Regional rainfall space-time prediction method and system based on multi-source data fusion

Abstract

The invention discloses a regional rainfall space-time prediction method and system based on multi-source data fusion, and belongs to the field of weather prediction.The method comprises the steps that a plurality of current vehicle-mounted data received by all base stations in a region at the current moment are obtained; taking each base station as a node, determining an environment where the corresponding vehicle is located according to the vehicle-mounted data, and fusing the vehicle-mounted data according to the environment to generate an embedded feature of each node; obtaining a wind direction vector collected by a sensor corresponding to each base station in the area at the current moment, generating a connection edge between the nodes according to the wind direction vector, and obtaining a meteorological propagation map corresponding to the current moment; and inputting the meteorological propagation atlas into a preset graph neural network to carry out graph signal transmission and feature aggregation operation so as to update the embedded features of each node, and predicting the rainfall intensity corresponding to the sub-region where each base station is located according to the updated embedded features of each node. The problems of low fineness and poor real-time performance of a regional rainfall intensity prediction result can be solved.

Description

Technical Field

[0001] This application relates to the field of meteorological monitoring, and in particular to a method and system for spatiotemporal prediction of regional precipitation based on multi-source data fusion. Background Technology

[0002] Precipitation forecasting is a key task in the field of meteorological monitoring and disaster early warning. Its results are directly related to the accuracy of decision-making in various application scenarios such as flood control scheduling, urban drainage, traffic safety and agricultural management.

[0003] Current regional precipitation forecasting methods primarily rely on meteorological forecasting internet platforms to collect observational data from various meteorological stations, including conventional meteorological elements such as precipitation, temperature, humidity, air pressure, and wind speed, for weather forecasting. However, due to numerous limitations in the deployment of meteorological stations and their limited observation range, the spatial non-uniformity of localized heavy rainfall means that data collected from these stations cannot accurately reflect precipitation intensity in densely built-up urban areas or mountainous terrain with restricted terrain, thus reducing the spatial resolution of precipitation forecasting internet platforms. Furthermore, the long update intervals of data collected from meteorological stations cannot adequately meet the stringent real-time requirements of rainfall intensity forecasting, impacting the accuracy and timeliness of meteorological forecasting internet platforms.

[0004] Therefore, improving the accuracy and real-time performance of meteorological forecasting internet platforms in predicting regional precipitation intensity during periods of localized heavy rainfall is a technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a method and system for spatiotemporal prediction of regional precipitation based on multi-source data fusion, which can solve the problems of low precision and poor real-time performance of existing meteorological forecasting Internet platforms for predicting regional precipitation intensity when local heavy precipitation occurs.

[0006] This application provides a regional precipitation spatiotemporal prediction method based on multi-source data fusion in some embodiments, applied to a meteorological forecasting internet platform. The regional precipitation spatiotemporal prediction method based on multi-source data fusion includes: Acquire several current vehicle data received by each base station in the area at the current moment; the current vehicle data includes: first wiper frequency, first communication signal strength data, and first satellite positioning; Using each base station as a node, the environment of the corresponding vehicle is determined based on the first satellite positioning, and the first wiper frequency and the first communication signal strength data are fused based on the environment to generate the embedding features of each node; the environment includes: a first interference environment with only communication interference, a second interference environment with only wiper interference, and an interference-free environment; Obtain the wind direction vectors collected by the sensors corresponding to each base station in the region at the current time, generate the connection edges between each node based on the wind direction vectors, and obtain the meteorological propagation map corresponding to the current time. The meteorological propagation map is input into a preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node, and the precipitation intensity of the corresponding sub-region of each base station is predicted based on the updated embedding features of each node. Compared to existing technologies, the above embodiments have the following beneficial effects: With the development of the Internet of Things and smart hardware devices, vehicle-mounted devices can access the meteorological forecasting internet platform through base stations, and receive and analyze vehicle-mounted data through the meteorological forecasting internet platform. Analysis of vehicle-mounted data through the meteorological forecasting internet platform reveals that when precipitation is excessive, dense raindrops can affect the communication quality between the vehicle and the base station. Furthermore, as precipitation increases, drivers adjust their windshield wiper speeds to ensure driving safety. Therefore, the wiper frequency and the strength of the communication signal between the vehicle and the base station reflect the intensity of localized rainfall to some extent. Moreover, in densely built-up urban areas or mountainous terrain where meteorological station deployment is limited, the data collected by vehicles is more extensive and comprehensive than that collected by meteorological stations. Therefore, this application, by introducing vehicle-mounted data, achieves high-density dynamic sampling of precipitation spatial distribution information, solving the problem of insufficient granularity in precipitation intensity prediction based on meteorological station observation data, and providing another feasible path for precipitation intensity prediction. Furthermore, since vehicles are dynamically distributed in different driving environments, such as roads with good signal but rain protection (i.e., the second interference environment) or mountain roads with poor signal but no rain protection (i.e., the first interference environment), the effective data collected from vehicle data varies in different driving environments. Therefore, when generating node embedding features, the vehicle's driving environment is identified, and the vehicle data is fused accordingly to improve the accuracy of the embedding features for each node. Finally, using each base station as a node and combining wind direction vectors, a meteorological propagation map is constructed. This map simulates the impact of wind field on precipitation intensity shifts. Furthermore, graph signal transmission and feature aggregation operations are performed in the graph neural network to accurately capture the propagation and evolution of precipitation intensity across different regions with meteorological factors. This improves the real-time performance and accuracy of precipitation intensity prediction within the region, providing more effective forecasting support for internet platforms with meteorological forecasting functions.

[0007] Furthermore, after acquiring several current vehicle-mounted data received by each base station in the area at the current moment, it also includes: Based on historical precipitation data and historical vehicle data within the area, a first road segment with communication interference and a second road segment with windshield wiper interference within the area are identified respectively. Identify the overlapping road segments between the first road segment and the second road segment; The first current vehicle data where the first satellite positioning is located in the overlapping road segment is filtered out from all the current vehicle data.

[0008] Compared with the prior art, the above embodiments have the following beneficial effects: by using historical precipitation data and historical vehicle data, it is possible to identify the first road segment where the communication signal was still weak when there was no precipitation in the past, and at the same time, it is possible to identify the second road segment where the wipers did not wipe due to rain-blocking factors when there was precipitation in the past. Thus, by superimposing the first road segment and the second road segment, noise data with multi-source interference can be identified, and this part of the data can be filtered out from the current vehicle data, thereby improving the data quality.

