Risk prediction method applied to typhoon rainstorm weather and related equipment

Abstract

The embodiment of the invention belongs to the technical field of artificial intelligence, and relates to a risk prediction method and related equipment applied to typhoon rainstorm weather, and the method comprises the steps: calculating the product of the danger density and the danger rate according to the current accident data and the current environment data, and obtaining a three-dimensional risk index; performing sliding time window cutting processing on a typhoon path and rainfall intensity of the dynamic meteorological data to obtain time sequence characteristic data; carrying out dangerous case prediction processing on the three-dimensional risk index according to an integrated learning part to obtain a basic prediction result; performing time sequence correction processing on the time sequence characteristic data according to the recurrent neural network part to obtain time sequence correction data; and performing weighted fusion processing on the basic prediction result according to the time sequence correction data to obtain a target prediction result. The method can be used for carrying out related risk prediction processing in a financial science and technology business system, and the risk condition of the target grid can be reflected more comprehensively and accurately.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a risk prediction method and related equipment for typhoon and rainstorm weather. Background Technology

[0002] Accurate risk prediction is crucial for taking preventative measures and minimizing losses when dealing with various natural disasters and emergencies.

[0003] Traditional risk prediction methods often consider only a single type of data, such as relying solely on meteorological data or historical incident data, neglecting the correlation and complementarity between different data. For example, predicting the potential impact of a typhoon based solely on meteorological data may fail to fully account for the amplifying or mitigating effects of local environmental factors and historical incidents on risk; while relying solely on historical incident data makes it difficult to capture the dynamic impact of real-time weather changes on risk.

[0004] This shows that traditional risk prediction methods suffer from low prediction accuracy. Summary of the Invention

[0005] The purpose of this application is to propose a risk prediction method and related equipment for typhoon and rainstorm weather, so as to solve the problem of low prediction accuracy in traditional risk prediction methods.

[0006] To address the aforementioned technical problems, this application provides a risk prediction method for typhoon and rainstorm weather, employing the following technical solution: Real-time collection of dynamic meteorological data, current incident data, and current environmental data of the target grid location; The three-dimensional risk index is obtained by multiplying the hazard density and the hazard rate based on the current incident data and the current environmental data. The typhoon path and rainfall intensity of the dynamic meteorological data are processed by sliding time window segmentation to obtain time-series feature data; The trained hybrid prediction model is invoked, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; Based on the integrated learning component, the three-dimensional risk index is processed for risk prediction to obtain basic prediction results; The time-series feature data is subjected to time-series correction processing based on the recurrent neural network part to obtain time-series corrected data; The target prediction result is obtained by weighted fusion processing of the basic prediction result based on the time-series corrected data.

[0007] To address the aforementioned technical problems, this application also provides a risk prediction device for typhoon and rainstorm weather, employing the following technical solution: The real-time data acquisition module is used to collect dynamic meteorological data, current incident data, and current environmental data of the target grid location in real time. The three-dimensional risk index calculation module is used to calculate the product of the hazard density and the hazard rate based on the current incident data and the current environmental data to obtain the three-dimensional risk index. The time-series feature acquisition module is used to perform sliding time window segmentation on the typhoon path and rainfall intensity of the dynamic meteorological data to obtain time-series feature data. The model invocation module is used to invoke a trained hybrid prediction model, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; The hazard prediction module is used to perform hazard prediction processing on the three-dimensional risk index based on the integrated learning part to obtain basic prediction results; The timing correction module is used to perform timing correction processing on the timing feature data based on the recurrent neural network part to obtain timing corrected data; The weighted fusion module is used to perform weighted fusion processing on the basic prediction results based on the time-series corrected data to obtain the target prediction result.

[0008] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the risk prediction method applied to typhoon and rainstorm weather as described above.

[0009] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the risk prediction method for typhoon and rainstorm weather as described above.

[0010] This application provides a risk prediction method for typhoon and rainstorm weather, comprising: real-time acquisition of dynamic meteorological data, current hazard data, and current environmental data of a target grid location; calculation of the product of hazard density and hazard rate based on the current hazard data and the current environmental data to obtain a three-dimensional risk index; processing the typhoon path and rainfall intensity of the dynamic meteorological data through a sliding time window to obtain time-series feature data; invoking a trained hybrid prediction model, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; performing hazard prediction processing on the three-dimensional risk index based on the ensemble learning part to obtain a basic prediction result; performing time-series correction processing on the time-series feature data based on the recurrent neural network part to obtain time-series corrected data; and performing weighted fusion processing on the basic prediction result based on the time-series corrected data to obtain a target prediction result. Compared with the prior art, this application can more comprehensively and accurately reflect the risk status of the target grid; in addition, by invoking a trained hybrid prediction model, utilizing the ensemble learning part to process non-time-series three-dimensional risk index data, and utilizing the recurrent neural network part to process time-series feature data, the advantages of the two models in processing different types of data are fully utilized, improving the accuracy and reliability of the prediction. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an exemplary system architecture diagram in which this application is applied; Figure 2 This is a flowchart illustrating the implementation of the risk prediction method for typhoon and rainstorm weather provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of the risk prediction device for typhoon and rainstorm weather provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein are combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, the system architecture 100 includes a terminal device 101, a network 102, and a server 103. The terminal device 101 is a laptop computer 1011, a tablet computer 1012, or a mobile phone 1013. The network 102 is used as a medium to provide a communication link between the terminal device 101 and the server 103. The network 102 includes various connection types, such as wired and wireless communication links or fiber optic cables, etc.

