A thunder short-time early warning algorithm based on atmospheric electric field and meteorological radar multi-source data fusion

Abstract

This invention discloses a short-term lightning warning algorithm based on the fusion of multi-source data from atmospheric electric fields and meteorological radar, belonging to the field of lightning monitoring and early warning technology. Addressing the industry pain points of existing lightning warning technologies—such as reliance on a single data source, asynchronous data timing, unpredictable cloud evolution, and poor adaptability to fixed threshold scenarios—leading to high false alarm rates, short warning times, and both missed and over-warning warnings, this invention proposes a multi-source data fusion AI early warning solution. This invention integrates real-time atmospheric electric field timing monitoring data with meteorological radar three-dimensional echo data. A dynamic timing registration and correction algorithm unifies the spatiotemporal reference of the multi-source heterogeneous data. A dual-dimensional cross-discrimination mechanism based on ground electric field timing characteristics and airborne radar three-dimensional cloud characteristics filters out electrostatic false interference. An AI cloud evolution fitting model is used to predict the cloud formation and dissipation and charge accumulation trends over the next 10 to 30 minutes. Simultaneously, a three-level graded early warning output is achieved based on adaptive dynamic thresholds using terrain and seasonal factors. This invention effectively reduces the false alarm rate of lightning warnings, improves the accuracy of warning identification, and extends the effective warning duration. It can adapt to all terrain scenarios such as mountains, plains, permafrost, and ancient buildings, as well as seasonal meteorological environments. It can link various lightning protection devices to build an integrated lightning protection system that combines warning, protection, and monitoring. It is widely applicable to short-term accurate warnings in high-risk areas such as new energy power plants, substations, industrial parks, and clusters of ancient buildings.

Description

Technical Field

[0001] This invention belongs to the technical fields of lightning monitoring and early warning, multi-source meteorological data fusion, short-term severe convective lightning forecasting, and lightning risk prevention and control for power and new energy power plants. Specifically, it involves an artificial intelligence short-term lightning early warning algorithm based on the fusion of atmospheric electric field temporal characteristics and meteorological radar three-dimensional echo, AI fitting of thundercloud spatiotemporal evolution, multi-source data temporal registration, and terrain seasonal adaptive threshold. It is mainly applied to all-area lightning early warning scenarios such as high-voltage substations, mountain wind power bases, centralized photovoltaic power stations, chemical industrial parks, clusters of cultural relics and ancient buildings, and mountain communication base stations. It can be equipped with lightning protection grounding, passive equipotential protection, and intelligent grounding impedance control equipment to achieve integrated closed-loop prevention and control of lightning risks across the entire area. Background Technology

[0002] Short-term lightning warning is a core pre-emptive technology for power safety production, ancient building security, and mountain engineering construction protection. Currently, the mainstream lightning warning methods in the industry mainly rely on a single atmospheric electric field instrument or independent meteorological radar for discrimination. The single monitoring mode has inherent technical defects, and there are also common problems such as misalignment of multi-source data time sequence and fixed discrimination threshold. This results in existing warning systems having many false alarms, many false alarms, short warning time, and poor regional adaptability, which cannot meet the requirements of high-precision and high-reliability lightning warning.

[0003] 1. Single atmospheric electric field meters have large discrimination errors and frequent false alarms. Ground-based atmospheric electric field meters only collect near-surface shallow electrostatic signals, making them highly susceptible to interference from external environmental factors such as wind and dust, human and vehicle activity, vegetation friction static electricity, and the insulating medium of the ground surface. Even in clear, cloudy, and thunderstorm-free weather, they can generate a large number of false charge disturbance signals, with conventional equipment having a lightning false alarm rate exceeding 40%. Furthermore, this equipment only has one-dimensional surface monitoring capabilities and cannot acquire three-dimensional spatial intensity and vertical structure information of upper-level thunderclouds. It cannot physically distinguish between localized environmental electrostatic interference and the actual thunderstorm cloud charge accumulation process, and relying solely on a single electric field threshold for judgment easily leads to a large number of invalid warnings.