[0009] Further, the historical vehicle data includes: second wiper frequency, second communication signal strength data, and second satellite positioning; the step of determining the first road segment with communication interference and the second road segment with wiper interference in the area based on historical precipitation data and historical vehicle data includes: Extract each road segment from the electronic map of the area, and calculate the average value of the second wiper brush frequency and the average value of the second communication signal strength data corresponding to each road segment based on the second satellite positioning. Based on the historical precipitation data, road segments in which the second communication signal strength data is less than a first threshold during non-precipitation periods are selected from all the road segments as the first road segments; Based on the historical precipitation data, road segments in which the average frequency of the second wiper during precipitation is less than a second threshold are selected from all the road segments as the second road segments.

[0010] Compared to existing technologies, the above embodiments have the following beneficial effects: If there is no precipitation during a historical period, theoretically there should be no problem of communication signals being affected by excessive precipitation intensity. If the communication signal is too weak at this time, it reflects to some extent the presence of other factors such as obstructions on the road segment that could affect the communication signal. Based on the above characteristics, the first road segment with communication interference can be accurately screened out. If there is precipitation during a historical period, theoretically the wiper frequency should be at least greater than a certain wiper frequency. However, if the wiper frequency is zero at this time, it reflects to some extent the presence of rain-blocking factors on the road. Based on the above characteristics, the second road segment with wiper interference can be accurately screened out, providing an accurate basis for subsequent noise removal and feature fusion.

[0011] Furthermore, before generating the embedding features corresponding to each of the nodes, the method further includes: Obtain the second current vehicle data collected under the interference-free environment from all the current vehicle data; Based on the first satellite positioning of each of the second current vehicle data, cluster analysis is performed on the spatial distribution of all the second current vehicle data to determine a number of first clusters whose density is greater than a third threshold. The environment in which the second current vehicle data is located in each of the first clusters is set as the first interference environment.

[0012] Compared to existing technologies, the above embodiments have the following beneficial effects: Communication signal quality is affected not only by objective obstructions but also by communication link congestion caused by traffic jams. This is especially true during periods of localized precipitation, when traffic congestion is more likely to occur. In such cases, the wiper frequency data collected from current vehicle data still holds significant value for precipitation intensity prediction. Therefore, it is necessary to accurately identify this portion of data and label it as data collected under the first interference environment to enhance the representation of wiper frequency data in the feature fusion process. First, second current data collected under interference-free conditions is acquired. Based on the spatial clustering characteristics of current vehicle data collected in traffic congestion scenarios, cluster analysis is performed on the spatial distribution characteristics of the first satellite positioning data within each second current vehicle data set. This allows for the selection of second current vehicle data corresponding to vehicle clustering areas, which is then used as data collected under the first interference environment. This improves the accuracy of the feature fusion process for this portion of data, thereby enhancing the accuracy of subsequent precipitation intensity prediction.

[0013] Further, the step of fusing the first wiper frequency and the first communication signal strength data based on the environment to generate the embedding features of each node includes: For the current vehicle data collected under the first interference environment, feature data of the current vehicle data is generated based on the first wiper brush frequency; For the current vehicle data collected under the second interference environment, feature data of the current vehicle data is generated based on the first communication signal strength data; For the current vehicle data collected under interference environment, the first wiper brush frequency and the first communication signal strength data are fused to generate feature data of the current vehicle data. For each base station, the mean value of the feature data corresponding to all current vehicle data received by the base station is used as the embedded feature of the corresponding node of the base station.

[0014] Compared to existing technologies, the above embodiments have the following beneficial effects: Since there is no wiper interference in the first interference environment, the first wiper frequency in the current vehicle data collected under the first interference environment still has high predictive value. Therefore, only the first wiper frequency is retained as the data used in subsequent feature data generation, thereby strengthening the influence of effective data under the first interference environment on the final prediction result and improving the accuracy of the final embedded feature generation. Similarly, since there is no communication interference in the second interference environment, only the first communication signal strength data is retained as the data used in subsequent feature data generation, thereby improving the accuracy of the final embedded feature generation.

[0015] Further, generating the connection edges between the nodes based on the wind direction vector includes: Generate connecting edges between the nodes, and for each node, calculate the projection of the wind direction vector corresponding to the node onto each connecting edge of the node to obtain the sub-vector corresponding to each connecting edge of the node; wherein, each connecting edge corresponds to two sub-vectors, and the starting points of the two sub-vectors correspond to the two nodes of the connecting edge respectively. For each of the connecting edges, the lengths of the two sub-vectors corresponding to the connecting edge are calculated, and the direction of the connecting edge is determined based on the directions of the two sub-vectors; For each connecting edge, if the directions of the two corresponding sub-vectors are opposite, the first weight of the connecting edge is determined according to the difference in length of the two corresponding sub-vectors; otherwise, the first weight of the connecting edge is determined according to the maximum value of the length of the two corresponding sub-vectors. Remove all connected edges whose weight is less than the fourth threshold from all the connected edges.

[0016] Compared to existing technologies, the above embodiments have the following beneficial effects: The wind direction vector includes a wind speed scalar and a wind direction vector. The wind speed scalar reflects the influence of the wind field on the speed of precipitation diffusion, while the wind direction vector reflects the influence of the wind direction on the direction of precipitation diffusion. Therefore, in order to more comprehensively simulate the influence of the wind field on precipitation diffusion, a connection edge is first established between each node. Based on the wind direction vector collected by each node, projection processing is performed with that node as the starting point to obtain the sub-vector of the wind direction vector on each connection edge. Since each connection edge corresponds to two nodes, each of the two nodes corresponding to each connection edge has a corresponding sub-vector on that connection edge. When the directions of the two sub-vectors are the same, it means that the wind field at the two nodes has the same influence on the direction of precipitation diffusion. Therefore, the strongest wind speed among the two nodes is selected as the first weight of the connection edge. When the directions of the two sub-vectors are different, it means that the wind field at the two nodes has a repulsive influence on the direction of precipitation diffusion. Therefore, the difference in wind speed between the two is selected to simulate the cancellation effect of two opposing wind fields, thereby improving the accuracy of the first weight setting of the connection edge.