[0017] Users use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications are installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 is a variety of electronic devices with a display screen that support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 is a server that provides various services, such as a backend server that supports the pages displayed on terminal device 101.

[0020] It should be noted that the risk prediction method for typhoon and rainstorm weather provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the risk prediction device for typhoon and rainstorm weather is generally installed in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Any number of terminal devices, networks, and servers can be included depending on implementation needs.

[0022] Continue to refer to Figure 2 The diagram shows a flowchart of an embodiment of the risk prediction method for typhoon and rainstorm weather according to this application. The above-described risk prediction method for typhoon and rainstorm weather includes steps S201, S202, S203, S204, S205, S206, and S207.

[0023] In step S201, dynamic meteorological data, current incident data, and current environmental data of the target grid location are collected in real time.

[0024] In this application embodiment, the above-mentioned input data may be the specific content that the user wants the system to answer. Specifically, the input data is "transaction data, payment data, business data, or purchase data" related to financial institutions (such as banks). The input data may also be medical data related to medical scenarios. For example, personal health records, prescriptions, examination reports, etc. It should be understood that the examples of input data here are only for ease of understanding and are not intended to limit this application.

[0025] In this embodiment of the application, dynamic meteorological data includes, but is not limited to, typhoon path, rainfall intensity, wind speed, and temperature.

[0026] In this application embodiment, the current accident data is the number of accidents and the type of accidents that have occurred at the target grid location within a certain period of time. Among them, the accident data is auto insurance business data related to financial institutions (such as banks). For example, it includes vehicle accident records, geographical distribution of policies, vehicle model risk coefficients, etc.

[0027] In this embodiment, the current environmental data includes topography, soil moisture, and vegetation cover. Multiple sensing devices and data acquisition systems are used to ensure the real-time nature and accuracy of the data.

[0028] In this embodiment, the application uses various sensing devices installed at the target grid location, such as weather stations, cameras, and soil moisture sensors, to collect dynamic meteorological data and current environmental data in real time. The collected data is transmitted to a data processing center for storage and processing via a wireless communication network.

[0029] In this embodiment of the application, the above-mentioned dynamic meteorological data can also be obtained by interfacing with the meteorological bureau's API.

[0030] In this embodiment of the application, the current accident data can be obtained by calling the system database and obtaining historical accident data of the target grid location from the system database.

[0031] In step S202, the product of the hazard density and the hazard rate is calculated based on the current hazard data and the current environmental data to obtain the three-dimensional risk index.

[0032] In the embodiments of this application, the hazard density refers to the number of hazards occurring per unit area or per unit time.

[0033] In this embodiment of the application, the incident rate refers to the probability of an incident occurring.

[0034] In this embodiment of the application, by multiplying these two indicators, the combined impact of historical incidents and environmental factors on risk can be comprehensively considered to obtain an index that more comprehensively reflects the risk status of the target grid.

[0035] In this embodiment, after receiving current incident data and current environmental data, the application calculates incident density and incident rate according to a preset algorithm, and multiplies them to obtain a three-dimensional risk index. For example, incident density can be obtained by counting the number of incidents at the target grid location within a certain period of time and dividing by the grid area; incident rate can be estimated based on historical data and current environmental factors.

[0036] In step S203, the typhoon path and rainfall intensity of the dynamic meteorological data are processed by sliding time window segmentation to obtain time-series feature data.

[0037] In this embodiment, the sliding time window is set according to actual needs and data characteristics, such as every hour, every half hour, or a shorter time interval.

[0038] In this embodiment of the application, by using a sliding time window to divide continuous meteorological data into multiple time segments, the features within each time segment can be extracted, such as the changing trend of typhoon path, the peak value of rainfall intensity and duration, etc., thereby better capturing the time dependence of meteorological data.

[0039] In this embodiment, the application performs sliding time window segmentation on the collected dynamic meteorological data, including typhoon paths and rainfall intensity. For example, the sliding time window is set to one hour, dividing the continuous meteorological data into multiple one-hour time segments, and extracting features within each time segment, such as the average movement speed of the typhoon path, the maximum value and average value of rainfall intensity, etc.

[0040] In step S204, the trained hybrid prediction model is invoked, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part.

[0041] In the embodiments of this application, the hybrid prediction model refers to the combination of XGBoost (ensemble learning part) and LSTM (recurrent neural network part), which aims to integrate the feature selection and ensemble learning advantages of XGBoost and the time series modeling capabilities of LSTM for time series prediction.

