A geoscience information space-time prediction method and device

Abstract

The application discloses a geoscience information space-time prediction method and device, electronic equipment and storage medium, and belongs to the technical field of geoscience information prediction. The method comprises the following steps: abstracting a monitoring site of geoscience information as a node in a graph structure; constructing a space-time ordinary differential equation module to predict the hidden state of the node and improve the nonlinear fitting capability and method interpretability of the prediction method; and constructing a space-time attention module to fuse the hidden state to improve the long-term dependence mining capability of the model. The method introduces the spatial position information by abstracting the monitoring site as the node in the graph structure, improves the nonlinear fitting capability and method interpretability of the space-time ordinary differential equation module through the hidden state, and can capture the long dependence relationship in the space-time data through the space-time attention module fusing the hidden state, thereby improving the prediction accuracy and interpretability of the prediction method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of geoscience information prediction, and particularly relates to a geoscience information spatiotemporal prediction method and device, an electronic device and a storage medium. BACKGROUND

[0002] Spatiotemporal prediction is a basic research proposition of geoscience spatial artificial intelligence, which depends on spatiotemporal data to predict the unknown state of adjacent time periods in the spatiotemporal domain. At present, spatiotemporal prediction technology has been widely applied in intelligent transportation, weather forecasting, earthquake warning and other practical applications.

[0003] Existing spatiotemporal prediction models can be roughly divided into knowledge-driven spatiotemporal prediction models and data-driven spatiotemporal prediction models. The knowledge-driven model establishes a specific mathematical equation to describe the complex spatiotemporal pattern in the data according to the prior knowledge accumulated by predecessors, which leads to high interpretability but poor prediction performance of the knowledge-driven model. Compared with the knowledge-driven model, the data-driven model ignores the prior knowledge accumulated by predecessors. Although the data-driven model can achieve very good prediction performance, the data-driven model is a pure black box model, which has poor interpretability. In the field of geoscience, it is the top priority of GeoAI to improve the prediction accuracy of the model while enhancing the interpretability and transparency of the model. However, the existing spatiotemporal prediction model still cannot balance the model prediction accuracy and interpretability.

[0004] The existing NODE-based prediction model uses a derivative network parameterized by a neural network to establish a relationship between deep learning and an ordinary differential equation, and then predicts the unknown state of a spatiotemporal system. However, the existing NODE-based prediction model sacrifices the prediction accuracy of the model while improving the interpretability of the model. SUMMARY

[0005] The purpose of the present application is to provide a geoscience information spatiotemporal prediction method, a geoscience information spatiotemporal prediction device, an electronic device and a storage medium to solve the problem that the existing geoscience information prediction cannot balance the prediction accuracy and interpretability.

[0006] According to a first aspect of the embodiments of the present application, a geoscience information spatiotemporal prediction method is provided, which can include:

[0007] abstracting a monitoring site of geoscience information as a node in a graph structure;

[0008] constructing a spatiotemporal ordinary differential equation module to predict the hidden state of the node in an interpretable manner;

[0009] constructing a spatiotemporal attention module for fusing the hidden state based on a spatiotemporal attention mechanism;

[0010] The spatiotemporal ordinary differential equation module and the spatiotemporal attention module are fused to obtain a geoscience information prediction model.

[0011] The historical spatiotemporal data of the geoscience information is substituted into the geoscience information prediction model to obtain a prediction result of the geoscience information.

[0012] In some optional embodiments of the present application, the spatiotemporal ordinary differential equation module is constructed to predict a hidden state of the node, comprising:

[0013] The definition of the hidden state of the node is obtained.

[0014] A spatiotemporal derivative network is constructed based on time information and space information.

[0015] The spatiotemporal ordinary differential equation module is constructed to iteratively solve the hidden state at each time point through the spatiotemporal derivative network.

[0016] In some optional embodiments of the present application, before the spatiotemporal ordinary differential equation module and the spatiotemporal attention module are fused to obtain a geoscience information prediction model, the method further comprises:

[0017] A gating mechanism and a residual connection are introduced into the spatiotemporal ordinary differential equation module.