[0017] Furthermore, the graph neural network includes multiple neural network layers; the step of inputting the meteorological propagation map into the preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node includes: Based on the direction of each connecting edge, determine several adjacent nodes corresponding to each node when it is the target node; The meteorological propagation map is sequentially input into each of the neural network layers. After each input into the neural network layer, for each node in the meteorological propagation map, the embedding features of each of the node's corresponding neighboring nodes are fused according to the first weight of the connection edge between the node and each of its corresponding neighboring nodes, and the embedding features of the node are updated according to the fused embedding features.

[0018] Compared to existing technologies, the above embodiments have the following advantages: Existing technologies often employ Long Short-Term Memory (LSTM) networks to learn the changing patterns of meteorological time series in order to improve the accuracy of weather forecasts. However, these methods require long data inputs and time-step calculations, making parallel computing impossible. Existing internet platforms are mostly cloud platforms with high parallel computing capabilities, thus these methods waste the computing power of these platforms. In contrast, this application constructs a meteorological propagation map including spatial information and uses graph neural networks for information propagation. Neighborhood information between nodes can be aggregated in parallel, resulting in higher computational efficiency. Furthermore, this application does not require the collection of large amounts of time series data; it only needs the data collected at the current moment to infer the data changes at the next moment based on the spatial structure, effectively reducing the amount of data processing and improving the real-time performance of forecasts.

[0019] Another embodiment of this application provides a regional precipitation spatiotemporal prediction system based on multi-source data fusion, which is applied to a meteorological forecasting internet platform. The regional precipitation spatiotemporal prediction system based on multi-source data fusion includes: a data acquisition module, a node construction module, a connection edge construction module, and a graph propagation module. The data acquisition module is used to acquire several current vehicle data received by each base station in the area at the current moment; the current vehicle data includes: a first wiper brush frequency, a first communication signal strength data, and a first satellite positioning. The node construction module is used to use each base station as a node, determine the environment of the corresponding vehicle based on the first satellite positioning, and fuse the first wiper frequency and the first communication signal strength data based on the environment to generate the embedding features of each node; the environment includes: a first interference environment with only communication interference, a second interference environment with only wiper interference, and an interference-free environment; The connection edge construction module is used to obtain the wind direction vector collected by the corresponding sensor of each base station in the region at the current time, generate the connection edge between each node according to the wind direction vector, and obtain the meteorological propagation map corresponding to the current time. The graph propagation module is used to input the meteorological propagation map into a preset graph neural network for graph signal transmission and feature aggregation operations, so as to update the embedding features of each node, and predict the precipitation intensity of the corresponding sub-region of each base station based on the updated embedding features of each node. Another embodiment of this application also provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps of the regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in this application.

[0020] Another embodiment of this application also provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the regional precipitation spatiotemporal prediction method based on multi-source data fusion of this application. Attached Figure Description

[0021] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1This is a flowchart illustrating a regional precipitation spatiotemporal prediction method based on multi-source data fusion provided in some embodiments of this application; Figure 2 This is a schematic diagram of a weather forecasting internet platform architecture provided in some embodiments of this application; Figure 3 This is a schematic diagram of the structure of a regional precipitation spatiotemporal prediction system based on multi-source data fusion provided in some embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0025] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0026] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0027] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0028] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0029] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0030] Current regional precipitation forecasting methods primarily rely on meteorological forecasting internet platforms to collect observational data from various meteorological stations, including conventional meteorological elements such as precipitation, temperature, humidity, air pressure, and wind speed, for weather forecasting. However, due to numerous limitations in the deployment of meteorological stations and their limited observation range, the spatial non-uniformity of localized heavy rainfall can prevent the data collected from these stations from accurately reflecting precipitation intensity in densely built-up urban areas or mountainous terrain with restricted terrain. This reduces the spatial resolution of precipitation forecasting by the meteorological forecasting internet platform. Furthermore, the long data update intervals from meteorological stations cannot adequately meet the stringent real-time requirements of rainfall intensity forecasting, impacting the accuracy and timeliness of the meteorological forecasting internet platform. It should be noted that existing meteorological stations generally need to be installed in flat, open, and unobstructed areas, ensuring that there are no tall buildings or dense forests or other potential obstacles within a predetermined distance. Densely built-up urban areas, as mentioned here, refer to urban areas with numerous tall buildings along roadsides. Due to the high number and density of these buildings, it is difficult to meet the installation requirements for meteorological stations.

[0031] Please refer to Figure 1 To address the issues of low precision and poor real-time performance in existing meteorological forecasting internet platforms for predicting regional precipitation intensity during periods of localized heavy rainfall, this application provides a regional precipitation spatiotemporal prediction method based on multi-source data fusion, applicable to meteorological forecasting internet platforms. (Reference) Figure 2 This is an architecture of a weather forecasting internet platform provided in some embodiments of this application, wherein the base station receives the current vehicle data sent by each vehicle, and then the base station uploads the received current vehicle data to the weather forecasting internet platform for execution of the regional precipitation spatiotemporal forecasting method based on multi-source data fusion provided in the embodiments of this application.

[0032] Furthermore, the regional precipitation spatiotemporal prediction method based on multi-source data fusion includes S101 to S104, specifically as follows: S101: Obtain several current vehicle data received by each base station in the area at the current moment; the current vehicle data includes: first wiper brush frequency, first communication signal strength data, and first satellite positioning.

[0033] Preferably, in some embodiments of this application, the first communication signal strength data is determined and obtained by the vehicle-mounted terminal through the vehicle-mounted communication module by measuring the communication delay between the vehicle and the base station; the first satellite positioning is collected and obtained by the vehicle-mounted terminal through the satellite navigation module.

[0034] Preferably, in some embodiments of this application, after receiving the first communication signal strength data, it is also necessary to preprocess it, including: taking the maximum and minimum values ​​of the first communication signal strength data, and subtracting the minimum value from the currently collected first communication signal strength data and dividing by the difference between the maximum and minimum values ​​to obtain the normalized first communication signal strength data.