[0042] In the embodiments of this application, the ensemble learning part (eXtreme Gradient Boosting, XGBoost) refers to an ensemble learning algorithm based on the gradient boosting framework, which has the advantages of high efficiency, flexibility and ability to handle high-dimensional data, and is suitable for processing non-time-series risk index data.

[0043] In this embodiment, the Long Short-Term Memory (LSTM) part refers to a recurrent neural network, which can effectively handle long-term dependencies in time-series data and is suitable for processing the time-series characteristics of dynamic meteorological data. By combining these two models, their advantages in processing different types of data are fully utilized.

[0044] In this embodiment, before performing risk prediction, the hybrid prediction model needs to be trained. Historical data is collected, including dynamic meteorological data, current incident data, current environmental data, and corresponding risk results, and the data is divided into training and testing sets. The hybrid prediction model is trained using the training set, and the model's parameters are adjusted to enable it to accurately predict risks. After training, the trained model is stored in the system and invoked when performing real-time risk prediction.

[0045] In step S205, the three-dimensional risk index is processed for hazard prediction based on the ensemble learning part to obtain the basic prediction result.

[0046] In this embodiment of the application, the integrated learning part can predict the probability or risk level of a dangerous situation occurring at a target grid location in the future by performing feature analysis and model training on the three-dimensional risk index, and use this as the basic prediction result.

[0047] In this embodiment of the application, the calculated three-dimensional risk index is input into the XGBoost part. The XGBoost part performs feature analysis and model calculation on the three-dimensional risk index and outputs basic prediction results, such as the probability of a dangerous situation occurring at the target grid location in the next 24 hours.

[0048] In step S206, the time-series feature data is processed by the recurrent neural network to obtain time-series corrected data.

[0049] In this embodiment, the recurrent neural network (RNN) part learns and analyzes temporal feature data to capture the temporal variation patterns of meteorological data and corrects the basic prediction results. For example, if current meteorological data shows that the typhoon path will change or the rainfall intensity will increase significantly, the LSTM part adjusts the basic prediction results based on these temporal features to make them more consistent with the actual situation.

[0050] In this embodiment of the application, the extracted time-series feature data is input into the LSTM part. The LSTM part learns and analyzes the time-series feature data to capture the temporal variation pattern of meteorological data and outputs time-series correction data, such as the adjustment range of the basic forecast result based on the changes in the current meteorological data.

[0051] In step S207, the basic prediction results are weighted and fused based on the time-series correction data to obtain the target prediction results.

[0052] In this embodiment, the weights of the weighted fusion are set according to actual needs and data importance. For example, based on the degree of impact of time-series feature data on risk, the time-series corrected data output by the LSTM is given a larger weight. Through weighted fusion, the basic prediction results and time-series corrected data can be organically combined to obtain more accurate and comprehensive target risk prediction results.

[0053] In this embodiment, the application performs weighted fusion processing on the basic prediction result and the time-series corrected data according to preset weights to obtain the target risk prediction result. For example, if the basic prediction result is that the probability of a dangerous situation occurring is 30%, the adjustment range of the time-series corrected data is 20%, and the weight of the time-series corrected data is 0.6, and the weight of the basic prediction result is 0.4, then the target risk prediction result is: 30%×0.4+(30%+20%)×0.6=42%.

[0054] In practical applications, assuming Typhoon Hagasha is expected to make landfall 3 hours beforehand, real-time meteorological data (wind speed of level 12, 3-hour cumulative rainfall of 80mm), historical accident data (18% car insurance claim rate during last year's typhoon season in this area), and environmental data (high building density, aging drainage system) for a target grid (such as Tianhe District, Guangzhou) are calculated. Based on this data, the hazard density is calculated to be 1.2 times / year·km², the accident rate to be 25%, and the three-dimensional risk index to be 0.3. A sliding time window is then applied to this real-time meteorological data to determine the typhoon's path will move westward in the next hour. If the typhoon path shifts 5km westward and the rainfall intensity increases by 10mm per hour, the above three-dimensional risk index is input into the ensemble learning part of the hybrid prediction model, resulting in a predicted basic claim probability of 30%. Based on the recurrent neural network part, the time-series correction processing of "the typhoon path will shift westward by 5km in the next hour and the rainfall intensity will increase by 10mm per hour" obtained from the above sliding time window segmentation process is performed, resulting in a predicted typhoon path shift that will lead to a 5% increase in the claim probability and a 3% increase in the rainfall intensity. Therefore, the final target prediction result is 38% = 30% + 5% + 3%.