[0018] In some optional embodiments of the present application, the spatiotemporal attention module for fusing the hidden state is constructed based on a spatiotemporal attention mechanism, comprising:

[0019] A spatial attention module for fusing the hidden state in a spatial dimension is constructed.

[0020] A temporal attention module for fusing the hidden state in a temporal dimension is constructed.

[0021] The spatiotemporal attention module is constructed based on the spatial attention module and the temporal attention module.

[0022] In some optional embodiments of the present application, before the spatiotemporal ordinary differential equation module and the spatiotemporal attention module are fused to obtain a geoscience information prediction model, the method further comprises:

[0023] A residual connection is introduced into the spatiotemporal attention module.

[0024] In some optional embodiments of the present application, before the historical spatiotemporal data of the geoscience information is substituted into the geoscience information prediction model to obtain a prediction result of the geoscience information, the method further comprises:

[0025] The geoscience information prediction model is optimized based on a loss function.

[0026] The loss function comprises a temporal dimension loss term and a spatial dimension loss term.

[0027] In some optional embodiments of the present application, the geoscience information includes any one of traffic flow, PM2.5 concentration and air temperature.

[0028] According to a second aspect of embodiments of the present application, a geoscience information spatio-temporal prediction apparatus is provided, which can include:

[0029] an abstracting module configured to abstract a monitoring site of geoscience information as a node in a graph structure;

[0030] a first constructing module configured to construct a spatio-temporal ordinary differential equation module to predict a hidden state of the node;

[0031] a second constructing module configured to construct a spatio-temporal attention module for fusing the hidden state based on a spatio-temporal attention mechanism;

[0032] a fusing module configured to fuse the spatio-temporal ordinary differential equation module and the spatio-temporal attention module to obtain a geoscience information prediction model;

[0033] a predicting module configured to substitute historical spatio-temporal data of the geoscience information into the geoscience information prediction model to obtain a prediction result of the geoscience information.

[0034] According to a third aspect of embodiments of the present application, an electronic device is provided, which can include:

[0035] a processor;

[0036] a memory for storing processor-executable instructions;

[0037] wherein the processor is configured to execute the instructions to implement the geoscience information spatio-temporal prediction method as shown in any one of the embodiments of the first aspect.

[0038] According to a fourth aspect of embodiments of the present application, a storage medium is provided, when instructions in the storage medium are executed by a processor of an information processing apparatus or a server, to enable the information processing apparatus or the server to implement the geoscience information spatio-temporal prediction method as shown in any one of the embodiments of the first aspect.

[0039] The above technical solutions of the present application have the following beneficial technical effects:

[0040] The geoscience information spatio-temporal prediction method provided in the embodiments of the present application introduces spatial position information by abstracting a monitoring site as a node in a graph structure, improves the nonlinear fitting capability of the ordinary differential equation module by predicting a hidden state of the node through the spatio-temporal ordinary differential equation module, and can capture long dependency in spatio-temporal data by fusing the hidden state through the spatio-temporal attention module, thereby being able to improve the prediction accuracy and interpretability of the prediction method at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a flowchart of a method for spatiotemporal prediction of geosciences information in an example embodiment of the present application;

[0042] Figure 2 is a structural diagram of a prediction model for geosciences information in an example embodiment of the present application;

[0043] Figure 3 is a structural diagram of a spatiotemporal derivative network in an example embodiment of the present application;

[0044] Figure 4 is a diagram for solving hidden states in an example embodiment of the present application;

[0045] Figure 5 is a structural diagram of a spatiotemporal attention module in an example embodiment of the present application;

[0046] Figure 6a is a diagram of distribution of traffic monitoring sites in an example embodiment of the present application;

[0047] Figure 6b is a diagram of distribution of air quality monitoring sites in an example embodiment of the present application;

[0048] Figure 6c is a diagram of distribution of temperature monitoring sites in an example embodiment of the present application;

[0049] Figure 7 is a structural diagram of a device for spatiotemporal prediction of geosciences information in an example embodiment of the present application;

[0050] Figure 8 is a structural diagram of an electronic device in an example embodiment of the present application;

[0051] Figure 9 is a structural diagram of a hardware of an electronic device in an example embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and in conjunction with the specific embodiments. It should be understood that the description is only exemplary and is not intended to limit the scope of the present application. In addition, in the following description, the description of known structures and techniques is omitted to avoid unnecessary confusion of the concept of the present application.