[0035] Furthermore, after acquiring several current vehicle-mounted data received by each base station in the area at the current moment, it also includes: Based on historical precipitation data and historical vehicle data within the area, a first road segment with communication interference and a second road segment with windshield wiper interference within the area are identified respectively. Identify the overlapping road segments between the first road segment and the second road segment; The first current vehicle data where the first satellite positioning is located in the overlapping road segment is filtered out from all the current vehicle data.

[0036] Preferably, in some embodiments of this application, historical precipitation data includes the actual precipitation intensity corresponding to each road segment at a historical time point.

[0037] It should be noted that communication interference refers to factors that affect communication signals, such as mountain tunnels, underground passages, or mountain roads; windshield wiper interference refers to factors that affect rainwater falling on vehicles, such as mountain tunnels, underground passages, tree-lined roads, or rain shelters.

[0038] It is further understood that communication interference and wiper interference in the embodiments of this application can coexist in one environment, or only one type of interference can exist in a certain environment. For example, when a vehicle is driving on a valley road without rain shelter, the vehicle will still be affected by the communication signal due to rocks. In this case, only communication interference exists, but no wiper interference exists, which is the first interference environment. When a vehicle is driving on a tree-lined urban road with rain shelter, the communication signal on the road is good, but the rainfall will be blocked by the trees. In this case, only wiper interference exists, but no communication interference exists, which is the second interference environment. When a vehicle is driving in an environment such as a tunnel, the tunnel not only affects the vehicle's communication signal, but also blocks the rainfall. In this case, both communication interference and wiper interference exist. An interference-free environment means that neither communication interference nor wiper interference exists.

[0039] This application uses historical precipitation data and historical vehicle data to identify the first road segment where communication signals were still weak when there was no precipitation in the past. It can also identify the second road segment where the windshield wipers did not operate due to rain cover when there was precipitation in the past. By superimposing the first road segment and the second road segment, noise data with multi-source interference can be identified and filtered out from the current vehicle data, thereby improving data quality.

[0040] Furthermore, in some embodiments of this application, the historical vehicle data includes: a second wiper frequency, second communication signal strength data, and second satellite positioning; the step of determining the first road segment with communication interference and the second road segment with wiper interference in the area based on historical precipitation data and historical vehicle data includes: Extract each road segment from the electronic map of the area, and calculate the average value of the second wiper brush frequency and the average value of the second communication signal strength data corresponding to each road segment based on the second satellite positioning. Based on the historical precipitation data, road segments in which the second communication signal strength data is less than a first threshold during non-precipitation periods are selected from all the road segments as the first road segments; Based on the historical precipitation data, road segments in which the average frequency of the second wiper during precipitation is less than a second threshold are selected from all the road segments as the second road segments.

[0041] Preferably, in some embodiments of this application, the first threshold can be set and obtained based on the average value of communication signal strength data collected from unobstructed urban road segments during periods of non-rainfall intensity.

[0042] Preferably, in some embodiments of this application, the second threshold can be set to the minimum wiper frequency among all vehicle models.

[0043] Understandably, if there is no precipitation during a historical period, theoretically there shouldn't be any issue with communication signals being affected by excessive rainfall. However, if the communication signal is weak at this time, it reflects, to some extent, the presence of other obstructions or other factors on that road segment that could affect the signal. Based on these characteristics, the first road segment with communication interference can be accurately identified. If there was precipitation during a historical period, theoretically, the windshield wiper frequency should be at least higher than a certain frequency. However, if the wiper frequency is zero, it reflects, to some extent, the presence of rain-blocking factors on the road. Based on these characteristics, the second road segment with wiper interference can be accurately identified, providing an accurate basis for subsequent noise removal and feature fusion.

[0044] S102: Using each base station as a node, determine the environment of the corresponding vehicle based on the first satellite positioning, and fuse the first wiper brush frequency and the first communication signal strength data based on the environment to generate the embedding features of each node; the environment includes: a first interference environment with only communication interference, a second interference environment with only wiper interference, and an interference-free environment.

[0045] Preferably, in some embodiments of this application, the first interference environment is determined in the following way: A first road segment with communication interference can be obtained through the above embodiments. At this time, some road segments within the first road segment are also affected by windshield wiper interference, and these road segments need to be filtered out. Furthermore, the above embodiments also obtain overlapping road segments between the first and second road segments. Therefore, by filtering out the overlapping road segments in the first road segment, a third road segment with only communication interference can be obtained. The environment in which the third road segment is located is then taken as the first interference environment.

[0046] Preferably, in some embodiments of this application, the second interference environment is determined in the following way: similar to the way the first interference environment is determined, since a second road segment with wiper interference can be obtained through the above embodiment, and there are also some road segments in the second road segment that are simultaneously interfered with by communication, the overlapping road segments in the second road segment are filtered out to obtain a fourth road segment with only wiper interference. At this time, the environment in which the fourth road segment is located is taken as the second interference environment.

[0047] Furthermore, in some embodiments of this application, before generating the embedding features corresponding to each of the nodes, the method further includes: Obtain the second current vehicle data collected under the interference-free environment from all the current vehicle data; Based on the first satellite positioning of each of the second current vehicle data, cluster analysis is performed on the spatial distribution of all the second current vehicle data to determine a number of first clusters whose density is greater than a third threshold. The environment in which the second current vehicle data is located in each of the first clusters is set as the first interference environment.

[0048] Preferably, in some embodiments of this application, the density of the cluster refers to the ratio of the number of second current vehicle data in the cluster to the area of ​​the cluster.

[0049] Preferably, in some embodiments of this application, the third threshold can be determined based on the minimum vehicle density when traffic congestion occurs in each road segment of the traffic monitoring system.

[0050] Communication signal quality is affected not only by objective obstructions but also by communication link congestion caused by traffic jams, especially during periods of localized precipitation. In such cases, the wiper frequency data collected from current vehicle data remains valuable for precipitation intensity prediction. Therefore, it is necessary to accurately identify this data and label it as data collected under the first interference environment to enhance the representation of wiper frequency data in the feature fusion process. First, second current data collected under interference-free conditions is acquired. Given the spatial clustering characteristics of current vehicle data collected in traffic congestion scenarios, cluster analysis is used to analyze the spatial distribution characteristics of the first satellite positioning within each second current vehicle data set. This allows for the selection of second current vehicle data corresponding to vehicle clustering areas, which is then used as data collected under the first interference environment. This improves the accuracy of the feature fusion process for this data, thereby enhancing the accuracy of subsequent precipitation intensity prediction.