[0004] 2. Pure weather radar monitoring lacks corroborating evidence of ground charge, leading to frequent false alarms. Doppler weather radar can only identify cloud morphology and structure parameters such as cloud echo intensity, vertical liquid water content, and cloud top height. It cannot directly perceive the charging state and charge accumulation growth pattern within clouds, often resulting in false alarms where cloud echo intensity is high but no cloud charging or lightning occurs. Furthermore, radar data cannot quantify the rate of thundercloud charge growth, making it difficult to predict thunderstorm formation time in advance. This results in a severely insufficient lead time for warnings, failing to meet short-term disaster avoidance needs.

[0005] 3. Asynchronous sampling times from multiple devices lead to feature misalignment during direct fusion. Atmospheric electric field meters use a continuous sampling mode on the order of seconds, while weather radar uses a periodic scanning mode on the order of minutes. The sampling periods of these two types of monitoring equipment differ significantly. Furthermore, the variable transmission delays caused by wireless transmission and gateway forwarding result in a noticeable time offset between electric field data and radar data for the same monitoring area. Direct fusion of heterogeneous data without time-series correction causes a spatiotemporal mismatch between ground electric field characteristics and aerial cloud characteristics, significantly reducing the accuracy of early warning models.

[0006] 4. Fixed discrimination thresholds have poor universality, leading to both missed reports and over-warnings. Different regions exhibit significant differences in terrain elevation, soil resistivity, and surface conductivity. The atmospheric electrostatic conduction characteristics of high-resistivity rock layers in mountains, permafrost insulation environments, soft and moist strata in plains, and high-insulation roofs of ancient buildings are completely different. Furthermore, temperature, humidity, air ion concentration, and cloud electrification patterns show significant seasonal differences throughout the year. Traditional early warning algorithms use globally fixed discrimination thresholds, which cannot adapt to complex terrain and seasonal changes. In the dry and cold winter environment of mountainous areas, missed lightning warnings are prone to occur, while in the high-humidity summer environment of plains, meaningless over-warnings are likely to occur, resulting in a severe deficiency in comprehensive and refined early warning capabilities.

[0007] In summary, existing lightning warning technologies suffer from drawbacks such as single data source, low data fusion accuracy, inability to predict lightning cloud evolution, and fixed and rigid discrimination thresholds. They lack an integrated algorithm system that combines multi-source data collaborative correction, dual-dimensional cross-validation, AI evolutionary inference, and adaptive hierarchical discrimination, making it difficult to adapt to high-precision short-term lightning warning scenarios across all terrains and seasons. Summary of the Invention

[0008] The purpose of this invention is to solve the technical problems of existing short-term lightning warning technologies, such as high false alarm rate of single-source monitoring, asynchronous time sequence of multi-source data, inability to predict the evolution trend of thunderclouds, and poor adaptability of fixed threshold scenarios. It provides a short-term lightning warning algorithm that integrates atmospheric electric field and meteorological radar multi-source data. Through four core technologies, it achieves short-term lightning warnings with high accuracy and long duration, covering all terrains and all seasons.

[0009] 1. The core of this invention employs a dual-dimensional fusion and cross-discrimination mechanism combining atmospheric electric field temporal characteristics and radar three-dimensional echo intensity. It simultaneously extracts temporal features such as the mean, fluctuation variance, charge rise / fall slope, and short-term pulse abrupt changes of the ground atmospheric electric field, while also extracting three-dimensional spatial features such as the vertical profile of radar cloud reflectivity, vertical liquid water content, cloud top height, and cloud coverage, constructing a ground-air joint discrimination feature set. Through dual cross-verification of ground electrostatic disturbances and upper-air thundercloud electrification conditions, it accurately distinguishes between local environmental electrostatic interference and actual thunderstorm charge accumulation signals, fundamentally eliminating false alarms and false alarms caused by single-device monitoring.