[0053] The layer structure diagrams according to the embodiments of the present application are shown in the drawings. The drawings are not drawn to scale, in which some details are exaggerated for the purpose of clarity and some details can be omitted. The shapes of various regions, layers and the relative size and position relationship therebetween shown in the drawings are only exemplary, and in actuality, there can be deviations due to manufacturing tolerances or technical limitations, and regions / layers with different shapes, sizes and relative positions can be additionally designed by a person skilled in the art according to actual needs.

[0054] Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0055] In the description of the present application, it should be noted that the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0056] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as there is no conflict.

[0057] It is found through research that existing NODE-based prediction models, such as space-time ordinary differential equations, space-time graph ordinary differential equations, recurrent neural network ordinary differential equations, long short-term memory network ordinary differential equations, etc. However, the above NODE-based prediction models sacrifice the prediction accuracy of the model while improving the interpretability of the model.

[0058] Firstly, the derivative network (DN) of the above NODE model only depends on the time factor, and does not explicitly model the spatial position information into the output of the derivative network (DN), which makes the above NODE-based prediction model perform poorly in the space-time prediction task. Secondly, the NODE-based prediction model is essentially solving the ODE initial value problem, which makes the NODE-based prediction model heavily dependent on the ODE initial value and difficult to mine the long-time dependence relationship in the space-time data.

[0059] To solve the above problems, the present application provides a geoscience information space-time prediction method, a geoscience information space-time prediction device, an electronic equipment and a storage medium.

[0060] The definitions in the present application include:

[0061] Graph structure: the graph G = <V, E> represents the graph structure abstracted by the monitoring sites, where V = {v ij} represents n nodes in the graph G, i.e. n monitoring sites; E = {e i} represents the node v jThe relationships between them.

[0062] Spatiotemporal State: Spatiotemporal state Represents node v i The geoscientific information values ​​monitored within a time window t, such as traffic flow or air quality per unit time (this application mainly focuses on individual geoscientific information parameters monitored by monitoring stations). The spatiotemporal state of all nodes and all time windows in Figure G can be represented as a spatiotemporal state matrix. in Represents node v i Time series across all time windows Let represent the spatial sequence of all nodes within time window t, where n represents the number of nodes and T represents the total number of time windows.

[0063] The research objective of this application is to establish a geoscientific information prediction model. This model predicts future spatiotemporal data based on the graph structure G and the spatiotemporal state matrix X. Specifically, the modeling process in this paper is shown in Equation (1).

[0064]

[0065] In the formula: The historical data required for the geoscientific information prediction model are represented by k, which represents the time-dependent step size; G represents the graph structure of the study area as an abstraction. Spatiotemporal data representing future (predicted) geoscientific information; This indicates the geoscientific information prediction model proposed in this application, namely the STA-ODE model; Θ represents the learnable parameters in the geoscientific information prediction model.

[0066] The following description, in conjunction with the accompanying drawings, details the spatiotemporal prediction method, spatiotemporal prediction device, electronic device, and storage medium for geoscientific information provided in this application, through specific embodiments and application scenarios.

[0067] like Figure 1 As shown, in a first aspect of the embodiments of this application, a spatiotemporal prediction method for geoscientific information is provided, the method including:

[0068] Step S101: Abstract the monitoring stations of geoscience information as nodes in a graph structure;

[0069] Step S102: Construct a spatiotemporal ordinary differential equation module to predict the hidden state of nodes;

[0070] Step S103: Construct a spatiotemporal attention module for fusing hidden states based on the spatiotemporal attention mechanism;

[0071] Step S104: Fuse the spatiotemporal ordinary differential equation module and the spatiotemporal attention module to obtain the geoscience information prediction model;

[0072] Step S105: Substitute the historical spatiotemporal data of geoscience information into the geoscience information prediction model to obtain the prediction results of geoscience information.