[0051] Furthermore, in some embodiments of this application, the step of fusing the first wiper frequency and the first communication signal strength data based on the environment to generate the embedding features of each node includes: For the current vehicle data collected under the first interference environment, feature data of the current vehicle data is generated based on the first wiper brush frequency; For the current vehicle data collected under the second interference environment, feature data of the current vehicle data is generated based on the first communication signal strength data; For the current vehicle data collected under interference environment, the first wiper brush frequency and the first communication signal strength data are fused to generate feature data of the current vehicle data. For each base station, the mean value of the feature data corresponding to all current vehicle data received by the base station is used as the embedded feature of the corresponding node of the base station.

[0052] Preferably, in some embodiments of this application, since the first wiper frequency and the first communication signal strength data are not physical quantities of the same dimension, when generating feature data, it is necessary to convert the data of the two different physical quantities to the same dimension. The methods that can be used include, but are not limited to: normalizing the maximum and minimum values ​​to the same range or dimensionless conversion based on Z-score standardization.

[0053] After converting the first wiper frequency and the first communication signal strength data to the same dimension, for the first interference environment, the first wiper frequency converted to the same dimension is used as the feature data of the current vehicle data; for the second interference environment, the first communication signal strength data converted to the same dimension is used as the feature data of the current vehicle data; for the interference environment, the first communication signal strength data converted to the same dimension and the first wiper frequency can be directly fused with a weight of 0.5, or the fusion weight between the first communication signal strength data and the first wiper frequency in the interference-free environment can be dynamically adjusted by using a feedforward neural network and introducing an attention mechanism. This application does not limit the fusion process.

[0054] Since there is no wiper interference in the first interference environment, the first wiper frequency in the current vehicle data collected under this environment still has high predictive value. Therefore, only the first wiper frequency is retained as the data used in subsequent feature data generation, thereby strengthening the influence of effective data under the first interference environment on the final prediction result and improving the accuracy of the final embedded feature generation. Similarly, since there is no communication interference in the second interference environment, only the first communication signal strength data is retained as the data used in subsequent feature data generation, thereby improving the accuracy of the final embedded feature generation.

[0055] S103: Obtain the wind direction vector collected by the corresponding sensor of each base station in the area at the current time, generate the connection edge between each node according to the wind direction vector, and obtain the meteorological propagation map corresponding to the current time.

[0056] Furthermore, in some embodiments of this application, generating the connection edges between the nodes based on the wind direction vector includes: Generate connecting edges between the nodes, and for each node, calculate the projection of the wind direction vector corresponding to the node onto each connecting edge of the node to obtain the sub-vector corresponding to each connecting edge of the node; wherein, each connecting edge corresponds to two sub-vectors, and the starting points of the two sub-vectors correspond to the two nodes of the connecting edge respectively. For each of the connecting edges, the lengths of the two sub-vectors corresponding to the connecting edge are calculated, and the direction of the connecting edge is determined based on the directions of the two sub-vectors; For each connecting edge, if the directions of the two corresponding sub-vectors are opposite, the first weight of the connecting edge is determined according to the difference in length of the two corresponding sub-vectors; otherwise, the first weight of the connecting edge is determined according to the maximum value of the length of the two corresponding sub-vectors. Remove all connected edges whose weight is less than the fourth threshold from all the connected edges.

[0057] Preferably, in some embodiments of this application, the process of determining the fourth threshold includes: calculating the average distance between all two base stations; determining the base station combinations whose distance between base stations is less than the average distance, and calculating the average weight of the first weight of the connection edge between the two nodes corresponding to all base station combinations; and using the average weight as the fourth threshold.

[0058] Preferably, in some embodiments of this application, the wind direction vector includes a wind speed scalar and a wind direction vector, wherein the wind speed scalar reflects the influence of the wind field on the speed of precipitation diffusion, and the wind direction vector reflects the influence of the wind direction on the direction of precipitation diffusion.

[0059] To more comprehensively simulate the impact of wind fields on precipitation diffusion, this application first establishes connecting edges between each node. Based on the wind direction vector collected from each node, projection processing is performed with that node as the starting point to obtain a sub-vector of the wind direction vector on each connecting edge. Since each connecting edge corresponds to two nodes, each node corresponding to each connecting edge has a corresponding sub-vector on that connecting edge. When the directions of the two sub-vectors are the same, it means that the wind fields at the two nodes have the same impact on the direction of precipitation diffusion; therefore, the strongest wind speed between the two nodes is selected as the first weight of the connecting edge. When the directions of the two sub-vectors are different, it means that the wind fields at the two nodes have a repulsive impact on the direction of precipitation diffusion; therefore, the difference in wind speed between the two is selected to simulate the canceling effect of two opposing wind fields, thereby improving the accuracy of the first weight setting of the connecting edge.

[0060] S104: Input the meteorological propagation map into a preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node, and predict the precipitation intensity of the corresponding sub-region of each base station based on the updated embedding features of each node. Furthermore, in some embodiments of this application, the graph neural network includes multiple neural network layers; the step of inputting the meteorological propagation map into a preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node includes: Based on the direction of each connecting edge, determine several adjacent nodes corresponding to each node when it is the target node; The meteorological propagation map is sequentially input into each of the neural network layers. After each input into the neural network layer, for each node in the meteorological propagation map, the embedding features of each of the node's corresponding neighboring nodes are fused according to the first weight of the connection edge between the node and each of its corresponding neighboring nodes, and the embedding features of the node are updated according to the fused embedding features.

[0061] Preferably, in some embodiments of this application, after each neural network layer completes the graph signal transmission and feature aggregation operation, the updated node embedding features are used as the input of the next neural network layer, and the connection edges between nodes are kept unchanged. The graph signal transmission and feature aggregation operation corresponding to the next neural network layer is then performed until the last neural network layer outputs the finally updated node embedding features.