[0010] 2. This invention establishes an AI fitting model for the spatiotemporal evolution of thunderclouds. Taking continuous multi-frame radar time-series data and time-aligned atmospheric electric field changes as input, the model automatically fits the movement trajectory, translation speed, cloud expansion and dissipation rate, and cloud charge accumulation growth rate of thunderclouds through a time-series convolutional network. This accurately predicts the spatial evolution trend and charge accumulation state of cloud clusters in the next 10 to 30 minutes, and forecasts the time of lightning occurrence and the scope of impact, effectively extending the lead time for lightning warnings.

[0011] 3. This invention configures a multi-source data dynamic temporal registration and correction algorithm. To address the heterogeneous data differences between atmospheric electric field instruments with second-level sampling and radar with minute-level scanning, it uses the radar scanning timestamp as the reference axis to perform piecewise linear interpolation resampling on the high-frequency electric field data. Combined with historical transmission delay data, it completes dynamic delay compensation, unifies the time reference and spatial coordinate system of all monitoring data, completely solves the problem of temporal misalignment of multi-source data, ensures the spatiotemporal consistency of fused data, and improves the model discrimination accuracy.

[0012] 4. This invention constructs a terrain-seasonal dual-factor adaptive threshold network, collects sample data such as regional terrain elevation, soil resistivity, surface medium type, seasonal temperature and humidity, and annual average lightning density to train the threshold model offline. When the system is running online, it can dynamically adjust the lightning initiation judgment threshold according to the terrain type of the monitored area and real-time seasonal meteorological parameters, and divides the warning levels into three levels: attention, warning, and danger. This effectively adapts to the lightning initiation patterns of different terrains and seasons, and eliminates the problems of missed reports and over-warning caused by fixed thresholds.

[0013] This invention also provides an AI-powered short-term lightning warning system that integrates multi-source data fusion from atmospheric electric fields and meteorological radar. The system comprises a multi-source monitoring data access unit, a time-series registration and correction module, a two-dimensional feature fusion and discrimination module, a thundercloud spatiotemporal evolution fitting AI calculation module, and a terrain-adaptive seasonal graded warning output module. 1. The multi-source monitoring data access unit connects to distributed atmospheric electric field meters, meteorological Doppler radar, and regional meteorological stations, collecting real-time electric field time-series data, radar three-dimensional echo raster data, and basic topographic and meteorological parameters, and completing standardized decoding and caching. 2. The time-series registration and correction module unifies the spatiotemporal reference of heterogeneous data through interpolation resampling and time delay compensation algorithms. 3. The two-dimensional feature fusion and discrimination module extracts multi-dimensional features from the ground and air, cross-filters false electrostatic interference signals, and performs initial screening for thunderstorm authenticity. 4. The thundercloud spatiotemporal evolution fitting AI calculation module extrapolates the thundercloud evolution trend and lightning occurrence probability based on the time-series fused data. 5. The terrain-adaptive seasonal graded warning output module dynamically matches and discriminates thresholds, outputs standardized graded warning information, and pushes it to terminal devices.

[0014] The lightning short-term warning algorithm of this invention includes the following complete execution steps: 1. Real-time access to global atmospheric electric field time-series data, meteorological radar three-dimensional echo data, regional topography, and seasonal meteorological basic parameters, completing data standardization processing and local caching. 2. Running a multi-source data time-series registration and correction algorithm to interpolate, align, and compensate for delays in heterogeneous monitoring data, unifying the spatiotemporal reference of all data. 3. Synchronously extracting atmospheric electric field time-series features and radar three-dimensional cloud features, conducting ground-air dual-dimensional cross-fusion discrimination, and filtering false lightning signals generated by local electrostatic interference. 4. Calling the lightning cloud spatiotemporal evolution fitting AI model to calculate lightning cloud movement parameters and charge accumulation rate, predicting the spatiotemporal range of lightning occurrence in the next 10 to 30 minutes. 5. Combining regional topography and seasonal parameters to match adaptive dynamic discrimination thresholds, classifying three levels of lightning risk warning. 6. Generating gridded lightning warning messages and visualized risk maps, synchronously pushing them to various control terminals to complete the warning closed loop.