[0055] This application provides a risk prediction method for typhoon and rainstorm weather, comprising: real-time acquisition of dynamic meteorological data, current hazard data, and current environmental data of the target grid location; calculation of the product of hazard density and hazard rate based on the current hazard data and current environmental data to obtain a three-dimensional risk index; processing the typhoon path and rainfall intensity of the dynamic meteorological data through a sliding time window to obtain time-series feature data; calling a trained hybrid prediction model, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; performing hazard prediction processing on the three-dimensional risk index based on the ensemble learning part to obtain a basic prediction result; performing time-series correction processing on the time-series feature data based on the recurrent neural network part to obtain time-series corrected data; and performing weighted fusion processing on the basic prediction result based on the time-series corrected data to obtain the target prediction result. Compared with the prior art, this application can more comprehensively and accurately reflect the risk status of the target grid; in addition, by calling a trained hybrid prediction model, using the XGBoost part to process non-time-series three-dimensional risk index data and the LSTM part to process time-series feature data, the advantages of the two models in processing different types of data are fully utilized, improving the accuracy and reliability of the prediction.

[0056] In some optional implementations of the embodiments of this application, the following steps are included before the step of calling the trained hybrid prediction model: Acquire historical meteorological data, historical incident data, and historical environmental data corresponding to the target grid location; Historical meteorological data, historical accident data, and historical environmental data are preprocessed to obtain preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data. Historical static features are constructed based on preprocessed historical accident data and preprocessed historical environmental data; Historical dynamic characteristics are constructed based on historical meteorological data; Construct an initial hybrid prediction model, which includes an input layer, a feature processing layer, a fusion layer, and an output layer. The feature processing layer includes an initial ensemble learning part and an initial recurrent neural network part. The initial ensemble learning part is trained using the first model based on historical static features. The initial recurrent neural network part is trained using a second model based on historical dynamic characteristics. Once the training processes for the first and second models are completed, a well-trained hybrid prediction model is obtained.

[0057] In this embodiment, historical meteorological data, historical accident data, and historical environmental data can be obtained from meteorological departments, insurance company databases, geographic information systems (GIS), and other sources. Historical meteorological data includes data on temperature, humidity, precipitation, wind speed, and other data that change over time; historical accident data records insurance claim events that occurred at the target grid location over a period of time, such as accident type, time of incident, and claim amount; historical environmental data covers information such as topography, distribution of surrounding facilities, and population density.

[0058] In the embodiments of this application, the preprocessing operations include data cleaning (removing outliers, filling in missing values, etc.), data standardization (unifying data of different dimensions to the same range), feature encoding (encoding categorical variables), etc., to improve the quality and usability of the data.

[0059] In this embodiment of the application, historical static features reflect the relatively stable attributes of the target grid location. For example, a static feature reflecting the accident-prone type in the area can be constructed by analyzing the distribution of accident types in historical accident data and the topography in historical environmental data; or a static feature representing the potential risk level of the area can be constructed based on the distribution of surrounding facilities and population density.

[0060] In the embodiments of this application, historical dynamic features capture factors that affect the risk of the target grid location over time. For example, precipitation data over a period of time can be processed according to a time series to construct a dynamic feature that reflects the trend of precipitation changes; or a dynamic feature representing the degree of wind force influence can be constructed based on changes in wind speed data.

[0061] In this embodiment, the initial hybrid prediction model includes an input layer, a feature processing layer (including an initial ensemble learning part and an initial recurrent neural network part), a fusion layer, and an output layer. The input layer receives preprocessed historical static features and historical dynamic features. The initial ensemble learning part of the feature processing layer (such as the XGBoost algorithm) is suitable for processing static features and can capture complex nonlinear relationships between features. The initial recurrent neural network part (such as the LSTM algorithm) is good at processing time series data and can effectively process historical dynamic features. The fusion layer merges the outputs of the ensemble learning part and the recurrent neural network part to comprehensively consider the impact of static and dynamic features on risk. The output layer outputs the risk prediction results for the target grid location.

[0062] In the embodiments of this application, during the training process, historical static features are used as input, and known risk labels (such as whether an accident has occurred) are used as targets. The parameters of the ensemble learning part are adjusted by an optimization algorithm so that it can accurately predict risk situations based on static features.

[0063] In this embodiment of the application, similarly, using historical dynamic features as input and risk labels as targets, the parameters of the recurrent neural network are optimized so that it can capture the impact of changes in dynamic features over time on risk.

[0064] Compared with existing technologies, this application improves the accuracy of risk prediction for target grid locations by integrating historical meteorological data, historical incident data, and historical environmental data to construct historical static and dynamic features, and by using a hybrid prediction model for training and prediction.

[0065] In some optional implementations of the embodiments of this application, the steps of preprocessing historical meteorological data, historical accident data, and historical environmental data to obtain preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data specifically include the following steps: Time alignment processing is performed on historical meteorological data, historical accident data, and historical environmental data; and / or Missing values ​​were handled for historical meteorological data, historical accident data, and historical environmental data; and / or Outlier correction processing is performed on historical meteorological data, historical accident data, and historical environmental data to obtain preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data.

[0066] In the embodiments of this application, various types of data can be unified to the same time scale according to actual time granularity requirements. For example, if historical meteorological data is recorded in hours and historical accident data only records the date, the historical meteorological data can be aggregated by day to calculate the average meteorological indicators (such as average temperature, average humidity, etc.) for each day, making it consistent with the time granularity of the historical accident data. For historical environmental data, if its changes are relatively slow, its time information can be appropriately matched with meteorological data and accident data according to the changes in the data, ensuring that the information of different types of data can be accurately associated in the same time period during analysis.