[0062] Preferably, in some embodiments of this application, the step of predicting the precipitation intensity of the corresponding sub-region of each base station based on the updated embedding features of each node includes: inputting the embedding features of each node into a support vector machine (SVM), and outputting the precipitation intensity through the SVM; wherein the SVM is trained and obtained based on historical precipitation data and historical vehicle data. If, during the training process, the feedforward neural network corresponding to the first weight combination and the second weight combination is used as the input layer of the SVM, that is, the wiper frequency and communication signal strength data in the vehicle data are fused through the weight combination generated by the feedforward neural network and then input into the SVM. For vehicle data in an interference-free environment, data fusion is performed using a fixed third weight combination.

[0063] To improve the accuracy of weather forecasts, existing technologies often employ Long Short-Term Memory (LSTM) networks to learn the changing patterns of meteorological time series. However, these methods require long data inputs and involve time-step calculations, making parallel computing impossible. Existing internet platforms are mostly cloud platforms with high parallel computing capabilities, thus these methods waste the computing power of these platforms. This application, however, constructs a meteorological propagation map including spatial information and uses graph neural networks for information propagation. Neighborhood information between nodes can be aggregated in parallel, resulting in higher computational efficiency. Furthermore, this application does not require collecting large amounts of time series data; it only needs the data collected at the current moment to infer the data changes at the next moment based on the spatial structure, effectively reducing the data processing load and improving the real-time performance of forecasts.

[0064] In summary, the regional precipitation spatiotemporal prediction method based on multi-source data fusion provided in this application has the following advantages compared to existing technologies: With the development of the Internet of Things and smart hardware devices, vehicle-mounted devices can access the meteorological forecasting internet platform through base stations, receiving and analyzing vehicle-mounted data through the platform. Analysis of vehicle-mounted data through the meteorological forecasting internet platform reveals that when precipitation is heavy, dense raindrops can affect the communication quality between the vehicle and the base station. Furthermore, as precipitation increases, drivers adjust their windshield wiper speeds to ensure driving safety. Therefore, the wiper frequency and the strength of the communication signal between the vehicle and the base station reflect the intensity of local rainfall to some extent. Moreover, in densely built-up urban areas or mountainous terrain where meteorological station deployment is limited, the data collected by vehicles is more extensive and comprehensive than that collected by meteorological stations. Therefore, this application, by introducing vehicle-mounted data, achieves high-density dynamic sampling of precipitation spatial distribution information, solving the problem of insufficient granularity in precipitation intensity prediction based on meteorological station observation data, and providing another feasible path for precipitation intensity prediction. Furthermore, since vehicles are dynamically distributed in different driving environments, such as roads with good signal but rain protection (i.e., the second interference environment) or mountain roads with poor signal but no rain protection (i.e., the first interference environment), the effective data collected from vehicle data varies in different driving environments. Therefore, when generating node embedding features, the vehicle's driving environment is identified, and the vehicle data is fused accordingly to improve the accuracy of the embedding features for each node. Finally, using each base station as a node and combining wind direction vectors, a meteorological propagation map is constructed. This map simulates the impact of wind field on precipitation intensity shifts. Furthermore, graph signal transmission and feature aggregation operations are performed in the graph neural network to accurately capture the propagation and evolution of precipitation intensity across different regions with meteorological factors. This improves the real-time performance and accuracy of precipitation intensity prediction within the region, providing more effective forecasting support for internet platforms with meteorological forecasting functions.

[0065] like Figure 3 As shown, based on the above method embodiments, this application provides a regional precipitation spatiotemporal prediction system based on multi-source data fusion, which is applied to a meteorological prediction Internet platform. The regional precipitation spatiotemporal prediction system based on multi-source data fusion includes: a data acquisition module 201, a node construction module 202, a connection edge construction module 203, and a graph propagation module 204.

[0066] Further, in some embodiments of this application, the data acquisition module 201 is used to acquire several current vehicle data received by each base station in the area at the current time; the current vehicle data includes: a first wiper brush frequency, a first communication signal strength data, and a first satellite positioning; the node construction module 202 is used to use each base station as a node, determine the environment of the corresponding vehicle based on the first satellite positioning, and fuse the first wiper brush frequency and the first communication signal strength data based on the environment to generate the embedding features of each node; the environment includes: a first interference where only communication interference exists. The system includes three environments: a disturbed environment, a second interference environment with only wiper interference, and an interference-free environment. The connection edge construction module 203 is used to obtain the wind direction vector collected by the corresponding sensors of each base station in the region at the current moment, generate connection edges between each node based on the wind direction vector, and obtain the meteorological propagation map corresponding to the current moment. The graph propagation module 204 is used to input the meteorological propagation map into a preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node, and predict the precipitation intensity of the corresponding sub-region of each base station based on the updated embedding features of each node. Furthermore, in some embodiments of this application, the regional precipitation spatiotemporal prediction system based on multi-source data fusion further includes: a first data filtering module; the first data filtering module includes: a road segment determination unit, a road segment overlap unit, and a first current vehicle data filtering unit; the road segment determination unit is used to determine, based on historical precipitation data and historical vehicle data within the region, a first road segment with communication interference and a second road segment with windshield wiper interference within the region; the road segment overlap unit is used to determine the overlapping road segment between the first road segment and the second road segment; the first current vehicle data filtering unit is used to filter out the first current vehicle data whose first satellite positioning is located in the overlapping road segment from all the current vehicle data.

[0067] Further, in some embodiments of this application, the historical vehicle data includes: a second wiper frequency, second communication signal strength data, and second satellite positioning; the road segment determination unit is used to determine, based on historical precipitation data and historical vehicle data within the area, a first road segment with communication interference and a second road segment with wiper interference within the area, respectively, including: extracting each road segment from the electronic map of the area, and calculating, based on the second satellite positioning, the average value of the second wiper frequency and the average value of the second communication signal strength data corresponding to each road segment; selecting, based on the historical precipitation data, road segments from all the road segments whose second communication signal strength data is less than a first threshold during non-precipitation periods as the first road segment; and selecting, based on the historical precipitation data, road segments from all the road segments whose average second wiper frequency is less than a second threshold during precipitation periods as the second road segment.

