Meteorological prediction method and system of multi-modal meteorological data based on deep learning

Abstract

The invention discloses a meteorological prediction method and system for multi-modal meteorological data based on deep learning, relates to the technical field of meteorological prediction methods, and determines meteorological characteristics based on the recognition of a meteorological deep learning model on a plurality of meteorological data combinations. The multi-modal meteorological data is determined according to the plurality of meteorological characteristics, the position of the meteorological detector and the previous meteorological event at the position, so that the accuracy of the multi-modal meteorological data is ensured, and the management and control of the multi-dimensional meteorological data are realized. Determining a meteorological event list according to the detection of the multi-modal meteorological data, and marking the priority of each predicted meteorological event in the meteorological event list and a corresponding core meteorological element; and determining a corresponding matching coefficient based on the matching between each core meteorological element and the meteorological data set, and determining a target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environment image detected by the meteorological detector.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological forecasting methods, and in particular to a meteorological forecasting method and system based on multimodal meteorological data based on deep learning. Background Art

[0002] With the development of science and technology, meteorological detectors are gradually used in life and perform meteorological detection on the surrounding environment. Meteorological detectors collect multiple meteorological data during operation. In the existing technology, the corresponding meteorological events are determined based on the detection of multiple meteorological data, and the location of the meteorological detector and the previous meteorological events at that location are not considered as a whole. As a result, the accuracy of the meteorological data cannot be achieved, and the management and control of meteorological data in multiple dimensions cannot be realized, which affects the precise control of meteorological events. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art. The present invention provides a weather forecasting method and system based on multimodal meteorological data based on deep learning.

[0004] An embodiment of the present invention provides a weather forecasting method based on multimodal weather data using deep learning, comprising: Collecting the position of the meteorological detector and triggering meteorological detection of the meteorological detector according to the position of the meteorological detector to determine the meteorological data set; Determine multiple meteorological data combinations of different dimensions based on the division of the meteorological data set, and input them into a preset meteorological deep learning model; Determining meteorological features based on recognition of a combination of multiple meteorological data by the meteorological deep learning model, and determining multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location; Determine a meteorological event list based on the detection of multimodal meteorological data, and mark the priority of each predicted meteorological event in the meteorological event list and the corresponding core meteorological elements; The corresponding matching coefficient is determined based on the matching of each core meteorological element with the meteorological data set, and the target meteorological event is determined according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector.

[0005] An embodiment of the present invention provides a weather forecasting system for multimodal meteorological data based on deep learning. The weather forecasting system for multimodal meteorological data based on deep learning is applied to the above-mentioned weather forecasting method for multimodal meteorological data based on deep learning. The weather forecasting system for multimodal meteorological data based on deep learning includes: A meteorological detection module is used to collect the position of the meteorological detector and trigger the meteorological detection of the meteorological detector according to the position of the meteorological detector to determine the meteorological data set; A meteorological data combination module is used to determine multiple meteorological data combinations of different dimensions based on the division of the meteorological data set, and input them into a preset meteorological deep learning model; A multimodal meteorological data module is configured to determine meteorological features based on the meteorological deep learning model's recognition of multiple meteorological data combinations, and to determine multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location; A meteorological event list module is used to determine a meteorological event list based on the detection of multimodal meteorological data, and mark the priority of each predicted meteorological event in the meteorological event list and the corresponding core meteorological elements; The target meteorological event module is used to determine the corresponding matching coefficient based on the matching of each core meteorological element with the meteorological data set, and to determine the target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector.

[0006] Compared with the prior art, the present invention has the following beneficial effects: In an embodiment of the present invention, through the method in the embodiment of the present invention, meteorological characteristics are determined based on the recognition of multiple meteorological data combinations by the meteorological deep learning model, and multimodal meteorological data are determined according to multiple meteorological characteristics, the location of the meteorological detector and the previous meteorological events at the location. Meteorological characteristics are introduced, and the overall consideration of multiple meteorological characteristics, the location of the meteorological detector and the previous meteorological events at the location is compatible, thereby ensuring the accuracy of multimodal meteorological data and realizing the management and control of meteorological data in multiple dimensions.

[0007] Therefore, a meteorological event list is determined based on the detection of multimodal meteorological data, and the priority of each predicted meteorological event and the corresponding core meteorological element in the meteorological event list are marked; the corresponding matching coefficient is determined based on the matching of each core meteorological element with the meteorological data set, and the target meteorological event is determined according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector. The meteorological event list is introduced to further control each predicted meteorological event, and the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector are considered as a whole to ensure the precise control of the target meteorological event. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 1 is a flow chart of a weather forecasting method based on multimodal weather data of deep learning in an embodiment of the present invention; Figure 2 1 is a flow chart of step S11 in the weather forecasting method for multimodal weather data based on deep learning in an embodiment of the present invention; Figure 3 1 is a flow chart of step S12 in the weather forecasting method for multimodal weather data based on deep learning in an embodiment of the present invention; Figure 4 3 is a flow chart of step S13 in the weather forecasting method based on multimodal weather data of deep learning in an embodiment of the present invention; Figure 5 1 is a flow chart of step S14 in the weather forecasting method for multimodal weather data based on deep learning in an embodiment of the present invention; Figure 6 1 is a flow chart of step S15 in the weather forecasting method for multimodal weather data based on deep learning in an embodiment of the present invention; Figure 7 Schematic diagram of the structural composition of a weather forecasting system for multimodal weather data based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0010] See also Figures 1 to 7 A meteorological forecasting method based on multimodal meteorological data based on deep learning is applied to a meteorological forecasting scenario based on multimodal meteorological data based on deep learning; the meteorological forecasting method based on multimodal meteorological data based on deep learning includes: Step S11: collecting the position of the meteorological detector and triggering the meteorological detector to detect the weather according to the position of the meteorological detector to determine the meteorological data set; Step S12: determining multiple meteorological data combinations of different dimensions based on the division of the meteorological data set, and inputting them into a preset meteorological deep learning model; Step S13: determining meteorological features based on the recognition of multiple meteorological data combinations by the meteorological deep learning model, and determining multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location; Step S14: determining a meteorological event list based on the detection of multimodal meteorological data, and marking the priority and corresponding core meteorological elements of each predicted meteorological event in the meteorological event list; Step S15: determining a corresponding matching coefficient based on the matching of each core meteorological element with the meteorological data set, and determining a target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element, and the environmental image detected by the meteorological detector; refer to Figure 2 In step S11, the position of the meteorological detector is collected, and meteorological detection of the meteorological detector is triggered according to the position of the meteorological detector to determine a meteorological data set; In the specific implementation process of the present invention, the specific steps are: S111: determining the location of the meteorological detector according to the name of the meteorological detector and the meteorological detection database, collecting the detection task list of the meteorological detector, and determining the current detection task of the meteorological detector according to the detection of the detection task list of the meteorological detector; S112: determining a detection mode of the meteorological detector based on the current detection task of the meteorological detector and the position of the meteorological detector, and triggering online meteorological detection of the meteorological detector according to the detection mode; S113: Real-time monitoring of the online meteorological detection of the meteorological detector, collecting multiple meteorological data based on the online meteorological detection of the meteorological detector, and determining the meteorological data set according to the multiple meteorological data, the corresponding directions and the detection time. At this time, each meteorological data, the corresponding direction and the detection time are in a corresponding relationship.

[0011] In an embodiment of the present application, each meteorological detector will have a unique name or identifier, which will match the record in the meteorological detection database; the database stores detailed information about each meteorological detector, including its geographic location (such as longitude and latitude coordinates), affiliated organization, installation date, equipment type, etc.; when it is necessary to determine the location of a meteorological detector, the system will first query the name or identifier of the detector; then, search the database for a record that matches the name or identifier; once a matching record is found, the system will read and record the geographic location information of the meteorological detector.

[0012] Each meteorological detector will have a detection task list, which details the various detection tasks that the detector needs to perform. These tasks include regular meteorological observations, monitoring of special weather events, data collection requested by users, etc. The task list is usually stored in the task management module of the meteorological detection system. After determining the location of the meteorological detector, the system will query the detection task list of the detector, which is achieved by communicating with the interface of the task management module. The system will read all tasks in the task list and sort them according to factors such as task priority and time requirements.