[0073] This embodiment provides a geoscientific information spatiotemporal prediction method that introduces spatial location information by abstracting monitoring stations as nodes in a graph structure. It improves the nonlinear fitting capability and interpretability of the spatiotemporal ordinary differential equation module by predicting the hidden states of the nodes through the spatiotemporal ordinary differential equation module. Furthermore, it captures long-term dependencies in spatiotemporal data by fusing the hidden states through the spatiotemporal attention module, thereby simultaneously improving the prediction accuracy and interpretability of the method. Figure 2 The geospatial information prediction model, namely the STA-ODE model, mainly consists of two modules: a spatiotemporal ordinary differential equation (ST-ODE) module and a spatiotemporal attention (STA) module. In the ST-ODE module, the hidden state of the geospatial information at each time step is defined as a solution to the ordinary differential equation, and the hidden state at each time step is iteratively solved based on a defined spatiotemporal derivative network and gating mechanism. In the STA module, spatial attention and temporal attention are used to fuse the hidden states from multiple historical time steps, thereby capturing long-term dependencies in the spatiotemporal data.

[0074] Specifically, in the STA-ODE model, a hidden state is defined for each node in the graph, and the hidden state at each time step is treated as a solution to an ordinary differential equation, thus enabling interpretable iterative solutions for hidden states at multiple time steps. During the iteration process, a spatiotemporal derivative network that considers both temporal and spatial information is defined to improve the accuracy of the hidden state solution. After the hidden states are solved, a spatiotemporal attention mechanism is integrated into the STA-ODE model to fuse multiple hidden states, thereby further capturing long-term dependencies in the spatiotemporal data.

[0075] With the spatiotemporal state matrix For example, the forward propagation process of the STA-ODE model can be defined by formula (3).

[0076]

[0077] In the formula: This represents the spatiotemporal data of geoscientific information monitored by a node over k historical time points; DN STThe spatiotemporal derivative network represents the hidden state; G represents the graph structure abstracted from the study region; STODE represents the ST-ODE module, used to solve the hidden state of nodes at k time steps; Let d represent the k hidden states obtained by solving the spacetime ordinary differential equation, where d h The dimension represents the hidden state; STA represents the spatiotemporal attention module, used to fuse the hidden states at k time steps. This indicates the result of the fusion of spatial dimensions; This indicates the fusion result along the time dimension; This represents the final output of the STA-ODE model, i.e., the prediction result; Θ STODE This represents the learnable parameters in the ST-ODE module; Θ STA This indicates the learnable parameters in the STA module; This represents the parameters that can be learned during the model's output process.

[0078] Traditional derivative networks are essentially time series models and do not consider the case of multiple monitoring stations, thus resulting in large prediction errors. In view of this, this application defines a derivative network that takes into account both temporal and spatial information, namely the Spatiotemporal Derivative Network (STDN). Compared with traditional derivative networks, the spatiotemporal derivative network has two differences. First, this application regards the hidden state at each time step as a solution to the ordinary differential equation, rather than the observation value at each time step as a solution to the ordinary differential equation, thereby improving the nonlinear fitting ability of the STA-ODE model; second, this application models spatial location information into the derivative network, so that the STA-ODE model can be used for spatiotemporal prediction tasks with multiple monitoring stations, rather than time series prediction tasks with multiple monitoring stations. Specifically, the mathematical definition of the spatiotemporal derivative network is shown in formula (4).

[0079]

[0080] In the formula: This represents the hidden state of n monitoring stations at time t, where This represents the hidden state of the i-th monitoring station at time t. This represents the spatial location information of different monitoring stations, which is further encoded in the spatiotemporal derivative network. For example... Figure 3 As shown, in the spatiotemporal derivative network, due to the different encodings of i and j, node v i and v j The derivative network output g(h) at time t3 i (t3), t3, i) and g(h j (t3), t3, j) will also be closer to the true derivative value, thus making the final predicted value closer to the true value.