[0015] The beneficial effects of this invention are as follows: 1. This invention adopts a ground-air dual-dimensional cross-verification mode, effectively filtering various environmental electrostatic interferences, reducing the lightning false alarm rate by more than 60%, and significantly reducing the workload of invalid early warnings. 2. Through the AI ​​lightning cloud evolution fitting model, the effective early warning duration is extended to 30 minutes, reserving sufficient handling windows for on-site risk avoidance and equipment maintenance. 3. The time-series registration correction algorithm eliminates the problem of multi-source data misalignment, improving the early warning recognition accuracy by 25%. 4. The adaptive dynamic threshold can adapt to all terrains, including mountains, plains, permafrost, and ancient buildings, as well as seasonal meteorological environments, completely solving the problems of missed and over-warning. 5. This invention can seamlessly connect with various lightning protection devices to achieve integrated linkage management of early warning, protection, and monitoring, constructing a complete lightning protection system. 6. The overall solution has strong versatility and convenient deployment, can be adapted to various high-risk lightning areas, and has high industrial application value. Attached Figure Description

[0016] Figure 1 This is a block diagram of the overall modular architecture of the early warning system of the present invention, showing the structure of the five major functional modules and the flow of data transmission, processing and output across the entire domain.

[0017] Figure 2 This is a schematic diagram of the multi-source data time-series registration and correction algorithm of the present invention, showing the complete processing flow of interpolation resampling and dynamic delay compensation based on radar timestamps.

[0018] Figure 3 This is a flowchart of the ground-to-air dual-dimensional feature fusion discrimination logic of the present invention, which shows the complete discrimination logic of multi-dimensional feature extraction, cross-validation, and false signal filtering.

[0019] Figure 4 This is a schematic diagram of the spatiotemporal evolution AI model of thunderclouds in this invention, showing the movement trajectory of thunderclouds, expansion range, and potential impact area of ​​lightning in the next 10 to 30 minutes.

[0020] Figure 5 This is a logic diagram for the terrain-season adaptive threshold classification early warning determination of the present invention, showing the parameter input, dynamic threshold matching, and three-level risk classification output process.

[0021] Figure 6 This is a flowchart of the overall execution steps of the algorithm of the present invention, which shows the serial execution logic and data flow relationship of the entire algorithm. Detailed Implementation

[0022] The technical solution of the present invention will be described in complete and detailed exemplary form below with reference to the accompanying drawings. This embodiment takes a mountain-based centralized photovoltaic new energy base as a typical application scenario to verify the overall performance of the algorithm and system of the present invention, and is not intended to limit the scope of protection of the present invention.

[0023] 1. Equipment Deployment and Data Model Building. Distributed atmospheric electric field instruments are deployed throughout the mountainous photovoltaic early warning area, with a sampling cycle of 1 second to achieve continuous acquisition of near-surface electric field signals. These instruments are connected to a regional meteorological Doppler radar, which completes a full-area 3D echo scan with a 6-minute cycle, stably acquiring 3D data such as cloud echo intensity, vertical liquid water content, and cloud top height. Digital elevation data, soil resistivity survey data, temperature and humidity statistics for the past five seasons, and regional annual average lightning density data are collected for the target area. A localized topographic meteorological sample library is constructed, and an offline adaptive threshold network based on the topography-seasonal dual-factor model is trained based on this sample library. The trained model parameters are then stored locally in the system. Simultaneously, communication interfaces are established between the system and the photovoltaic power station's intelligent operation and maintenance platform, passive grounding impedance compensation devices, and field terminal equipment to ensure stable transmission of early warning signals and lightning protection equipment.