[0068] Furthermore, in some embodiments of this application, the regional precipitation spatiotemporal prediction system based on multi-source data fusion further includes: a second data filtering module; the second data filtering module includes: a second current vehicle-mounted data filtering unit, a clustering analysis unit, and an environmental labeling unit; the second current vehicle-mounted data filtering unit is used to obtain the second current vehicle-mounted data collected in the interference-free environment from all the current vehicle-mounted data; the clustering analysis unit is used to perform clustering analysis on the spatial distribution of all the second current vehicle-mounted data according to the first satellite positioning of each second current vehicle-mounted data, and determine a number of first clusters whose density is greater than a third threshold; the noise removal unit is used to set the environment in which the second current vehicle-mounted data in each of the first clusters is located as a first interference environment.

[0069] Further, in some embodiments of this application, the node construction module 202 includes: a first feature data generation unit, a second feature data generation unit, a third feature data generation unit, and an embedded feature generation unit; the node construction module 202 is used to fuse the first wiper brush frequency and the first communication signal strength data according to the environment to generate embedded features of each node, including: the first feature data generation unit is used to generate feature data of the current vehicle data collected under the first interference environment based on the first wiper brush frequency; the second feature data generation unit is used to generate feature data of the current vehicle data collected under the second interference environment based on the first communication signal strength data; the third feature data generation unit is used to fuse the first wiper brush frequency and the first communication signal strength data of the current vehicle data collected under the interference environment to generate feature data of the current vehicle data; the embedded feature generation unit is used to take the average value of the feature data corresponding to all current vehicle data received by each base station as the embedded feature of the node corresponding to that base station.

[0070] Further, in some embodiments of this application, the connection edge construction module 203 includes: a wind direction vector decomposition unit, a connection edge direction determination unit, a connection edge weight determination unit, and a connection edge filtering unit; the connection edge construction module 203 is used to generate connection edges between each of the nodes based on the wind direction vector, including: the wind direction vector decomposition unit is used to generate connection edges between each of the nodes, and for each node, calculate the projection of the wind direction vector corresponding to the node onto each connection edge of the node, to obtain the sub-vector corresponding to each connection edge of the node; wherein, each connection edge corresponds to two sub-vectors, and the starting points of the two sub-vectors respectively correspond to the connection edge. The two nodes of the connecting edge; the connecting edge direction determination unit is used to calculate the lengths of the two sub-vectors corresponding to each connecting edge, and determine the direction of the connecting edge according to the directions of the two sub-vectors; the connecting edge weight determination unit is used to determine the first weight of the connecting edge according to the difference between the lengths of the two sub-vectors corresponding to each connecting edge if the directions of the two sub-vectors corresponding to the connecting edge are opposite, otherwise determine the first weight of the connecting edge according to the maximum value of the lengths of the two sub-vectors corresponding to the connecting edge; the connecting edge filtering unit is used to filter out connecting edges whose first weight is less than a fourth threshold from all the connecting edges.

[0071] Further, in some embodiments of this application, the graph neural network includes multiple neural network layers; the graph propagation module 204 includes: an adjacency node determination unit and a propagation unit; the graph propagation module 204 is used to input the meteorological propagation map into a preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node, including: the adjacency node determination unit is used to determine several adjacency nodes corresponding to each node when it is a target node according to the direction of each connection edge; the propagation unit is used to sequentially input the meteorological propagation map into each neural network layer, and after each input into the neural network layer, for each node in the meteorological propagation map, according to the first weight of the connection edge between the node and each of its corresponding adjacency nodes, fuse the embedding features of each adjacency node corresponding to the node, and update the embedding features of the node according to the fused embedding features.

[0072] It is understood that the above system embodiments correspond to the method embodiments of this application, and can realize the regional precipitation spatiotemporal prediction method based on multi-source data fusion provided by any of the above method embodiments of this application.

[0073] In summary, the regional precipitation spatiotemporal prediction system based on multi-source data fusion provided in this application has the following advantages compared to existing technologies: With the development of the Internet of Things and smart hardware devices, vehicle-mounted devices can access the meteorological forecasting internet platform through base stations, and receive and analyze vehicle-mounted data through the meteorological forecasting internet platform. Analysis of vehicle-mounted data through the meteorological forecasting internet platform reveals that when precipitation is excessive, dense raindrops can affect the communication quality between the vehicle and the base station. Furthermore, as precipitation increases, drivers adjust the windshield wiper speed to ensure driving safety. Therefore, the wiper frequency and the strength of the communication signal between the vehicle and the base station reflect the intensity of local rainfall to some extent. Moreover, in densely built-up urban areas or mountainous terrain where meteorological station deployment is limited, the data collected by vehicles is from a wider and more comprehensive range of sources compared to meteorological stations. Therefore, by introducing vehicle-mounted data, this application achieves high-density dynamic sampling of precipitation spatial distribution information, solving the problem of insufficient granularity in precipitation intensity prediction based on meteorological station observation data, and providing another feasible path for precipitation intensity prediction. Furthermore, since vehicles are dynamically distributed in different driving environments, such as roads with good signal but rain protection (i.e., the second interference environment) or mountain roads with poor signal but no rain protection (i.e., the first interference environment), the effective data collected from vehicle data varies in different driving environments. Therefore, when generating node embedding features, the vehicle's driving environment is identified, and the vehicle data is fused accordingly to improve the accuracy of the embedding features for each node. Finally, using each base station as a node and combining wind direction vectors, a meteorological propagation map is constructed. This map simulates the impact of wind field on precipitation intensity shifts. Furthermore, graph signal transmission and feature aggregation operations are performed in the graph neural network to accurately capture the propagation and evolution of precipitation intensity across different regions with meteorological factors. This improves the real-time performance and accuracy of precipitation intensity prediction within the region, providing more effective forecasting support for internet platforms with meteorological forecasting functions.

[0074] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0075] Based on the above embodiments of the regional precipitation spatiotemporal prediction method based on multi-source data fusion, another embodiment of this application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the regional precipitation spatiotemporal prediction method based on multi-source data fusion of any embodiment of this application.

[0076] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete this application. The one or more module units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0077] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0078] The processor can be a Central Processing Unit (CPU), or 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. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0079] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in any of the above-described method embodiments of this application.