[0081] As the iterative model is prone to cause gradient vanishing / explosion phenomenon in the optimization process, thereby leading to slow model optimization, or even unable to optimize. In some embodiments, in order to alleviate the gradient vanishing / explosion phenomenon in the optimization process, the present application introduces a gating mechanism and residual connection in the ST-ODE module to speed up the optimization efficiency of the model, and thus improve the prediction accuracy of the model. As shown in Figure 4 The observation value of the monitoring station is based on the gating mechanism and the spatiotemporal derivative network to solve the hidden state at each time. In addition, since the dimensions of the observation value and the hidden state of the monitoring station are not the same, the present application uses a 1x1 convolution operation to align the two dimensions, thereby completing the residual connection of the observation value and the hidden state. Taking the spatiotemporal state matrix For example, the iterative solving process of the hidden state at each time is shown in formula (5), and the residual connection of the observation value and the hidden state is shown in formula (6). The long short-term memory network

[0082]

[0083]

[0084] In the formula: represents the hidden state of the k time after residual connection; represents the hidden state of all monitoring stations at time t solved by iteration, d h represents the dimension of the hidden state; represents the observation data of geoscience information of all monitoring stations at time t; Z t-1 , R t-1 , represents an intermediate variable in the iteration process; σ represents the sigmoid activation function; tanh represents the tanh activation function; ⊙ represents the hadamard product; [·||·] represents the concatenate function; represents the spatiotemporal derivative network; represents the convolution operation for residual connection, represents the convolution kernel of the convolution network; represents the weight in the iteration process of the hidden state.

[0085] The spatiotemporal attention module is composed of multiple spatial attention blocks and multiple temporal attention blocks, wherein the spatial attention is used to fuse the k hidden states in the spatial dimension, and the temporal attention is used to fuse the k hidden states in the time dimension. For simplicity, Figure 5The operation flow of the spatio-temporal attention module composed of a single spatial attention block and a single temporal attention block is shown. In the spatio-temporal attention module, the present application also integrates the residual connection into the spatio-temporal attention module to accelerate the optimization efficiency of the model. Specifically, Figure 5 The calculation flow of is shown in equations (7) (8) (9).

[0086]

[0087]

[0088]

[0089] In the formula: represents the hidden state at k time, which is obtained by section 4.1.2; represents The transpose of represents the final prediction result of the model at t+1 time; represents the fusion result of the spatial attention at t+1 time; represents the fusion result of the temporal attention at t+1 time; represents the spatial attention matrix, represents the weighted sum of the spatial dimension, which is equivalent to the spatial-based graph convolution operation (reference); represents the temporal attention matrix; represents the weighted sum of the time dimension; represents the learnable parameters of the fully connected layer in the spatio-temporal attention; represents the convolution operation in the spatio-temporal attention; represents the learnable parameters of the convolution layer in the spatio-temporal attention; Intermediate variable in spatio-temporal attention; exp represents the exponential function.

[0090] In the forward propagation process, the STA-ODE model predicts the future spatio-temporal data through the spatio-temporal data of the previous k time windows In theory, the square loss between the minimum prediction value and the observed true value x t+1 can obtain the final prediction model STA-ODE. However, only optimizing the square loss between and x t+1 ignores the alignment problem of the fusion results in the time dimension and the space dimension. Specifically, when the fusion of a single dimension or ) the final prediction result x t+1 will deviate from the true value with a larger probability. In real-world scenarios, whether the fusion result of spatial dimension or the fusion result of temporal dimension both describe the intrinsic characteristics of spatio-temporal data, which makes the fusion result of each dimension should be as close to the true value as possible. Therefore, the alignment problem of the fusion result is integrated into the optimization process of STA-ODE, and the corresponding loss function is shown in equation (10).