[0024] 2. Multi-source data access and temporal registration correction. The system routinely collects high-frequency time-series data of atmospheric electric fields, 3D raster data from meteorological radar, static terrain parameters, and seasonal dynamic meteorological parameters in real time, completing the decoding, cleaning, and caching of the raw data. Using the radar's 6-minute scan timestamp as a unified time reference, the system performs piecewise linear interpolation resampling on the second-level high-frequency electric field time-series data, ensuring a perfect match between the electric field data time nodes and the radar scan time. It retrieves historical transmission delay data from the equipment over the past 24 hours, calculates dynamic delay compensation coefficients, and uniformly corrects the timestamps of the two types of heterogeneous data, completely eliminating the spatiotemporal misalignment caused by differences in sampling periods and network transmission delays. The system outputs a standardized fusion dataset with complete spatiotemporal alignment, providing accurate data support for subsequent feature discrimination.

[0025] 3. Ground-to-Air Dual-Dimensional Feature Fusion and False Signal Filtering. The system synchronously extracts multi-dimensional features from the corrected data. From the ground dimension, it extracts temporal features such as the 10-minute mean electric field, 5-minute electric field fluctuation variance, charge rise and fall slope, and short-term pulse abrupt change amplitude. From the air dimension, it extracts three-dimensional structural features such as radar cloud vertical reflectivity profile, vertical liquid water content, cloud top height, cloud cluster horizontal coverage radius, and cloud layer vertical thickness. A joint ground-to-air discrimination mechanism is constructed. When a single electric field data is abnormal and the radar lacks strong convective cloud characteristics, it is judged as electrostatic interference signals caused by wind, sand, vegetation, or human activities, and directly filtered out. When a single radar has a strong echo and the electric field does not show a continuous charge accumulation trend, it is judged as a no-charge false alarm sample and no warning is issued. Only when the ground electric field continuously and stably increases and the radar meets the three-dimensional characteristics of strong convective clouds is it judged as a real thunderstorm cloud, proceeding to the next evolutionary simulation process. Field measurements show that this discrimination method can reduce the number of false warnings in a region by 62% per day, significantly improving the effectiveness of warnings.

[0026] 4. AI-powered Thundercloud Spatiotemporal Evolution Fitting and Lightning Trend Prediction. The system selects three consecutive frames of time-aligned fused feature data and inputs them into a pre-trained thundercloud spatiotemporal evolution fitting model. The model calculates key parameters in real time, such as thundercloud movement direction, translational speed, cloud expansion rate, and cloud charge accumulation rate. Based on these parameters, it predicts the spatial evolution path and charge accumulation trend of the thundercloud over the next 10 to 30 minutes, accurately marking potential lightning occurrence areas and estimating lightning strike times. In this implementation scenario, the model can stably predict the lightning risk in the field area over the next 25 to 30 minutes. Compared to traditional early warning methods, the lead time for early warning is increased by more than two times, effectively solving the problem of insufficient timeliness in short-term early warnings.

[0027] 5. Terrain-Adaptive Seasonal Threshold Classification. The system identifies the mountainous terrain attributes and current seasonal meteorological parameters of the site in real time, automatically calls the appropriate dynamic discrimination threshold, and quantifies the lightning risk level by combining real-time ground-air fusion feature data, distinguishing between three levels of warning: attention, alert, and danger. In the high temperature and humidity environment of summer, the system automatically optimizes the electrostatic interference filtering threshold to avoid over-warning; in the dry, cold, and high-resistivity geological environment of winter, the system lowers the electrification discrimination threshold to effectively prevent missed lightning warnings. For different scenarios such as plains, permafrost, and ancient buildings, only the localized terrain and meteorological parameter library needs to be replaced to automatically adapt the corresponding discrimination threshold, without the need for manual modification of the core algorithm parameters, demonstrating strong adaptability across all scenarios.