[0013] After collecting the detection task list of the meteorological detector, the system needs to determine the detection task that needs to be performed currently based on the current time and task priority. This usually involves traversing the task list, checking the time requirements and priority of each task, and selecting the task that matches the current time and has the highest priority as the current detection task.

[0014] Furthermore, the system will determine the most appropriate detection mode based on the current detection task and location information of the meteorological detector; the detection mode usually involves multiple parameters, such as detection frequency, data sampling rate, resolution, observation range, etc. The selection of these parameters will directly affect the quality and efficiency of meteorological data collection; for example, if the current detection task is to monitor an impending rainstorm event, then the system will select a higher detection frequency and data sampling rate in order to more accurately capture changes in the rainfall process; at the same time, if the meteorological detector is located in a key area affected by the rainstorm, the system will also adjust the observation range to ensure that the meteorological conditions in the area can be fully covered.

[0015] Therefore, the online meteorological detection of the meteorological detector is monitored in real time, and multiple meteorological data are collected based on the online meteorological detection of the meteorological detector. The meteorological data set is determined according to the multiple meteorological data, the corresponding directions and the detection time. At this time, each meteorological data, the corresponding direction and the detection time are in a corresponding relationship, which is compatible with the overall consideration of multiple meteorological data, the corresponding directions and the detection time, and ensures the accuracy of the meteorological data set.

[0016] At this time, the system will monitor the online meteorological detection process of the meteorological detector in real time, which usually involves real-time communication with the meteorological detector to ensure that the system can obtain the working status of the detector and the collected data at any time; the importance of real-time monitoring is that it allows the system to promptly discover and resolve any potential problems, such as data transmission interruption, equipment failure or data anomaly; through continuous monitoring, the system ensures the continuity and accuracy of meteorological data, providing a reliable basis for subsequent data analysis and meteorological forecasting; at this time, status query commands are regularly sent to the meteorological detector, and responses returned by the detector are received; by analyzing these responses, the system determines whether the working status of the detector is normal and whether further measures need to be taken.

[0017] While monitoring the meteorological detector in real time, the system will collect multiple meteorological data based on the detector's online meteorological detection. These data include temperature, humidity, air pressure, wind speed, wind direction and other meteorological elements; the system will parse and process the data packets returned by the detector, extract useful meteorological information, and store it in the database for subsequent use; in the process of collecting data, the system will ensure the integrity and accuracy of the data to avoid any data loss or errors; at this time, the communication interface will establish a stable communication connection with the meteorological detector and receive data in a predetermined format and frequency; after receiving the data, the system will perform necessary verification and processing to ensure the accuracy and reliability of the data.

[0018] After collecting multiple meteorological data, the system will determine a meteorological data set based on these data, the corresponding directions (such as longitude and latitude coordinates) and the detection time. This set contains all relevant meteorological information, and each data point is associated with a specific direction and detection time. This correspondence is crucial for subsequent data analysis and meteorological forecasting; it allows the system to organize and process meteorological data according to the dimensions of time and space, so as to more accurately reveal the changing laws and trends of meteorological phenomena; at this time, it is assumed that "Observation Station A" is conducting online meteorological detection and sending data to the system in real time; the system monitors that the detector is working normally and starts collecting data; at a certain moment (such as 14:00 on May 1, 2023), the system collects the following meteorological data: temperature: 28°C; humidity: 75%; air pressure: 1013.2hPa; wind speed: 3m / s; wind direction: north wind.

[0019] At the same time, the system also records the corresponding position of these data (such as the latitude and longitude coordinates of observation station A) and the detection time (14:00 on May 1, 2023); combining these data together forms a meteorological data set, which contains the meteorological conditions in area A at that moment and is used for subsequent data analysis and meteorological forecasting. By real-time monitoring of online meteorological detection by meteorological detectors, collecting multiple meteorological data, and determining the meteorological data set based on these data, the corresponding position and detection time, the integrity, accuracy and spatiotemporal correlation of the meteorological data are ensured, which provides a reliable basis for subsequent meteorological analysis and forecasting.

[0020] In some embodiments of the present application, a meteorological data set matching table is collected, and the meteorological data set matching table is shown in Table 1: Table 1 Meteorological data set matching table Detection time bearing (longitude, latitude) Temperature (°C) humidity(%) Air pressure (hPa) Wind speed (m / s) wind direction 2023-05-01 14:00 (113.26, 23.13) 28 75 1013.2 3 north wind 2023-05-01 14:05 (113.26, 23.13) 28.5 74 1013.0 3.5 Northeast Wind refer to Figure 3 , in step S12, multiple meteorological data combinations of different dimensions are determined according to the division of the meteorological data set, and are input into a preset meteorological deep learning model; In the specific implementation process of the present invention, the specific steps are: S121: collecting a meteorological data set, and determining a plurality of meteorological data combinations according to the division of the meteorological data set; S122: Marking the multiple meteorological data combinations and marking corresponding meteorological dimensions. The meteorological dimensions of the multiple meteorological data combinations are inconsistent. The meteorological dimensions include thermodynamic dimension, dynamic dimension, cloud physics dimension, and pressure dimension. S123: Determine the corresponding detection space based on the name and location of the meteorological detector, determine the preset meteorological deep learning model based on the tracing of the detection space, and input each meteorological data combination into the preset meteorological deep learning model to obtain the output result of the meteorological deep learning model.

[0021] In an embodiment of the present application, the system collects meteorological data from various sources (such as meteorological observation stations, satellites, radars, etc.). These data include various meteorological elements such as temperature, humidity, air pressure, wind speed, wind direction, precipitation, etc., and usually exist in the form of time series; the collected data will be integrated into a meteorological data set to provide a basis for subsequent processing and analysis.

[0022] After collecting the meteorological data set, the system will divide this data into multiple meteorological data combinations according to specific division criteria. These combinations are based on time intervals (such as hourly or daily data), geographical locations (such as data from different cities or regions), types of meteorological elements (such as temperature, humidity, wind speed, etc.) or any other meaningful division methods; at this time, it needs to be determined according to the needs of analysis or prediction; each combination should contain sufficient information to support subsequent analysis or prediction tasks; ensure that the data between different combinations are consistent in time and space for comparison and analysis.

[0023] Furthermore, the system will mark the multiple meteorological data combinations generated in the previous steps; the purpose of marking is to distinguish different data combinations and facilitate subsequent analysis and processing; the tag is an identifier in any form such as numbers, letters, strings, etc., as long as it can uniquely represent each data combination; at this time, ensure that each data combination has a unique tag to avoid confusion; the tag should be easy to understand and remember to facilitate subsequent processing and analysis; the tag usually needs to be stored together with the data combination so that it can be quickly retrieved when needed.

[0024] After labeling the data combinations, the system needs to assign a corresponding meteorological dimension to each combination. Meteorological dimension is a way to classify meteorological data, which divides data into different categories based on its characteristics and usage. In this case, meteorological dimensions include thermodynamic dimension, kinetic dimension, cloud physics dimension, and pressure dimension. At this time, the appropriate meteorological dimension is selected based on the meteorological elements contained in the data combination and the characteristics of the data. The meaning and scope of each meteorological dimension must be clarified to ensure data consistency and comparability. A unique identifier is assigned to each meteorological dimension and associated with the label of the data combination.

[0025] Specifically, suppose there are three meteorological data combinations: Combination A: contains temperature data for a certain area for three consecutive days; Combination B: contains wind speed and direction data for the same area for three consecutive days; Combination C: contains precipitation and cloud cover data for a rainfall event in the area; now, these data combinations need to be labeled and assigned corresponding meteorological dimensions.

[0026] Combination A: labeled "Temp_3days"; Combination B: labeled "Wind_3days"; Combination C: labeled "Rain_Event"; Combination A (Temp_3days): thermodynamic dimension; because temperature is an important parameter in thermodynamic processes, this set of data is classified into the thermodynamic dimension; Combination B (Wind_3days): dynamic dimension; wind speed and direction reflect the dynamic processes in the atmosphere, so this set of data is classified into the dynamic dimension; Combination C (Rain_Event): cloud physics dimension; precipitation and cloud cover are closely related to cloud physics processes such as cloud formation, development and dissipation, so this set of data is classified into the cloud physics dimension.