[0091]

[0092] where x t+1 represents the spatio-temporal state ground truth of the t+1th time window; represents the spatio-temporal state prediction value of the t+1th time window; represents the fusion value of the t+1th time window in the temporal dimension; represents the fusion value of the t+1th time window in the spatial dimension; represents a function for solving the 2-norm of a vector; Θ represents the learnable parameters in the STA-ODE model; α and β represent the regularization terms, which respectively penalize the deviation of the fusion result of the two dimensions from the true value.

[0093] The STA-ODE model is compared with the first type of model and the second type of model by using three historical spatio-temporal data sets, i.e., traffic volume data, PM2.5 monitoring data and temperature monitoring data. The first type is a pure black-box data-driven model. The second type is a data-driven model based on NODE.

[0094] The traffic volume data set comes from 67 monitoring cameras in Wuhan, China. Figure 6a The spatial distribution of the monitoring cameras is shown. The time span of the traffic volume data set is from March 1, 2021 to March 28, 2021, and the time window size is 5 minutes. Each traffic volume data contains the unique identification of the monitoring camera, the coordinates of the monitoring camera, the monitoring time window and the traffic volume in the time window. The PM2.5 monitoring data comes from 36 air quality monitoring stations in Beijing, China. Figure 6b The spatial distribution of the air quality monitoring stations is shown. The time span of the PM2.5 monitoring data set is from May 1, 2014 to August 31, 2014, and the time window size is 60 minutes. Each PM2.5 data contains the unique identification of the monitoring station, the coordinates of the monitoring station, the monitoring time window and the PM2.5 air content in the time window. The temperature monitoring data set comes from the Copernicus Climate Database, which records the air temperature (time window size is 60 minutes) above 2 meters above the inland water surface. As Figure 6cAs shown, the present application selects 64 grids in Wuhan and its surrounding areas for experiments, and the grid resolution is 0.25°x0.25°. Each temperature data contains the unique identification of the grid, the center point coordinates of the grid, the monitoring time window and the average temperature in the time window.

[0095] The results show that the prediction accuracy of the first type of model is higher than that of the second type of model, but the first type of model has the problem of poor interpretability. The prediction accuracy of the second type of model is still slightly lower than that of the first type of model. The main reason for the above results is that the NODE-based prediction model improves the interpretability of the model at the expense of the prediction accuracy of the model. Specifically, the derivative network of the NODE-based prediction model mostly depends on the time factor, and does not explicitly model the spatial position information into the output of the derivative network. Secondly, the NODE-based prediction model depends on the initial value of the ODE, and it is difficult to find the long-time dependence in the spatio-temporal data. Compared with the benchmark experiment, the STA-ODE model solves the above two shortcomings, and obtains the optimal prediction result while considering the interpretability of the model.

[0096] As shown in Figure 7 In the second aspect of the embodiment of the present application, a geoscience information spatio-temporal prediction device is provided, which can include:

[0097] The abstract module 11 is configured to abstract the monitoring site of the geoscience information into a node in a graph structure.

[0098] The first construction module 12 is configured to construct a spatio-temporal ordinary differential equation module to predict the hidden state of the node.

[0099] The second construction module 13 is configured to construct a spatio-temporal attention module for fusing the hidden state based on a spatio-temporal attention mechanism.

[0100] The fusion module 14 is configured to fuse the spatio-temporal ordinary differential equation module and the spatio-temporal attention module to obtain a geoscience information prediction model.

[0101] The prediction module 15 is configured to substitute the historical spatio-temporal data of the geoscience information into the geoscience information prediction model to obtain the prediction result of the geoscience information.

[0102] The geoscience information space-time prediction apparatus in the embodiments of the present application can be an apparatus, or a component, integrated circuit, or chip in a terminal. The apparatus can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, tablet computer, notebook computer, palm computer, vehicle-mounted electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., and the non-mobile electronic device can be a server, network attached storage (NAS), personal computer (PC), television (TV), teller machine, or self-service machine, etc., and the embodiments of the present application are not limited in this regard.