[0067] In the embodiments of this application, appropriate processing methods can be adopted for different types of data and missing values. For numerical data, such as temperature and humidity in meteorological data, if there are few missing values, linear interpolation can be used to calculate the missing values ​​based on data from adjacent time points before and after the missing values; if there are many missing values, the overall mean or median of the data sequence can be used for filling. For categorical data, such as accident types in accident data, if there are missing values, they can be inferred and filled based on other relevant information (such as accident time and location), or the most frequently occurring category in the data can be used for filling. For missing values ​​in historical environmental data, data from similar surrounding areas can be used for reasonable estimation and filling.

[0068] In this application's embodiments, for numerical data, outliers can be identified by setting a reasonable threshold range. For example, for temperature in meteorological data, a reasonable temperature range is set based on historical data and meteorological common sense; values ​​exceeding this range are considered outliers. For outliers, correction methods can be used, such as correcting them to the mean or median of the data sequence, or making reasonable adjustments based on data from adjacent time points. For categorical data, if obviously illogical categories appear, such as negative claim amounts in accident data, they can be corrected to reasonable values ​​or marked for further processing. For outliers in historical environmental data, judgment and correction can be made by combining actual conditions and relevant expert knowledge.

[0069] Compared with existing technologies, this application obtains high-quality preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data by performing time alignment processing, missing value processing, and outlier correction processing on historical meteorological data, historical accident data, and historical environmental data, providing a reliable data foundation for subsequent data analysis and model building.

[0070] In some optional implementations of the embodiments of this application, the step of performing time-series correction processing on the time-series feature data based on the recurrent neural network to obtain time-series corrected data specifically includes the following steps: Spatial weights are constructed based on temporal feature data and a Gaussian kernel function, where the spatial weights are expressed as:

[0071] in, This indicates the time from the target grid location (x, y) to the typhoon center. Euclidean distance, This indicates the radius of influence of a typhoon. Time weights are constructed based on time-series characteristic data and a risk time-series decay function, where the real-time time weight is expressed as:

[0072]

[0073]

[0074] in, Indicates the time of typhoon landfall. This indicates a sensitive time window before login. Indicates the decay time constant; Spatial and temporal weights are fused using a spatiotemporal weighting process to obtain time-series corrected data.

[0075] In this application embodiment, the sensitive time window This refers to the 1 to 6 hours before a typhoon makes landfall, especially the 3 to 4 hours, which is the "critical point" when the typhoon's impact on land is most severe. At this time, the typhoon's eyewall (the area with the strongest winds) is about to cover the landfall point, and the intensity of wind and rain may suddenly increase.

[0076] In the embodiments of this application, the decay time constant It is a physical quantity that describes how quickly the intensity of a typhoon decreases over time. It refers to the time required for the central air pressure (or maximum wind speed) of a typhoon to decrease from its initial value to 1 / e (approximately 36.8%) of its initial value.

[0077] In some optional implementations of the embodiments of this application, time weighting In the context of typhoon movement speed exceeding 20 km / h, the decay time constant is... .

[0078] In this embodiment, the typhoon's moving speed (V) is related to its attenuation rate. Specifically, when the typhoon's moving speed is greater than 20 km / h, its path is usually relatively straight, with a short contact time with land, but it covers a wide area of ​​land per unit time. If its attenuation time constant is small (e.g., τ < 6 hours), it indicates that the typhoon can still maintain strong intensity during rapid movement, and its wind and rain impact range needs to be closely monitored. If τ is large (e.g., τ > 12 hours), the typhoon may gradually weaken during its movement, reducing the threat to subsequent landfall points. When the typhoon's moving speed is less than or equal to 20 km / h, it is prone to lingering on land, and its attenuation time constant is usually large (e.g., τ > 12 hours). However, prolonged stagnation may lead to extremely large local cumulative rainfall, causing severe urban flooding.

[0079] In this embodiment of the application, when the typhoon's moving speed is greater than 20 km / h, the weight coefficients in the hybrid prediction model can be dynamically adjusted by accelerating the decay time constant, thereby improving the model's prediction accuracy.

[0080] In some optional implementations of the embodiments of this application, after the step of constructing spatial weights based on time-series feature data and Gaussian kernel function, the following steps are further included: Obtain target terrain data corresponding to the target grid location based on the current environmental data; Spatial weights are optimized based on the digital elevation model and target terrain data.

[0081] In this embodiment, the application obtains target terrain data corresponding to the target grid location from a pre-constructed terrain database based on current environmental data. The target grid location refers to a specific spatial unit that requires spatial weight analysis and optimization; it can be a geographic grid, an administrative division unit, or other defined spatial area. For example, if current environmental data shows a change in the vegetation cover type of a certain area, terrain data matching the vegetation cover type of that area can be filtered from the terrain database based on this change, including information such as terrain slope, aspect, and altitude.