[0027] Therefore, the corresponding detection space is determined based on the name and location of the meteorological detector, and the preset meteorological deep learning model is determined based on the tracing of the detection space. Each meteorological data combination is input into the preset meteorological deep learning model to obtain the output result of the meteorological deep learning model, which is compatible with the overall consideration of the name and location of the meteorological detector, and ensures the accuracy of the corresponding detection space.

[0028] At this time, the system will determine the corresponding detection space based on the name of the meteorological detector (such as site name, device ID, etc.) and location information (such as latitude and longitude coordinates, administrative divisions, etc.); the detection space is usually a geographical area, which defines the range within which the detector can accurately measure and record meteorological data. This range is a fixed geographical area and a dynamically changing space, depending on the type, accuracy and layout of the detector; at this time, ensure that the name and location information of the detector are accurate and reliable to avoid determining the wrong detection space; clarify the boundaries and scope of the detection space based on the characteristics and requirements of the detector; if the location or measurement range of the detector changes, the detection space information needs to be updated in a timely manner.

[0029] After determining the detection space, the system will trace and select preset meteorological deep learning models based on this space. These models are usually trained based on historical meteorological data and can predict and analyze specific regions or meteorological phenomena. The system will select the most appropriate model based on factors such as the geographical location, climate characteristics, and historical meteorological data of the detection space. At this time, multiple deep learning models for different regions and meteorological phenomena are prepared in advance so that they can be selected as needed. Appropriate model selection criteria are formulated based on the characteristics and needs of the detection space. As new data accumulates and model performance improves, the model needs to be updated and optimized regularly.

[0030] After selecting a suitable deep learning model, the system will input various meteorological data combinations into the model to obtain the model's output results, which include future weather forecasts, trend analysis of meteorological elements, detection of abnormal meteorological events, etc. The accuracy and reliability of the output results depend on the performance of the model and the quality of the data. At this time, before inputting the data, the data needs to be preprocessed, such as normalization and denoising, to improve the model's predictive performance; the output results of the model are interpreted and analyzed to better understand meteorological phenomena and make decisions; the accuracy and reliability of the output results are verified by comparing with actual observation data or data from other reliable sources.

[0031] Specifically, suppose there is a weather detector named "Weather Station A", which is located in the center of City A and can collect meteorological data such as temperature, humidity, and air pressure in real time; now, it is necessary to predict the weather conditions of City A in the next week based on this data; Name: Weather Station A Location: City Center A (assuming longitude and latitude are 39.9°N, 116.4°E) Detection space: Urban area of ​​City A (determined according to the location and measurement range of the weather station).

[0032] Based on the geographical location, climate characteristics (such as distinct four seasons, rainy summers, etc.) and historical meteorological data of City A, a deep learning model specifically for weather forecasting in City A is selected; the model's prediction accuracy, the spatiotemporal coverage of the training data, the model's update frequency, etc.; the collected temperature, humidity, air pressure and other data are normalized to eliminate the impact of different dimensions on the data; the preprocessed data combination is input into the preset deep learning model; the model outputs the weather forecast for the next week, including daily maximum temperature, minimum temperature, precipitation and other information. Through this process, we can see how step S123 determines the detection space based on the name and location of the meteorological detector, selects the preset deep learning model, and inputs the data into the model to obtain the output results. This provides an efficient and accurate method for meteorological forecasting and analysis.

[0033] In some embodiments of the present application, a detection space matching table is collected, and the detection space matching table is shown in Table 2: Table 2 Detection space matching table Detector name Latitude and longitude Administrative divisions Detection space Beijing Railway Station 39.9°N, 116.4°E Beijing Beijing urban area Shanghai Station 31.2°N, 121.5°E Shanghai Downtown Shanghai Guangzhou Station 23.1°N, 113.3°E Guangzhou Guangzhou City When the system receives the name and location information of the weather detector, it will determine the corresponding detection space by querying the matching table; the model matching table is introduced, and the model matching table is shown in Table 3: Table 3 Model matching table Detection space Model Name Model Description Beijing urban area Beijing Model Deep learning model trained based on Beijing's historical meteorological data Downtown Shanghai Shanghai Model Deep learning model trained based on Shanghai historical meteorological data Guangzhou City Guangzhou Model Deep learning model trained based on Guangzhou's historical meteorological data Suppose there is a meteorological data combination that includes the temperature, humidity, air pressure and wind speed data of Beijing city on a certain day; input this data into the preset "Beijing Model" and calculate the output result; input data: temperature: 25°C; humidity: 60%; air pressure: 1013.25hPa; wind speed: 3m / s; forecast of the maximum temperature in the next 24 hours: predicted value: 28°C; weight: 0.8 (the relative importance of historical temperature data in the forecast); score: 0.9 (the accuracy of the model's temperature forecast); forecast of precipitation in the next 24 hours: predicted value: 5mm; weight: 0.6 (the relative importance of cloud cover and humidity data in precipitation forecast); score: 0.7 (the accuracy of the model's precipitation forecast).

[0034] refer to Figure 4 In step S13, meteorological features are determined based on the recognition of multiple meteorological data combinations by the meteorological deep learning model, and multimodal meteorological data are determined based on the multiple meteorological features, the location of the meteorological detector, and previous meteorological events at the location; In the specific implementation process of the present invention, the specific steps are: S131: When each meteorological data combination is input into a preset meteorological deep learning model, a plurality of recognition channels are determined based on detection by the meteorological deep learning model, and states of the plurality of recognition channels are marked. The meteorological deep learning model determines an order of recognition of the meteorological data combination based on the states of the plurality of recognition channels and the meteorological data combination; S132: triggering the meteorological deep learning model to autonomously identify each meteorological data combination along the recognition sequence, and sequentially outputting corresponding meteorological features, and determining a first form of the meteorological data based on the multiple meteorological features and the location of the meteorological detector; S133: Determine a second form of meteorological data based on the location of the meteorological detector and past meteorological events at the location, and determine multimodal meteorological data based on a synthesis of the first form of meteorological data and the second form of meteorological data.

[0035] In an embodiment of the present application, the system will input the collected meteorological data combinations into a preset meteorological deep learning model. These meteorological data combinations contain data on various meteorological elements such as temperature, humidity, air pressure, wind speed, and precipitation. At the same time, after receiving the data, the meteorological deep learning model will perform internal processing to identify the features in the data. In this process, the model will determine multiple identification channels based on its structure and design. These channels are understood as different processing paths or modules within the model, and each channel is specifically responsible for processing or identifying a specific type of meteorological data or features.

[0036] For example, a meteorological deep learning model contains the following recognition channels: temperature recognition channel: specifically processes temperature data and extracts temperature-related features; humidity recognition channel: specifically processes humidity data and extracts humidity-related features; air pressure recognition channel: specifically processes air pressure data and extracts air pressure-related features.

[0037] After determining the identification channels, the system will mark the status of each channel, which includes activated, inactivated, busy, idle, etc. The purpose of marking the status is to manage the usage of the channels and ensure that the data can be processed correctly in the predetermined order. For example, when temperature data is input into the model, the temperature identification channel will be marked as activated; when the channel is processing data, it will be marked as busy; after processing is completed, it will be marked as idle and wait for the arrival of the next data packet.

[0038] The system will determine the data recognition order based on the status of the recognition channel and the input meteorological data combination. This order is based on multiple factors such as the data priority, the channel processing capability, and the real-time requirements of the data. For example, if temperature data is very important for the current forecast task and the temperature recognition channel is idle, the system will give priority to processing the temperature data. If humidity data and pressure data arrive at the same time, but the humidity recognition channel is busy processing other data and the pressure recognition channel is idle, the system will process the pressure data first and then the humidity data.

[0039] Specifically, suppose there is a meteorological deep learning model, which contains three recognition channels: temperature, humidity and air pressure; now, there are two meteorological data combinations that need to be processed: data combination A: contains temperature data and humidity data; data combination B: contains air pressure data and additional temperature data (used to verify or update temperature predictions); the system will first detect the structure of the model and determine the three recognition channels of temperature, humidity and air pressure; then, it will mark the status of these channels; assume that initially, all channels are idle; next, the system will determine the recognition order based on the status of the channels and the content of the data combination; since the temperature data and humidity data in data combination A are both important and the channels are idle, the system will process these two data simultaneously or in any order; but to simplify the explanation, assume that the system processes the temperature data first and then the humidity data.