[0103] The geoscience information space-time prediction apparatus provided in the embodiments of the present application can implement the geoscience information space-time prediction method provided in the above embodiments, and thus the details are not repeated here.

[0104] Optionally, as shown in Figure 8 The embodiments of the present application further provide an electronic device 1100, which includes a processor 1101, a memory 1102, and a program or instruction stored in the memory 1102 and executable on the processor 1101. The program or instruction is executed by the processor 1101 to implement each process of the above geoscience information space-time prediction method or data processing method embodiments, and achieve the same technical effects. The details are not repeated here to avoid repetition.

[0105] It should be noted that the electronic device in the embodiments of the present application includes the above-mentioned mobile electronic device and non-mobile electronic device.

[0106] Figure 9 A hardware structure schematic diagram of an electronic device according to an embodiment of the present application is shown in FIG. 12.

[0107] The electronic device 1200 includes, but is not limited to, a radio frequency unit 1201, a network module 1202, an audio output unit 1203, an input unit 1204, a sensor 1205, a display unit 1206, a user input unit 1207, an interface unit 1208, a memory 1209, and a processor 1210, etc.

[0108] Those skilled in the art can understand that the electronic device 1200 can further include a power supply (such as a battery) for supplying power to each component, and the power supply can be logically connected to the processor 1210 through a power management system, so as to realize the functions of power management, such as charging, discharging, and power consumption management, through the power management system.Figure 9 The electronic device structure shown in the above-mentioned embodiments does not constitute a limitation on the electronic device, which can include more or fewer components than those shown, or combine certain components, or arrange different components, which are not described here again.

[0109] It should be understood that in the embodiments of the present application, the input unit 1204 can include a graphics processor (GPU) 12041 and a microphone 12042. The graphics processor 12041 processes image data of a still picture or a video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 1206 can include a display panel 12061, which can be configured in the form of a liquid crystal display, an organic light-emitting diode, etc. The user input unit 1207 includes a touch panel 12071 and other input devices 12072. The touch panel 12071 is also called a touch screen. The touch panel 12071 can include two parts of a touch detection device and a touch controller. The other input devices 12072 can include, but are not limited to, a physical keyboard, function keys (such as volume control keys, on-off keys, etc.), trackballs, mice, joysticks, etc., which are not described here again. The memory 1209 can be used to store software programs and various data, including but not limited to application programs and operating systems. The processor 1210 can integrate an application processor and a modem processor, wherein the application processor mainly processes operating systems, user interfaces and application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 1210.

[0110] The embodiments of the present application also provide a readable storage medium, which stores programs or instructions, and the programs or instructions are executed by a processor to realize the above-mentioned geoscience information spatio-temporal prediction method and achieve the same technical effects. To avoid repetition, details are not described here again.

[0111] The processor is the processor in the electronic device described in the above-mentioned embodiments. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0112] The embodiments of the present application further provide a chip, which includes a processor and a communication interface. The communication interface is coupled with the processor. The processor is used to run programs or instructions to realize various processes of the above-mentioned geoscience information spatio-temporal prediction method embodiments and achieve the same technical effects. To avoid repetition, details are not described here again.

[0113] It should be understood that the chip mentioned in the embodiments of the present application can also be referred to as a system-level chip, a system chip, a chip system, or a system-on-chip chip, etc.

[0114] It should be noted that in this document, the terms "comprise", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, so that processes, methods, articles, or devices that comprise a list of elements not only include those elements, but also include other elements not explicitly listed, or further include elements inherent in such processes, methods, articles, or devices. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but can also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method can be performed in an order different from that described, and various steps can also be added, omitted, or combined. In addition, the features described with reference to certain examples can be combined in other examples.

[0115] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and the necessary general hardware platform, of course, they can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disc, optical disc), and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, or network equipment, etc.) execute the methods described in various embodiments of the present application.

[0116] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative, not restrictive, and those skilled in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the scope protected by the claims.