[0082] In this application embodiment, the digital elevation model (DEM) is mainly used to describe the digital model of terrain elevation, which can be obtained through geographic information system software or professional surveying equipment.

[0083] In this embodiment, spatial weights are optimized based on a digital elevation model and acquired target terrain data. Specifically, the impact of terrain features (such as slope, aspect, and elevation) in the target terrain data on spatial interaction is first analyzed. For example, ={1.2} Low-lying areas (elevation < 5m), ={1.0} Plains area, ={0.8} Mountainous area (slope > 25°).

[0084] Compared with existing technologies, this application can overcome the limitations of traditional spatial weight determination methods, improve the adaptability of spatial weights to actual terrain and environmental changes, and thus enhance the accuracy and reliability of spatial data analysis.

[0085] In some optional implementations of the embodiments of this application, the step of performing spatiotemporal weight fusion processing on spatial weights and temporal weights to obtain time-series corrected data specifically includes the following steps: A three-dimensional tensor is constructed based on spatial and temporal weights, where the three-dimensional tensor is represented as:

[0086] Normalizing the three-dimensional tensor yields temporally corrected data, which is represented as follows: .

[0087] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0088] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0089] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments is accomplished by instructing related hardware through computer-readable instructions. These computer-readable instructions may be stored in a computer-readable storage medium. When executed, the program may include the processes of the embodiments of the above methods. The aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0090] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order constraint on the execution of these steps, and they may be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures include multiple sub-steps or multiple stages, which are not necessarily completed at the same time, but are executed at different times, and their execution order is not necessarily sequential, but rather alternates or rotates with other steps or at least some of the sub-steps or stages of other steps.

[0091] Further reference Figure 3 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of a risk prediction device applied to typhoon and rainstorm weather. This device embodiment is similar to... Figure 2 Corresponding to the method embodiments shown, this device is specifically applied in various electronic devices.

[0092] like Figure 3 As shown, the risk prediction device 200 applied to typhoon and rainstorm weather according to an embodiment of this application includes: The real-time data acquisition module 210 is used to collect dynamic meteorological data, current incident data, and current environmental data of the target grid location in real time. The three-dimensional risk index calculation module 220 is used to calculate the product of the hazard density and the hazard rate based on the current incident data and the current environmental data to obtain the three-dimensional risk index. The time-series feature acquisition module 230 is used to perform sliding time window segmentation on the typhoon path and rainfall intensity of dynamic meteorological data to obtain time-series feature data. The model calling module 240 is used to call the trained hybrid prediction model, which includes an ensemble learning part and a recurrent neural network part. The hazard prediction module 250 is used to perform hazard prediction processing on the three-dimensional risk index based on the ensemble learning part to obtain basic prediction results. The timing correction module 260 is used to perform timing correction processing on the timing feature data based on the recurrent neural network part to obtain timing corrected data; The weighted fusion module 270 is used to perform weighted fusion processing on the basic prediction results based on the time-series corrected data to obtain the target prediction results.

[0093] In this embodiment, a risk prediction device 200 for typhoon and rainstorm weather is provided, comprising: a real-time data acquisition module 210 for real-time acquisition of dynamic meteorological data, current hazard data, and current environmental data of a target grid location; a three-dimensional risk index calculation module 220 for calculating the product of hazard density and hazard rate based on the current hazard data and current environmental data to obtain a three-dimensional risk index; a time-series feature acquisition module 230 for performing sliding time window segmentation on the typhoon path and rainfall intensity of the dynamic meteorological data to obtain time-series feature data; a model invocation module 240 for invoking a trained hybrid prediction model, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; a hazard prediction module 250 for performing hazard prediction processing on the three-dimensional risk index based on the ensemble learning part to obtain a basic prediction result; a time-series correction module 260 for performing time-series correction processing on the time-series feature data based on the recurrent neural network part to obtain time-series corrected data; and a weighted fusion module 270 for performing weighted fusion processing on the basic prediction result based on the time-series corrected data to obtain a target prediction result. Compared with existing technologies, this application can more comprehensively and accurately reflect the risk status of the target grid. In addition, by calling the trained hybrid prediction model, the XGBoost part is used to process non-time-series three-dimensional risk index data, and the LSTM part is used to process time-series feature data. This fully leverages the advantages of the two models in processing different types of data, thereby improving the accuracy and reliability of the prediction.

[0094] In some optional implementations of the embodiments of this application, the risk prediction device 200 applied to typhoon and rainstorm weather further includes: The historical data acquisition module is used to acquire historical meteorological data, historical accident data, and historical environmental data corresponding to the target grid location; The preprocessing module is used to preprocess historical meteorological data, historical accident data, and historical environmental data to obtain preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data. The historical static feature construction module is used to construct historical static features based on preprocessed historical accident data and preprocessed historical environmental data. The historical dynamic feature construction module is used to construct historical dynamic features based on historical meteorological data; The initial model building module is used to build an initial hybrid prediction model, which includes an input layer, a feature processing layer, a fusion layer, and an output layer. The feature processing layer includes an initial ensemble learning part and an initial recurrent neural network part. The first model training module is used to perform first model training processing on the initial ensemble learning part based on historical static features. The second model training module is used to perform second model training on the initial recurrent neural network part based on historical dynamic features. The training completion module is used to obtain the trained hybrid prediction model after the training processes of the first model and the second model are completed.