[0040] When processing the temperature data of data combination A, the temperature identification channel will be marked as active and busy; once the processing is completed, it will be marked as idle; then, the system will process the humidity data. Similarly, the humidity identification channel will be marked as active and busy, and will become idle after the processing is completed.

[0041] When processing data combination B, the system notices that the pressure recognition channel is idle, while the temperature recognition channel, although it has just finished processing the temperature in data A, still needs some time to perform any necessary subsequent processing or state updates (this depends on the specific implementation of the model); therefore, the system processes the pressure data first, and then processes the additional temperature data (for verification or update).

[0042] Furthermore, the meteorological deep learning model is triggered to autonomously identify each meteorological data combination along the recognition sequence, and the corresponding meteorological features are output in sequence. The first form of the meteorological data is determined based on multiple meteorological features and the positions of meteorological detectors, which is compatible with the overall consideration of multiple meteorological features and the positions of meteorological detectors, ensuring the accuracy of the first form of meteorological data.

[0043] At this point, in step S131, the recognition order of each meteorological data combination by the meteorological deep learning model has been determined; in this step, the system will trigger the model's autonomous recognition process for each data combination one by one in this order, which means that the model will independently process each data combination according to the preset logic and algorithm without the need for external real-time intervention.

[0044] When the model autonomously identifies a certain data combination, it extracts a series of meteorological features from the data. These features are the hidden layer representations learned within the model and are also the direct result of the model output. They can reflect certain key attributes or patterns of the data; meteorological features include temperature change trends, humidity fluctuation ranges, and the rate of rise and fall of air pressure.

[0045] After obtaining the meteorological characteristics, the system will combine the location information of the meteorological detector to determine the first form of the meteorological data; the first form is understood as a preliminary classification or representation of the data, which is based on the intrinsic characteristics of the data and the external location information; for example, if a meteorological detector is located in the city center and the data it collects shows the characteristics of high temperature and high humidity, then the system will determine the first form of the data as "high temperature and high humidity environment in the city center."

[0046] Specifically, suppose there is a meteorological deep learning model that can autonomously identify temperature, humidity, and wind speed data and output corresponding features; now, there is a meteorological detector located in the suburbs of the city, which collects a set of meteorological data; the system first triggers the model to autonomously identify the data collected by the suburban meteorological detector; the model will process the temperature, humidity, and wind speed data in a preset order; after the model processes the data, it outputs a series of meteorological features; for example, the temperature feature shows that the suburbs are currently at a lower temperature level; the humidity feature indicates that the air is relatively dry; and the wind speed feature shows slight wind activity.

[0047] The system combines these meteorological characteristics with the location information of the detector (urban suburbs) to determine a preliminary first form of meteorological data. In this example, due to the low temperature, dry humidity and weak wind in the suburbs, the system will determine the first form of the data as "low temperature, low humidity and light breeze environment in urban suburbs." Through this example, we can see how step S132 triggers autonomous identification and output of meteorological characteristics along the recognition sequence, and determines the first form of meteorological data based on these characteristics and the location of the detector. This process provides an important foundation for subsequent data analysis and application.

[0048] Therefore, the second form of meteorological data is determined based on the location of the meteorological detector and the previous meteorological events at that location, and multimodal meteorological data is determined based on the synthesis of the first form of meteorological data and the second form of meteorological data. This is compatible with the overall consideration of the synthesis of the first form of meteorological data and the second form of meteorological data, ensuring the accuracy of multimodal meteorological data. At the same time, meteorological features are introduced, which is compatible with the overall consideration of multiple meteorological features, the location of the meteorological detector and the previous meteorological events at that location, ensuring the accuracy of multimodal meteorological data and realizing the management and control of meteorological data in multiple dimensions.

[0049] At this time, the system will consider the geographical location of the weather detector and the meteorological events that have occurred in the history of that location to determine the second form of the meteorological data; the second form is a deeper and more specific description of the meteorological data, which is based on the combination of location information and historical meteorological events; for example, if a weather detector is located in an area that is often hit by heavy rains, then the system will determine the second form of the meteorological data at that location as a "rainstorm-frequent area" based on this and historical rainstorm records; similarly, if the detector is located in a dry and rainy area, the system will determine its second form as a "drought area". This process involves analysis, statistics and induction of historical meteorological data, as well as reasoning and judgment based on geographical location.

[0050] After determining the first and second forms of the meteorological data, the system will synthesize the two forms to generate multimodal meteorological data; multimodal meteorological data refers to meteorological data that contains multiple different information sources, different representation forms or different analysis dimensions; for example, if there is a meteorological detector located in an area that has experienced both heavy rain and drought; through step S132, the first form of the location is determined to be "environment under specific climatic conditions" (such as "mild and humid environment" or "dry and hot environment", which depends on the current meteorological data); and through the first step of step S133, the second form of the location is determined to be "climate-variable area" (because there are both heavy rain and drought); when synthesizing these two forms, the system will combine them to form a more comprehensive description, such as "mild and humid / dry and hot environment in a climate-variable area", which includes both the current meteorological conditions (first form) and the long-term climate characteristics of the location (second form).

[0051] Specifically, suppose there is a weather detector located in a mountainous area, which has a history of both abundant snowfall records and drought years; now, a set of current weather data is collected; based on the location of the weather detector (mountainous area) and past weather events (abundant snowfall records and drought years), the system determines the second form of the weather data at this location as "mountainous area with alternating snowfall and drought"; through step S132, the current first form is determined to be "current climate state in the mountainous area" (such as "mild and snowy" or "dry and rainy", depending on the current weather data).

[0052] When synthesizing these two forms, the system generates a description, such as "the current mild snowy / dry and rainy climate state in mountainous areas with alternating snowfall and drought." This description reflects both the current meteorological conditions and the long-term climate characteristics of the location. Through this example, we can see how step S133 determines the second form based on the location of the meteorological detector and past meteorological events, and synthesizes the first and second forms to determine multimodal meteorological data. This process helps provide more comprehensive and in-depth meteorological information, and provides strong support for applications such as meteorological forecasting and disaster warning.

[0053] In an embodiment of the present application, the first form score (based on current meteorological data): temperature (0.3), humidity (0.2), wind speed (0.1); the second form score (based on historical meteorological events): snowfall frequency (0.2), proportion of drought years (0.2); Now, there is a meteorological detector located in a mountainous area, which collects the following current meteorological data: temperature = 5°C, humidity = 80%, wind speed = 5m / s; at the same time, it is known that in history, snowfall has been frequent in this mountainous area and the proportion of drought years is high.

[0054] Temperature score = 5°C (assuming a high score within the low temperature range, the specific value is determined by the scoring criteria); Humidity score = 80% (assuming a high score within the high humidity range); Wind speed score = 5m / s (assuming a medium score within the medium wind speed range); First form comprehensive score = Temperature score × 0.3 + Humidity score × 0.2 + Wind speed score × 0.1; Snowfall frequency score = high (assuming frequent snowfall equals high); drought year proportion score = high (assuming a high proportion of drought years equals high); second form comprehensive score = snowfall frequency score × 0.2 + drought year proportion score × 0.2; comprehensive score = first form comprehensive score + second form comprehensive score.

[0055] Assume that after calculation, the following results are obtained: the comprehensive score of the first form = 7.5; the comprehensive score of the second form = 4.0; the comprehensive score of the multimodal meteorological data = 11.5; the system will output the following multimodal meteorological data description: "The mountainous area is currently in a climate state of low temperature, high humidity and moderate wind speed (the comprehensive score of the first form is 7.5), and historically, snowfall has been frequent and drought years have accounted for a high proportion (the comprehensive score of the second form is 4.0). The comprehensive score is 11.5, indicating that the current and long-term climate characteristics of the mountainous area are relatively significant;"; Through this method, combined with the location of the meteorological detector, current meteorological data and historical meteorological events, multimodal meteorological data containing rich information is generated.