[0080] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented 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 files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

Claims

1. A method for spatiotemporal prediction of regional precipitation based on multi-source data fusion, characterized in that, The regional precipitation spatiotemporal prediction method based on multi-source data fusion, applied to an internet-based meteorological forecasting platform, includes: Acquire several current vehicle data received by each base station in the area at the current moment; the current vehicle data includes: first wiper frequency, first communication signal strength data, and first satellite positioning; Using each base station as a node, the environment of the corresponding vehicle is determined based on the first satellite positioning, and the first wiper frequency and the first communication signal strength data are fused based on the environment to generate the embedding features of each node; the environment includes: a first interference environment with only communication interference, a second interference environment with only wiper interference, and an interference-free environment; Obtain the wind direction vectors collected by the sensors corresponding to each base station in the region at the current time, generate the connection edges between each node based on the wind direction vectors, and obtain the meteorological propagation map corresponding to the current time. The meteorological propagation map is input into a preset graph neural network for graph signal transmission and feature aggregation operations to update the embedding features of each node, and the precipitation intensity of the corresponding sub-region of each base station is predicted based on the updated embedding features of each node.

2. The regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in claim 1, characterized in that, After acquiring several current vehicle-mounted data received by each base station in the area at the current moment, it also includes: Based on historical precipitation data and historical vehicle data within the area, a first road segment with communication interference and a second road segment with windshield wiper interference within the area are identified respectively. Identify the overlapping road segments between the first road segment and the second road segment; The first current vehicle data where the first satellite positioning is located in the overlapping road segment is filtered out from all the current vehicle data.

3. The regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in claim 2, characterized in that, The historical vehicle data includes: second wiper frequency, second communication signal strength data, and second satellite positioning; the step of determining the first road segment with communication interference and the second road segment with wiper interference in the area based on historical precipitation data and historical vehicle data includes: Extract each road segment from the electronic map of the area, and calculate the average value of the second wiper brush frequency and the average value of the second communication signal strength data corresponding to each road segment based on the second satellite positioning. Based on the historical precipitation data, road segments in which the second communication signal strength data is less than a first threshold during non-precipitation periods are selected from all the road segments as the first road segments; Based on the historical precipitation data, road segments in which the average frequency of the second wiper during precipitation is less than a second threshold are selected from all the road segments as the second road segments.

4. The regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in claim 1, characterized in that, Before generating the embedding features corresponding to each of the nodes, the process further includes: Obtain the second current vehicle data collected under the interference-free environment from all the current vehicle data; Based on the first satellite positioning of each of the second current vehicle data, cluster analysis is performed on the spatial distribution of all the second current vehicle data to determine a number of first clusters whose density is greater than a third threshold. The environment in which the second current vehicle data is located in each of the first clusters is set as the first interference environment.

5. The regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in claim 4, characterized in that, The step of fusing the first wiper frequency and the first communication signal strength data based on the environment to generate the embedding features of each node includes: For the current vehicle data collected under the first interference environment, feature data of the current vehicle data is generated based on the first wiper brush frequency; For the current vehicle data collected under the second interference environment, feature data of the current vehicle data is generated based on the first communication signal strength data; For the current vehicle data collected under interference environment, the first wiper brush frequency and the first communication signal strength data are fused to generate feature data of the current vehicle data. For each base station, the mean value of the feature data corresponding to all current vehicle data received by the base station is used as the embedded feature of the corresponding node of the base station.

6. The regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in claim 1, characterized in that, The step of generating connection edges between the nodes based on the wind direction vector includes: Generate connecting edges between the nodes, and for each node, calculate the projection of the wind direction vector corresponding to the node onto each connecting edge of the node to obtain the sub-vector corresponding to each connecting edge of the node; wherein, each connecting edge corresponds to two sub-vectors, and the starting points of the two sub-vectors correspond to the two nodes of the connecting edge respectively. For each of the connecting edges, the lengths of the two sub-vectors corresponding to the connecting edge are calculated, and the direction of the connecting edge is determined based on the directions of the two sub-vectors; For each connecting edge, if the directions of the two corresponding sub-vectors are opposite, the first weight of the connecting edge is determined according to the difference in length of the two corresponding sub-vectors; otherwise, the first weight of the connecting edge is determined according to the maximum value of the length of the two corresponding sub-vectors. Remove all connected edges whose weight is less than the fourth threshold from all the connected edges.

7. The regional precipitation spatiotemporal prediction method based on multi-source data fusion as described in claim 6, characterized in that, The graph neural network includes multiple neural network layers; the step of inputting the meteorological propagation map into the preset graph neural network for graph signal transmission and feature aggregation operations to update the embedded features of each node includes: Based on the direction of each connecting edge, determine several adjacent nodes corresponding to each node when it is the target node; The meteorological propagation map is sequentially input into each of the neural network layers. After each input into the neural network layer, for each node in the meteorological propagation map, the embedding features of each of the node's corresponding neighboring nodes are fused according to the first weight of the connection edge between the node and each of its corresponding neighboring nodes, and the embedding features of the node are updated according to the fused embedding features.

8. A regional precipitation spatiotemporal prediction system based on multi-source data fusion, characterized in that, The regional precipitation spatiotemporal prediction system based on multi-source data fusion, which is applied to the meteorological forecasting internet platform, includes: a data acquisition module, a node construction module, a connection edge construction module, and a graph propagation module; The data acquisition module is used to acquire several current vehicle data received by each base station in the area at the current moment; the current vehicle data includes: a first wiper brush frequency, a first communication signal strength data, and a first satellite positioning. The node construction module is used to use each base station as a node, determine the environment of the corresponding vehicle based on the first satellite positioning, and fuse the first wiper frequency and the first communication signal strength data based on the environment to generate the embedding features of each node; the environment includes: a first interference environment with only communication interference, a second interference environment with only wiper interference, and an interference-free environment; The connection edge construction module is used to obtain the wind direction vector collected by the corresponding sensor of each base station in the region at the current time, generate the connection edge between each node according to the wind direction vector, and obtain the meteorological propagation map corresponding to the current time. The graph propagation module is used to input the meteorological propagation map into a preset graph neural network for graph signal transmission and feature aggregation operations, so as to update the embedding features of each node, and predict the precipitation intensity of the corresponding sub-region of each base station based on the updated embedding features of each node.

9. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a spatiotemporal prediction method for regional precipitation based on multi-source data fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform a spatiotemporal prediction method for regional precipitation based on multi-source data fusion as described in any one of claims 1 to 7.

Images