Claims

1. A geoscience information spatio-temporal prediction method, characterized in that, The method comprises the steps of: abstracting monitoring sites of geosciences information as nodes in a graph structure; constructing a space-time ordinary differential equation module to predict hidden states of the nodes; constructing a space-time attention module for fusing the hidden states based on a space-time attention mechanism; fusing the space-time ordinary differential equation module and the space-time attention module to obtain a geosciences information prediction model; substituting historical space-time data of the geosciences information into the geosciences information prediction model to obtain a prediction result of the geosciences information; the step of constructing the space-time ordinary differential equation module to predict the hidden states of the nodes comprises the steps of: obtaining a definition of hidden states of the nodes; constructing a space-time derivative network based on time information and space information; constructing the space-time ordinary differential equation module to iteratively solve the hidden states at each time point through the space-time derivative network; the space-time derivative network is mathematically defined as shown in the following formula: In the formula: express Each monitoring station at time The hidden state, where DN ST The spacetime derivative representing the hidden state. Represents the spatiotemporal derivative network. Indicates the first Each monitoring station at time The hidden state, This represents the spatial location information of different monitoring stations, which is further encoded in the spatiotemporal derivative network.

2. The method according to claim 1, wherein, before the step of fusing the space-time ordinary differential equation module and the space-time attention module to obtain the geosciences information prediction model, the method further comprises the step of: introducing a gating mechanism and a residual connection into the space-time ordinary differential equation module.

3. The method according to claim 1, wherein, the step of constructing the space-time attention module for fusing the hidden states based on the space-time attention mechanism comprises the steps of: constructing a spatial attention module for fusing the hidden states in a spatial dimension; constructing a temporal attention module for fusing the hidden states in a temporal dimension; constructing the space-time attention module based on the spatial attention module and the temporal attention module.

4. The method according to claim 1, wherein, before the step of fusing the space-time ordinary differential equation module and the space-time attention module to obtain the geosciences information prediction model, the method further comprises the step of: introducing a residual connection into the space-time attention module.

5. The method according to claim 1, wherein, the step of substituting the historical space-time data of the geosciences information into the geosciences information prediction model to obtain the prediction result of the geosciences information further comprises the step of: optimizing the geosciences information prediction model based on a loss function; the loss function comprises a temporal dimension loss term and a spatial dimension loss term.

6. The method according to claim 1, wherein, The geosciences information comprises any one of the following: traffic flow, PM2.5 concentration and air temperature.

7. A geoscience information spatio-temporal prediction apparatus, characterized in that, The method comprises the steps of: an abstracting module is configured to abstract monitoring sites of geosciences information as nodes in a graph structure; a first constructing module is configured to construct a space-time ordinary differential equation module to predict hidden states of the nodes; a second constructing module is configured to construct a space-time attention module for fusing the hidden states based on a space-time attention mechanism; a fusing module is configured to fuse the space-time ordinary differential equation module and the space-time attention module to obtain a geosciences information prediction model; a prediction module is configured to substitute historical space-time data of the geosciences information into the geosciences information prediction model to obtain a prediction result of the geosciences information; the first constructing module is specifically configured to: obtain a definition of hidden states of the nodes; construct a space-time derivative network based on time information and space information; construct the space-time ordinary differential equation module to iteratively solve the hidden states at each time point through the space-time derivative network; the space-time derivative network is mathematically defined as shown in the following formula: wherein: represents the hidden state of the kth monitoring site at time , wherein DN ST represents the spatiotemporal derivative of the hidden state, represents the spatiotemporal derivative network, represents the hidden state of the kth monitoring site at time , wherein DN represents the hidden state of the kth monitoring site at time represents the spatial location information of different monitoring sites, which is further encoded in the spatiotemporal derivative network.

8. An electronic device, comprising: The method comprises the steps of: A processor, a memory, and a program or instructions stored on the memory and executable on the processor, the program or instructions being executed by the processor to implement the geosciences information spatiotemporal prediction method according to any one of claims 1-6.

9. A readable storage medium, characterized by, A program or instructions stored on the readable storage medium, the program or instructions being executed by the processor to implement the geosciences information spatiotemporal prediction method according to any one of claims 1-6.

Images