[0056] refer to Figure 5, in step S14, a meteorological event list is determined based on the detection of multimodal meteorological data, and the priority and corresponding core meteorological elements of each predicted meteorological event in the meteorological event list are marked; In the specific implementation process of the present invention, the specific steps are: S141: Collecting multimodal meteorological data, determining a plurality of prediction combinations based on the detection of the multimodal meteorological data, determining corresponding predicted meteorological events based on the identification of the plurality of prediction combinations, collecting a plurality of predicted meteorological events, and determining a meteorological event list based on the combination of the plurality of predicted meteorological events; S142: Determining first priority coefficients of the plurality of predicted meteorological events in the meteorological event list based on matching of the plurality of predicted meteorological events with the meteorological data, and determining the priority of each predicted meteorological event based on a mapping relationship between the event type, the first priority coefficient, and the priority of the plurality of predicted meteorological events; S143: Sort multiple predicted meteorological events in order of priority. At this time, determine the core element detection method based on the order of priority, the event content of the multiple predicted meteorological events and the correlation coefficient between the multiple predicted meteorological events, and autonomously detect the multiple predicted meteorological events based on the core element detection method to determine the core meteorological elements corresponding to each predicted meteorological event.

[0057] In an embodiment of the present application, multimodal meteorological data is collected, multiple prediction combinations are determined based on the detection of the multimodal meteorological data, and corresponding predicted meteorological events are determined based on the identification of the multiple prediction combinations. In order to collect multiple predicted meteorological events, a meteorological event list is determined based on the combination of the multiple predicted meteorological events. This is compatible with the overall consideration of the combination of multiple predicted meteorological events and ensures the accuracy of the meteorological event list.

[0058] At this time, multimodal meteorological data is collected. After collecting the multimodal meteorological data, the system will use machine learning or deep learning algorithms to detect patterns and associations in the data. These algorithms will identify statistical correlations between data, trends and cyclical changes in time series, or outliers in the data; based on these detections, the system will generate multiple forecast combinations, each combination is a set of associations or patterns between meteorological elements.

[0059] For each forecast combination, the system will further analyze it to determine whether it corresponds to a specific meteorological event; meteorological events are weather phenomena (such as heavy rain, drought, typhoon, frost, etc.) and meteorological conditions (such as high temperature, low temperature, strong wind, etc.); the system will use pre-defined rules, classification models or thresholds to determine whether the forecast combination meets the definition of a meteorological event.

[0060] After identifying the corresponding predicted meteorological events, the system will record these events, which occur simultaneously and at different time and spatial scales; the system needs to be able to handle this complexity and ensure that each event is accurately recorded; finally, the system combines all identified predicted meteorological events into a list, which contains detailed information about each event, such as event type, occurrence time, location, intensity, etc. This list is the basis for subsequent steps (such as S142 and S143) for further analysis and processing of meteorological events.

[0061] Specifically, suppose there is a weather monitoring system that is collecting data from multiple weather stations, including temperature, humidity, wind speed, and precipitation. The system collects data from different weather stations and preprocesses it to ensure the quality and consistency of the data. The system uses machine learning algorithms to analyze the data and identify some prediction combinations. For example, the system finds that when the temperature continues to rise and the humidity decreases, it is often accompanied by the emergence of hot weather. When the wind speed increases and the precipitation suddenly increases, it indicates the arrival of heavy rain.

[0062] Based on these prediction combinations, the system further identifies the corresponding predicted meteorological events; for example, the system determines that a certain area is about to experience high temperature weather because the temperature in the area has exceeded 35°C for several consecutive days and the humidity continues to drop; at the same time, the system also predicts heavy rain in another area because the wind speed in the area has accelerated and the precipitation has increased significantly in a short period of time; the system records these predicted meteorological events and assigns a unique identifier to each event; finally, the system generates a meteorological event list containing all identified predicted meteorological events. This list shows information such as the type, time of occurrence, location and intensity of each event, providing a basis for subsequent analysis and processing.

[0063] Furthermore, in the meteorological event list, the first priority coefficients of multiple predicted meteorological events are determined based on the matching of multiple predicted meteorological events with meteorological data, and the priority of each predicted meteorological event is determined based on the event type, first priority coefficient and priority mapping relationship of multiple predicted meteorological events. This is compatible with the overall consideration of the event type, first priority coefficient and priority mapping relationship of multiple predicted meteorological events, ensuring the accuracy of the priority of each predicted meteorological event.

[0064] At this time, the system will determine a first priority coefficient based on the degree of matching between each predicted meteorological event in the meteorological event list and the current or historical meteorological data. This coefficient reflects the degree of correlation or credibility between the predicted meteorological event and the actual meteorological conditions; the degree of matching is based on multiple factors, such as the temporal and spatial distribution of the event, the numerical range of meteorological elements, the probability of the event occurring, etc.; in order to calculate the first priority coefficient.

[0065] After determining the first priority coefficient, the system will determine the priority of each event based on the event type of multiple predicted meteorological events, the first priority coefficient, and a predefined priority mapping relationship; the priority mapping relationship is a scoring system based on factors such as the urgency of the event type, the scope of impact, and the uncertainty of the forecast.

[0066] The system assigns a basic priority to each event type, and then adjusts the basic priority based on the first priority coefficient; for example, for meteorological events that have a serious impact on human society (such as typhoons, rainstorms, etc.), the system will give a higher basic priority; and for events with smaller impacts or greater forecast uncertainty, the system will give a lower priority; finally, the system will sort the meteorological event list according to these adjusted priorities so that subsequent steps can give priority to the most important or most urgent meteorological events.

[0067] Specifically, assume there is a meteorological monitoring system that has generated a meteorological event list containing multiple predicted meteorological events, including heavy rain, high temperature, drought, and lightning; the system first calculates a first priority coefficient based on the degree of match between each predicted meteorological event and the current meteorological data; for example, for heavy rain events, the system compares the difference between the predicted precipitation and the actual observed precipitation, as well as the degree of overlap between the predicted precipitation area and the actual precipitation area; based on these comparison results, the system assigns a higher first priority coefficient to the heavy rain event because it has a higher degree of match with the actual meteorological data.

[0068] The system determines the priority of each event based on the event type, first priority coefficient and priority mapping relationship; in this example, it is assumed that the system has defined a priority mapping relationship, in which rainstorm events are given the highest basic priority due to their serious impact on human society; although high temperature and drought events also have a certain impact, their urgency and impact range are smaller, so they are given a lower basic priority; lightning events are also given a lower priority due to the large prediction uncertainty; after considering the first priority coefficient, the system adjusts the basic priority; for rainstorm events, due to their high first priority coefficient, the system makes a positive adjustment to their basic priority, making their final priority higher; and for high temperature, drought and lightning events, due to their low first priority coefficient or large prediction uncertainty, the system makes a negative adjustment to their basic priority or keeps it unchanged.

[0069] Finally, the system sorts the list of meteorological events according to the adjusted priority; in this example, the rainstorm event is ranked first because it has the highest priority, followed by high temperature and drought events, and finally the lightning event; through this example, we can see how step S142 determines the priority of each event based on the degree of match between the predicted meteorological event and the meteorological data and the mapping relationship between event type and priority. This process helps the system prioritize the most important or most urgent meteorological events, thereby improving the efficiency and accuracy of meteorological monitoring and early warning.

[0070] Therefore, multiple predicted meteorological events are sorted in order of priority. At this time, the core element detection method is determined according to the order of priority, the event content of multiple predicted meteorological events and the correlation coefficients between multiple predicted meteorological events, and the multiple predicted meteorological events are autonomously detected based on the core element detection method to determine the core meteorological elements corresponding to each predicted meteorological event. This is compatible with the overall consideration of the order of priority, the event content of multiple predicted meteorological events and the correlation coefficients between multiple predicted meteorological events, thereby ensuring the accuracy of the core element detection method.

[0071] At this time, the system will sort the predicted meteorological events according to their priority order determined in step S142, which means that events with high priority will be placed at the front of the list, while events with low priority will be placed at the back of the list; the purpose of sorting is to ensure that the system can handle the most important or urgent meteorological events first.

[0072] After ranking the forecasted meteorological events, the system needs to determine the core element detection method based on the event priority, event content, and the correlation coefficient between events. The core element detection method refers to the method or technology used to identify and analyze the key meteorological elements in meteorological events. These elements are the direct cause of the event and the factors that play a key role in the development of the event. To determine the core element detection method, the system will consider the following factors: event priority: high-priority events require more accurate and rapid detection methods; event content: different types of meteorological events require different detection methods; for example, rainstorm events require a focus on analyzing precipitation and precipitation intensity, while typhoon events require attention to factors such as wind speed, wind direction and air pressure; correlation coefficient between events: if there is a strong correlation between two or more events, the system needs to adopt a method that can analyze these events simultaneously to identify the common core elements between them; based on these factors, the system will choose to use a specific detection model to detect core elements.