[0095] In some optional implementations of the embodiments of this application, the preprocessing module includes: The time alignment submodule is used to perform time alignment processing on historical meteorological data, historical accident data, and historical environmental data; and / or The missing value processing submodule is used to process missing values ​​in historical meteorological data, historical accident data, and historical environmental data; and / or The outlier correction submodule is used to perform outlier correction processing on historical meteorological data, historical accident data, and historical environmental data to obtain preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data.

[0096] In some optional implementations of the embodiments of this application, the timing correction module includes: The spatial weight construction submodule is used to construct spatial weights based on temporal feature data and a Gaussian kernel function. The spatial weights are represented as follows:

[0097] in, This indicates the time from the target grid location (x, y) to the typhoon center. Euclidean distance, This indicates the radius of influence of a typhoon. The time weight construction submodule is used to construct time weights based on time-series feature data and the risk time-series decay function. The real-time time weight is expressed as:

[0098]

[0099]

[0100] in, Indicates the time of typhoon landfall. This indicates a sensitive time window before login. Indicates the decay time constant; The spatiotemporal weight fusion submodule is used to perform spatiotemporal weight fusion processing on spatial weights and temporal weights to obtain time-series corrected data.

[0101] In some optional implementations of the embodiments of this application, the timing correction module further includes: The terrain data acquisition submodule is used to acquire target terrain data corresponding to the target grid location based on the current environmental data; The spatial weight optimization submodule is used to optimize the spatial weights based on the digital elevation model and the target terrain data.

[0102] In some optional implementations of the embodiments of this application, the above-mentioned spatiotemporal weight fusion submodule includes: A 3D tensor construction unit is used to construct a 3D tensor based on spatial and temporal weights, where the 3D tensor is represented as:

[0103] The normalization unit is used to normalize the three-dimensional tensor to obtain temporally corrected data, which is represented as follows: .

[0104] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of a computer device according to an embodiment of this application.

[0105] Computer device 300 includes a memory 310, a processor 320, and a network interface 330 that are interconnected via a system bus. It should be noted that only computer device 300 with components 310-330 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and alternatively, more or fewer components may be implemented. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0106] Computer equipment includes desktop computers, laptops, handheld computers, and cloud servers. Human-computer interaction occurs between computer equipment and users through methods such as keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0107] The memory 310 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 310 is an internal storage unit of the computer device 300, such as the hard disk or memory of the computer device 300. In other embodiments, the memory 310 is also an external storage device of the computer device 300, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 300. Of course, the memory 310 also includes both internal storage units and external storage devices of the computer device 300. In the embodiments of this application, the memory 310 is typically used to store the operating system and various application software installed on the computer device 300, such as computer-readable instructions for risk prediction methods applied to typhoon and rainstorm weather. In addition, the memory 310 is also used to temporarily store various types of data that have been output or will be output.

[0108] In some embodiments, processor 320 is a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. Processor 320 is typically used to control the overall operation of computer device 300. In embodiments of this application, processor 320 is used to execute computer-readable instructions stored in memory 310 or process data, such as executing computer-readable instructions applied to a risk prediction method for typhoon and rainstorm weather.

[0109] The network interface 330 may include a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the computer device 300 and other electronic devices.

[0110] The computer equipment provided in this application can more comprehensively and accurately reflect the risk status of the target grid. In addition, by calling the trained hybrid prediction model, the XGBoost part is used to process non-time-series three-dimensional risk index data, and the LSTM part is used to process time-series feature data. This fully leverages the advantages of the two models in processing different types of data, thereby improving the accuracy and reliability of the prediction.

[0111] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the risk prediction method applied to typhoon and rainstorm weather as described above.

[0112] The computer-readable storage medium provided in this application can more comprehensively and accurately reflect the risk status of the target grid. In addition, by calling the trained hybrid prediction model, the XGBoost part is used to process non-time-series three-dimensional risk index data, and the LSTM part is used to process time-series feature data. This fully leverages the advantages of the two models in processing different types of data, thereby improving the accuracy and reliability of the prediction.

[0113] Through the above description of the embodiments, those skilled in the art will clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, is embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (such as a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0114] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application is implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A risk prediction method applied to typhoon and rainstorm weather, characterized in that, Includes the following steps: Real-time collection of dynamic meteorological data, current incident data, and current environmental data of the target grid location; The three-dimensional risk index is obtained by multiplying the hazard density and the hazard rate based on the current incident data and the current environmental data. The typhoon path and rainfall intensity of the dynamic meteorological data are processed by sliding time window segmentation to obtain time-series feature data; The trained hybrid prediction model is invoked, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; Based on the integrated learning component, the three-dimensional risk index is processed for risk prediction to obtain basic prediction results; The time-series feature data is subjected to time-series correction processing based on the recurrent neural network part to obtain time-series corrected data; The target prediction result is obtained by weighted fusion processing of the basic prediction result based on the time-series corrected data.

