prediction method of claim 1, wherein, The CNN convolutional neural network uses a 3x3 convolution kernel with a step of 1, and the network injects a ReLU activation function; 9. The prediction method of claim 1, wherein, The number of neurons of the feature fusion layer is not less than 64, and the activation function adopts Sigmoid. The device includes a data collection module, a time sequence extraction module, a spatial extraction module, and a feature fusion module, wherein:

10. An apparatus and an electronic device implementing the prediction method of claim 1, characterized in that, The data collection module is used to realize the work of steps 1 and 2; The time sequence extraction module is used to realize the work of step 3; The spatial extraction module is used to realize the work of step 4; The feature fusion module is used to realize the work of step 5; ​ The electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method according to any one of claims 1 to 9 when executing the computer program.

Images