[0073] After determining the core element detection method, the system will conduct autonomous detection on the sorted predicted meteorological events, which means that the system will automatically apply the selected detection method to analyze the key meteorological elements of each event. This process involves multiple steps such as data preprocessing, feature extraction, model operation and result interpretation. Through autonomous detection, the system can identify the core meteorological elements corresponding to each predicted meteorological event. These elements include the physical process that causes the event, the spatial distribution characteristics of the event, the time evolution law of the event, etc.

[0074] Specifically, assume that there is a meteorological monitoring system that has determined the priorities of predicted meteorological events according to step S142 and sorted them; now, step S143 will be used to determine the core element detection method and perform autonomous detection; the sorted list of predicted meteorological events is as follows (sorted from high to low priority): heavy rain event A (high priority); typhoon event B (medium priority); high temperature event C (low priority).

[0075] For rainstorm event A, the system will choose to use a high-resolution numerical weather forecast model to detect core elements. This is because rainstorm events usually involve complex precipitation processes and spatial distribution characteristics, which require high-precision models to accurately simulate and analyze. For typhoon event B, the system will choose to use a method that combines typhoon path prediction models and wind speed and direction observation data for detection. For high temperature event C, since it is relatively simple and has a limited impact range, the system will choose to use simple statistical methods to analyze the duration and intensity of the high temperature.

[0076] After determining the core element detection method, the system will conduct autonomous detection on each event; for example, for rainstorm event A, the system will run a high-resolution numerical weather forecast model to simulate the precipitation process and analyze core elements such as precipitation intensity, precipitation area and precipitation duration; for typhoon event B, the system will use the typhoon path prediction model to predict the typhoon's movement path and intensity, and combine wind speed and direction observation data to verify the accuracy of the model; for high temperature event C, the system will use statistical methods to calculate elements such as the duration and maximum temperature of the high temperature; through this process, the system can identify the core meteorological elements corresponding to each predicted meteorological event and provide key information for subsequent meteorological warnings and services.

[0077] Specifically, a core element detection method matching table is collected. The core element detection method matching table is shown in Table 4: Table 4 Core element detection method matching table Predicting meteorological events Priority Event content Correlation coefficient Core element detection method Heavy Rain A high Heavy rainfall 0.8 High-resolution precipitation model Typhoon B middle Strong wind speed 0.6 Typhoon track prediction model High temperature C Low The temperature remains high 0.4 Temperature statistical analysis Heavy fog middle Low visibility 0.5 Visibility monitoring model Thunderstorm E high Lightning activity 0.7 Lightning monitoring network Forecasted meteorological events are sorted from high to low priority. Based on the event content and correlation coefficient, an appropriate core element detection method is selected for each event. For example, due to the heavy rainfall and high correlation coefficient with other events, a high-resolution precipitation model is selected for detection. The core meteorological elements of Rainstorm A include precipitation amount, intensity, and regional distribution. The core meteorological elements of Typhoon B include wind speed, direction, typhoon path, and pressure changes. (This process is repeated in this way, with the corresponding core meteorological elements output for each event.)

[0078] At the same time, rainstorm A: priority 3, event content 3, average correlation coefficient with other events 0.7 (assuming it is related to two events), total score = 3+3+0.7×2=7.4; typhoon B: priority 2, event content 3, average correlation coefficient with other events 0.6, total score = 2+3+0.6×2=6.2; sort from high to low according to the total score, and choose a more complex or more accurate detection method for high-scoring events; for example, rainstorm A has the highest score, and a combination of high-resolution precipitation model and radar monitoring is selected for detection; therefore, the core meteorological elements of rainstorm A: based on the data of high-resolution model and radar monitoring, output elements such as precipitation amount, precipitation intensity, and precipitation duration; the core meteorological elements of typhoon B: based on the typhoon path prediction model and meteorological station observation data, output elements such as wind speed, wind direction, typhoon path, and air pressure change.

[0079] refer to Figure 6 In step S15, a corresponding matching coefficient is determined based on the matching between each core meteorological element and the meteorological data set, and a target meteorological event is determined according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element, and the environmental image detected by the meteorological detector; In the specific implementation process of the present invention, the specific steps are: S151: collecting core meteorological elements corresponding to each predicted meteorological event, determining a core meteorological element combination based on a combination of each core meteorological element, and dynamically matching the core meteorological element combination with a meteorological data set to determine a matching coefficient between each core meteorological element and the meteorological data set; S152: Determine a first target coefficient based on the priority of each predicted meteorological event and the matching coefficient corresponding to each core meteorological element among the multiple predicted meteorological events. Simultaneously, the meteorological detector performs environmental detection on the external environment in a working state to collect an environmental image detected by the meteorological detector. S153: Determine a second target coefficient based on the priorities of each predicted meteorological event and the environmental image detected by the meteorological detector, and determine a target meteorological event according to a mapping relationship among the first target coefficient, the second target coefficient, and the target meteorological event.

[0080] In an embodiment of the present application, core meteorological elements corresponding to each predicted meteorological event are collected, a core meteorological element combination is determined based on the combination of each core meteorological element, and the core meteorological element combination is dynamically matched with the meteorological data set to determine the matching coefficient of each core meteorological element and the meteorological data set, which is compatible with the overall consideration of the combination of each core meteorological element and ensures the accuracy of the core meteorological element combination.

[0081] At this time, the core meteorological elements corresponding to each predicted meteorological event are collected. At this time, the system needs to collect the core meteorological elements of each predicted meteorological event from the previous step (such as S143). These elements are usually the key factors that lead to the occurrence of meteorological events, such as precipitation, wind speed, wind direction, temperature, humidity, etc.; for each predicted meteorological event, the system will have a corresponding set of core meteorological elements.

[0082] After collecting the core meteorological elements of each predicted meteorological event, the system needs to determine a core meteorological element combination based on the correlation and importance between these elements. This combination includes multiple elements that together describe the main characteristics of the meteorological event; for example, for heavy rain events, the core meteorological element combination includes precipitation amount, precipitation intensity and precipitation duration.

[0083] The system needs to match the determined core meteorological element combination with the current or historical meteorological data set; the meteorological data set includes observation data, forecast data or model simulation data from different sources and different time scales; the matching process is dynamic, which means that the system will continuously adjust the matching results based on the real-time update of the data; the purpose of matching is to evaluate the similarity or consistency between the core meteorological element combination and the meteorological data set, so as to determine the degree of match between them.

[0084] Based on the matching results, the system calculates the matching coefficient of each core meteorological element and the meteorological data set. This coefficient is usually a numerical value that indicates the degree of fit or correlation between the element and the data set. The matching coefficient is determined based on statistical methods, machine learning algorithms or expert systems. The high or low matching coefficient reflects the degree of consistency between the core meteorological element and the meteorological data set. A high matching coefficient means that there is a good fit between the element and the data set, while a low matching coefficient indicates that there is a large difference or uncertainty between the element and the data set.

[0085] Specifically, suppose there is a meteorological monitoring system that predicts three meteorological events that will occur in the next three days: heavy rain, typhoon and high temperature; the system collects the core meteorological elements of heavy rain events as precipitation amount, precipitation intensity and precipitation duration; the core meteorological elements of typhoon events are wind speed, wind direction, air pressure and center position; the core meteorological elements of high temperature events are maximum temperature, duration and affected range.

[0086] According to the importance and relevance of these elements, the system determines that the core meteorological element combination of rainstorm events is precipitation and precipitation intensity; the core meteorological element combination of typhoon events is wind speed, wind direction and air pressure; the core meteorological element combination of high temperature events is maximum temperature and duration; at the same time, the system matches these core meteorological element combinations with the current meteorological data set; the meteorological data set includes observation and forecast data from weather stations, satellites, radars and numerical weather forecast models; therefore, based on the matching results, the system calculates the matching coefficient of each core meteorological element with the meteorological data set; for example, the core meteorological element combination of rainstorm ... The matching coefficient between the precipitation element and the data set is 0.9 (indicating a high degree of consistency), while the matching coefficient of the precipitation intensity element is 0.8 (indicating a moderate consistency); the matching coefficient of the wind speed element of the typhoon event is 0.7, the matching coefficient of the wind direction element is 0.6, and the matching coefficient of the air pressure element is 0.9; the matching coefficient of the maximum temperature element of the high temperature event is 0.8, and the matching coefficient of the duration element is 0.9; through these steps, the system can evaluate the degree of matching between the core meteorological elements of the predicted meteorological event and the current meteorological data set, thereby providing key information for subsequent steps (such as S152 and S153).