2. The risk prediction method for typhoon and rainstorm weather according to claim 1, characterized in that, Before the step of invoking the trained hybrid prediction model, the following steps are also included: Acquire historical meteorological data, historical incident data, and historical environmental data corresponding to the target grid location; The historical meteorological data, the historical accident data, and the historical environmental data are preprocessed to obtain preprocessed historical meteorological data, preprocessed historical accident data, and preprocessed historical environmental data. Historical static features are constructed based on the preprocessed historical incident data and preprocessed historical environmental data; Historical dynamic features are constructed based on the historical meteorological data; Construct an initial hybrid prediction model, wherein the initial hybrid prediction model includes an initial ensemble learning part and an initial recurrent neural network part; The initial ensemble learning part is trained using the historical static features as a first model. The initial recurrent neural network part is trained using the historical dynamic features as a second model. After the training of the first model and the training of the second model are completed, the trained hybrid prediction model is obtained.

3. The risk prediction method for typhoon and rainstorm weather according to claim 2, characterized in that, The step of preprocessing the historical meteorological data, historical incident data, and historical environmental data to obtain preprocessed historical meteorological data, preprocessed historical incident data, and preprocessed historical environmental data specifically includes the following steps: Time alignment processing is performed on the historical meteorological data, the historical accident data, and the historical environmental data; and / or Missing values ​​are processed for the historical meteorological data, the historical accident data, and the historical environmental data. and / or Outlier correction processing is performed on the historical meteorological data, the historical accident data, and the historical environmental data to obtain the preprocessed historical meteorological data, the preprocessed historical accident data, and the preprocessed historical environmental data.

4. The risk prediction method for typhoon and rainstorm weather according to claim 1, characterized in that, The step of performing time-series correction processing on the time-series feature data based on the recurrent neural network portion to obtain time-series corrected data specifically includes the following steps: Spatial weights are constructed based on the time-series feature data and the Gaussian kernel function, wherein the spatial weights are expressed as: in, This indicates the time from the target grid location (x, y) to the typhoon center. Euclidean distance, This indicates the radius of influence of a typhoon. A time weight is constructed based on the aforementioned time-series feature data and the risk time-series decay function, wherein the real-time time weight is expressed as: in, Indicates the time of typhoon landfall. This indicates a sensitive time window before login. Indicates the decay time constant; The spatial weights and temporal weights are subjected to spatiotemporal weight fusion processing to obtain the time-series corrected data.

5. The risk prediction method for typhoon and rainstorm weather according to claim 4, characterized in that, Following the step of constructing spatial weights based on the temporal feature data and the Gaussian kernel function, the following step is also included: Based on the current environmental data, obtain the target terrain data corresponding to the target grid location; The spatial weights are optimized based on the digital elevation model and the target terrain data.

6. The risk prediction method for typhoon and rainstorm weather according to claim 4, characterized in that, In the time weight In the context of typhoon movement speed exceeding 20 km / h, the attenuation time constant... .

7. The risk prediction method for typhoon and rainstorm weather according to claim 4, characterized in that, The step of performing spatiotemporal weight fusion processing on the spatial weights and the temporal weights to obtain the time-series corrected data specifically includes the following steps: A three-dimensional tensor is constructed based on the spatial weights and the temporal weights, wherein the three-dimensional tensor is represented as: The three-dimensional tensor is normalized to obtain the temporal correction data, wherein the temporal correction data is represented as follows: 。 8. A risk prediction device for typhoon and rainstorm weather, characterized in that, include: The real-time data acquisition module is used to collect dynamic meteorological data, current incident data, and current environmental data of the target grid location in real time. The three-dimensional risk index calculation module is used to calculate the product of the hazard density and the hazard rate based on the current incident data and the current environmental data to obtain the three-dimensional risk index. The time-series feature acquisition module is used to perform sliding time window segmentation on the typhoon path and rainfall intensity of the dynamic meteorological data to obtain time-series feature data. The model invocation module is used to invoke a trained hybrid prediction model, wherein the hybrid prediction model includes an ensemble learning part and a recurrent neural network part; The hazard prediction module is used to perform hazard prediction processing on the three-dimensional risk index based on the integrated learning part to obtain basic prediction results; The timing correction module is used to perform timing correction processing on the timing feature data based on the recurrent neural network part to obtain timing corrected data; The weighted fusion module is used to perform weighted fusion processing on the basic prediction results based on the time-series corrected data to obtain the target prediction result.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, it implements the steps of the risk prediction method for typhoon and rainstorm weather as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the risk prediction method for typhoon and rainstorm weather as described in any one of claims 1 to 7.

Images