[0087] Furthermore, among multiple predicted meteorological events, the first target coefficient is determined based on the priority of each predicted meteorological event and the matching coefficient corresponding to each core meteorological element. At the same time, the meteorological detector performs environmental detection on the external environment in a working state to collect environmental images detected by the meteorological detector, which is compatible with the overall consideration of the priority of each predicted meteorological event and the matching coefficient corresponding to each core meteorological element, thereby ensuring the accuracy of the first target coefficient.

[0088] At this time, the system needs to comprehensively consider the priorities of multiple predicted meteorological events and the matching coefficients of their respective core meteorological elements and meteorological data sets to determine a value called the "first target coefficient", which reflects the importance of each predicted meteorological event and its degree of matching with the current meteorological conditions; at this time, the priority is pre-set based on factors such as the potential impact, urgency or probability of occurrence of the meteorological event; high-priority events usually mean greater threats or higher levels of attention; the matching coefficient is calculated in step S151, reflecting the degree of fit between the core meteorological elements and the meteorological data set; the system will use weighted average, multiplication or other mathematical methods to combine the priority and matching coefficient to obtain the first target coefficient. The specific calculation method of this coefficient varies from system to system, but it is generally intended to balance the importance of the event and its degree of matching with the current meteorological conditions.

[0089] At the same time, meteorological detectors (such as cameras, radars, lidars, etc.) are in working condition, monitoring the external environment in real time. These devices can capture the visual characteristics of meteorological phenomena, such as cloud morphology, precipitation type, wind speed and direction, etc., and record this information in the form of images; at the same time, meteorological detectors collect environmental data through sensor arrays, including temperature, humidity, air pressure, wind speed, wind direction, precipitation, etc.; based on the collected data, meteorological detectors use imaging devices such as cameras or radars to generate environmental images. These images provide an intuitive representation of meteorological phenomena, which helps the system or operators to understand the current weather conditions more intuitively.

[0090] Specifically, suppose there is a weather warning system that predicts three weather events that will occur in the next three days: heavy rain, typhoon, and high temperature. The system first considers the priority of these three events. Assume that the priority of heavy rain is "high", the priority of typhoon is "medium", and the priority of high temperature is "low". Then, the system evaluates the degree of consistency between the core meteorological elements of each event and the current meteorological data set based on the matching coefficient calculated in step S151. For example, the matching coefficients of the precipitation and precipitation intensity elements of the heavy rain event with the data set are 0.9 and 0, respectively. .8; the matching coefficients of wind speed, wind direction and air pressure elements of typhoon events are 0.7, 0.6 and 0.9 respectively; the matching coefficients of maximum temperature and duration elements of high temperature events are 0.8 and 0.7 respectively; the system uses a weighted average method to calculate the first target coefficient, where the weights of priority and matching coefficient are set according to system design and business requirements; for example, assuming that the weights of priority and matching coefficient are 0.6 and 0.4 respectively, the first target coefficient of heavy rain events is calculated as (0.6 high priority weight + 0.4 (0.9 + 0.8) / 2) = high value (the specific value depends on the weight distribution); the first target coefficients of typhoon and high temperature events are calculated according to the corresponding methods and weights.

[0091] At the same time, the meteorological detector is monitoring the external environment in real time; for example, the camera captured images of gathering clouds and increasing precipitation, which suggests that heavy rain is about to occur; the radar equipment also captured precipitation echoes, further confirming the existence of heavy rain. These environmental images are recorded by the system and combined with the information of predicted meteorological events to provide a more comprehensive description of the meteorological conditions; through step S152, the system not only considers the priority and matching coefficient of the predicted meteorological events, but also captures environmental images in real time through the meteorological detector, providing richer information for subsequent decision-making. This information will be used in step S153 to determine the final target meteorological event.

[0092] Therefore, the second target coefficient is determined based on the priority of each predicted meteorological event and the environmental image detected by the meteorological detector, and the target meteorological event is determined according to the mapping relationship between the first target coefficient, the second target coefficient and the target meteorological event. The overall consideration of the mapping relationship between the first target coefficient, the second target coefficient and the target meteorological event is compatible to ensure the accuracy of the target meteorological event. At the same time, a meteorological event list is introduced to further control each predicted meteorological event, and the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector are considered as a whole to ensure the accurate control of the target meteorological event.

[0093] At this time, the system needs to comprehensively consider the priority of each predicted meteorological event and the environmental image detected by the meteorological detector to determine a value called the "second target coefficient". This coefficient reflects the consistency or degree of fit between the predicted meteorological event and the actual observed meteorological conditions; at this time, the priority is the same as in step S152, which is pre-set based on factors such as the potential impact, urgency or probability of occurrence of the meteorological event; the environmental image is a visual representation of the external environment captured in real time by the meteorological detector, including cloud morphology, precipitation type, wind speed and direction and other meteorological phenomena; the system will use image recognition, machine learning algorithms or expert systems to analyze the environmental image, extract key meteorological features, and compare them with the core meteorological elements of the predicted meteorological event; based on the comparison results, the system calculates a second target coefficient that reflects the consistency between the prediction and the actual situation.

[0094] The system needs to combine the first target coefficient calculated in step S152 and the second target coefficient calculated in step S153, as well as a predefined target meteorological event mapping relationship to determine the final target meteorological event; at this time, the first target coefficient reflects the priority of the predicted meteorological event and its degree of matching with the current meteorological data set; the second target coefficient reflects the consistency between the predicted meteorological event and the actual observed meteorological conditions; the target meteorological event mapping relationship is a predefined rule or model used to map the calculated coefficient to a specific meteorological event; it is constructed based on historical data, expert knowledge or machine learning algorithms; the system will use weighted average, multiplication or other mathematical methods to combine the first target coefficient and the second target coefficient to obtain a comprehensive score; then, the system compares this comprehensive score with the target meteorological event mapping relationship to determine the final target meteorological event.

[0095] Specifically, a second target coefficient matching table is collected, and the second target coefficient matching table is shown in Table 5: Table 5 Second target coefficient matching table Predicting meteorological events Priority Environmental image features Second target coefficient rainstorm high Dense clouds and heavy precipitation 0.9 typhoon middle Swirling clouds and fast winds 0.7 high temperature Low Clear sky, no clouds 0.5 Assume the following situation: predicted meteorological events: heavy rain, typhoon, high temperature; priority: heavy rain (high), typhoon (medium), high temperature (low); environmental image features detected by the meteorological detector: dense clouds and significantly increased precipitation; according to the matching table, the predicted meteorological event that best matches the feature of "dense clouds and significantly increased precipitation" is heavy rain, and its second target coefficient is 0.9.

[0096] Next, assume that the first target coefficient has been calculated: heavy rain: 0.8; typhoon: 0.6; high temperature: 0.4; now, combine the first target coefficient and the second target coefficient, as well as a predefined target meteorological event mapping relationship (here simplified to a threshold, for example, an event with a comprehensive score greater than 0.75 is considered a target meteorological event) to determine the final target meteorological event; the comprehensive score is simply calculated as the average of the first target coefficient and the second target coefficient (or other weighted method): heavy rain comprehensive score: (0.8 + 0.9) / 2 = 0.85; typhoon comprehensive score: (0.6 + 0.7) / 2 = 0.65; high temperature comprehensive score: (0.4 + 0.5) / 2 = 0.45; since heavy rain has the highest comprehensive score and exceeds the threshold of 0.75, heavy rain is determined as the target meteorological event.

[0097] See also Figure 7 , Figure 7 : is a schematic diagram of the structural composition of a weather forecasting system based on multimodal weather data of deep learning in an embodiment of the present invention; the weather forecasting system based on multimodal weather data of deep learning includes: A meteorological detection module 21 is used to collect the position of a meteorological detector and trigger meteorological detection of the meteorological detector according to the position of the meteorological detector to determine a meteorological data set; A meteorological data combination module 22 is used to determine multiple meteorological data combinations of different dimensions based on the division of the meteorological data set, and input them into a preset meteorological deep learning model; a multimodal meteorological data module 23 for determining meteorological features based on the recognition of multiple meteorological data combinations by the meteorological deep learning model, and determining multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location; A meteorological event list module 24 is configured to determine a meteorological event list based on the detection of multimodal meteorological data, and mark the priority of each predicted meteorological event in the meteorological event list and the corresponding core meteorological element; The target meteorological event module 25 is used to determine the corresponding matching coefficient based on the matching of each core meteorological element with the meteorological data set, and to determine the target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector.

[0098] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A weather forecasting method based on multimodal weather data based on deep learning, characterized in that: include: Collecting the position of the meteorological detector and triggering meteorological detection of the meteorological detector according to the position of the meteorological detector to determine the meteorological data set; Determine multiple meteorological data combinations of different dimensions based on the division of the meteorological data set, and input them into a preset meteorological deep learning model; Determining meteorological features based on recognition of a combination of multiple meteorological data by the meteorological deep learning model, and determining multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location; Determine a meteorological event list based on the detection of multimodal meteorological data, and mark the priority and corresponding core meteorological elements of each predicted meteorological event in the meteorological event list; The corresponding matching coefficient is determined based on the matching of each core meteorological element with the meteorological data set, and the target meteorological event is determined according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector.

2. The weather forecasting method based on multimodal weather data of deep learning according to claim 1, characterized in that: The collecting the position of the meteorological detector and triggering the meteorological detector to detect the meteorological data according to the position of the meteorological detector to determine the meteorological data set includes: Determine the location of the meteorological detector according to the name of the meteorological detector and the meteorological detection database, collect the detection task list of the meteorological detector, and determine the current detection task of the meteorological detector according to the detection of the detection task list of the meteorological detector; Determining a detection mode of the meteorological detector based on a current detection task of the meteorological detector and a position of the meteorological detector, and triggering online meteorological detection of the meteorological detector according to the detection mode; Real-time monitoring of the online meteorological detection of the meteorological detector collects multiple meteorological data based on the online meteorological detection of the meteorological detector, and determines the meteorological data set according to the multiple meteorological data, the corresponding directions and the detection time. At this time, each meteorological data, the corresponding direction and the detection time are in a corresponding relationship.

3. The weather forecasting method based on multimodal weather data of deep learning according to claim 1, characterized in that: The method of determining multiple meteorological data combinations of different dimensions based on the division of the meteorological data set and inputting them into a preset meteorological deep learning model includes: Collecting a meteorological data set, and determining a plurality of meteorological data combinations according to the division of the meteorological data set; Mark multiple meteorological data combinations and mark the corresponding meteorological dimensions. The meteorological dimensions of multiple meteorological data combinations are inconsistent. The meteorological dimensions include thermodynamic dimension, dynamic dimension, cloud physics dimension and pressure dimension. Based on the name and location of the meteorological detector, the corresponding detection space is determined, and the preset meteorological deep learning model is determined based on the tracing of the detection space. Each meteorological data combination is input into the preset meteorological deep learning model to obtain the output result of the meteorological deep learning model.

4. The weather forecasting method based on multimodal weather data of deep learning according to claim 1, characterized in that: The method of determining meteorological features based on the recognition of multiple meteorological data combinations by the meteorological deep learning model, and determining multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location, includes: When each meteorological data combination is input into a preset meteorological deep learning model, multiple identification channels are determined based on the detection of the meteorological deep learning model, and the status of the multiple identification channels is marked. The recognition order of each meteorological data combination by the meteorological deep learning model is determined based on the status of the multiple identification channels and each meteorological data combination.

5. The weather forecasting method based on multimodal weather data of deep learning according to claim 4, characterized in that: The method further includes: determining meteorological features based on the recognition of multiple meteorological data combinations by the meteorological deep learning model, and determining multimodal meteorological data based on the multiple meteorological features, the location of the meteorological detector, and past meteorological events at the location. Following this recognition sequence, the meteorological deep learning model is triggered to autonomously recognize each meteorological data combination and output corresponding meteorological features in sequence. The first form of the meteorological data is determined based on the multiple meteorological features and the location of the meteorological detector; A second form of the meteorological data is determined based on the location of the meteorological detector and past meteorological events at the location, and multimodal meteorological data is determined based on a synthesis of the first form of the meteorological data and the second form of the meteorological data.

6. The weather forecasting method based on multimodal weather data of deep learning according to claim 1, characterized in that: The method of determining a meteorological event list based on the detection of multimodal meteorological data and marking the priority and corresponding core meteorological elements of each predicted meteorological event in the meteorological event list includes: Collect multimodal meteorological data, determine multiple prediction combinations based on the detection of multimodal meteorological data, determine corresponding predicted meteorological events based on the identification of multiple prediction combinations, collect multiple predicted meteorological events, and determine a meteorological event list based on the combination of multiple predicted meteorological events.

7. The weather forecasting method based on multimodal weather data of deep learning according to claim 6, characterized in that: The method of determining a meteorological event list based on the detection of multimodal meteorological data and marking the priority and corresponding core meteorological elements of each predicted meteorological event in the meteorological event list further includes: In the meteorological event list, first priority coefficients of the plurality of predicted meteorological events are determined based on a match between the plurality of predicted meteorological events and the meteorological data, and priorities of the respective predicted meteorological events are determined based on a mapping relationship among the event types, the first priority coefficients, and the priorities of the plurality of predicted meteorological events; Multiple predicted meteorological events are sorted in order of priority. At this time, a core element detection method is determined according to the order of priority, the event content of the multiple predicted meteorological events and the correlation coefficients between the multiple predicted meteorological events, and the multiple predicted meteorological events are autonomously detected based on the core element detection method to determine the core meteorological elements corresponding to each predicted meteorological event.

8. The weather forecasting method based on multimodal weather data of deep learning according to claim 1, characterized in that: The method includes determining a corresponding matching coefficient based on the matching of each core meteorological element with the meteorological data set, and determining a target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element, and the environmental image detected by the meteorological detector, including: The core meteorological elements corresponding to each predicted meteorological event are collected, the core meteorological element combination is determined based on the combination of each core meteorological element, and the core meteorological element combination is dynamically matched with the meteorological data set to determine the matching coefficient of each core meteorological element and the meteorological data set.

9. The weather forecasting method based on multimodal weather data of deep learning according to claim 8, characterized in that: The method further includes determining a corresponding matching coefficient based on the matching of each core meteorological element with the meteorological data set, and determining a target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element, and the environmental image detected by the meteorological detector. Among the multiple predicted meteorological events, a first target coefficient is determined based on the priority of each predicted meteorological event and the matching coefficient corresponding to each core meteorological element. At the same time, the meteorological detector performs environmental detection on the external environment in a working state to collect an environmental image detected by the meteorological detector; The second target coefficient is determined based on the priority of each predicted meteorological event and the environmental image detected by the meteorological detector, and the target meteorological event is determined according to the mapping relationship among the first target coefficient, the second target coefficient and the target meteorological event.

10. A weather forecasting system based on multimodal weather data based on deep learning, characterized in that: The meteorological forecasting system based on multimodal meteorological data of deep learning is applied to the meteorological forecasting method based on multimodal meteorological data of deep learning according to any one of claims 1 to 9, and the meteorological forecasting system based on multimodal meteorological data of deep learning includes: A meteorological detection module is used to collect the position of the meteorological detector and trigger the meteorological detection of the meteorological detector according to the position of the meteorological detector to determine the meteorological data set; A meteorological data combination module is used to determine multiple meteorological data combinations of different dimensions based on the division of the meteorological data set, and input them into a preset meteorological deep learning model; A multimodal meteorological data module is configured to determine meteorological characteristics based on the meteorological deep learning model's recognition of multiple meteorological data combinations, and to determine multimodal meteorological data based on the multiple meteorological characteristics, the location of the meteorological detector, and past meteorological events at the location; A meteorological event list module is used to determine a meteorological event list based on the detection of multimodal meteorological data, and mark the priority of each predicted meteorological event in the meteorological event list and the corresponding core meteorological elements; The target meteorological event module is used to determine the corresponding matching coefficient based on the matching of each core meteorological element with the meteorological data set, and to determine the target meteorological event according to the priority of each predicted meteorological event, the matching coefficient corresponding to each core meteorological element and the environmental image detected by the meteorological detector.

Images