[0028] 6. Tiered Early Warning Output and Closed-Loop Lightning Protection Linkage. Based on the determined risk level, the system generates a comprehensive gridded lightning warning message and a visualized risk heat map, simultaneously pushing them to the site's central control platform, maintenance handheld terminals, and security management system. Upon triggering a high-level hazard warning, the system automatically activates the passive grounding impedance compensation device and the intelligent lightning protection maintenance system to dynamically optimize the site's grounding impedance to a safe range, simultaneously sending equipment shutdown and personnel evacuation alerts to complete on-site safety and hazard mitigation control. Once the lightning clouds dissipate and the risk level returns to a safe threshold, the system automatically deactivates the enhanced lightning protection mode, restoring normal site operation and achieving a closed-loop management system encompassing lightning risk warning, active protection, and online monitoring.

[0029] 7. Explanation of Full-Scene Adaptation and Expansion. The core architecture of this invention's algorithm is universal and requires no modification to the core logic. By simply replacing the corresponding regional terrain resistivity, seasonal weather, and lightning density parameter libraries, it can quickly adapt to various high-risk lightning scenarios such as wind power bases, substations, industrial parks, permafrost mining areas, and clusters of cultural relics and ancient buildings. The accuracy of full-domain measured early warning recognition is steadily improved by more than 25%, completely solving the problem of poor full-scene adaptability of traditional early warning technologies, and possessing extremely high engineering implementation and promotion value.

Claims

1. A short-term lightning warning system that fuses multi-source data from atmospheric electric fields and meteorological radar, characterized in that, The system includes a multi-source monitoring data access unit, a time-series registration and correction module, a dual-dimensional feature fusion and discrimination module, a thundercloud spatiotemporal evolution fitting AI calculation module, and a terrain-seasonal adaptive graded early warning output module. The time-series registration and correction module is equipped with a multi-source data time-series registration and correction algorithm to eliminate spatiotemporal data deviations caused by differences in sampling periods and transmission delays between heterogeneous monitoring devices such as meteorological radar and distributed atmospheric electric field meters. The dual-dimensional feature fusion and discrimination module integrates continuous temporal features of the atmospheric electric field with three-dimensional intensity features of radar echoes to achieve ground-to-air cross-discrimination, filtering false alarms and false alarms from single monitoring sources. The thundercloud spatiotemporal evolution fitting AI calculation module fits the movement speed, cloud expansion rate, and cloud charge accumulation rate of thunderclouds based on continuous temporal fusion data to predict the short-term spatiotemporal evolution trend of lightning over the next 10–30 minutes. The terrain-seasonal adaptive graded early warning output module incorporates a terrain-seasonal dual-factor threshold network, which can automatically and dynamically adjust the lightning ignition discrimination threshold according to the terrain type and real-time seasonal meteorological conditions of the monitoring area, and output multi-level standardized lightning early warning signals.

2. The lightning short-term early warning system based on the fusion of atmospheric electric field and meteorological radar multi-source data according to claim 1, characterized in that, The system outputs a graded lightning warning signal that can be linked with the intelligent operation and maintenance monitoring system for lightning protection and grounding, the passive grounding impedance compensation device, and the passive equipotential protection equipment. Based on the warning risk level, the system can perform dynamic optimization of the grounding system impedance and activate the active lightning protection equipment in advance, thereby realizing the linkage control between lightning risk warning and on-site lightning protection.

3. A short-term lightning warning algorithm based on the fusion of atmospheric electric field and meteorological radar multi-source data, characterized in that, The execution steps include:

1. Real-time access to distributed atmospheric electric field time-series monitoring data, meteorological radar three-dimensional echo grid data, regional topographic elevation, soil resistivity, seasonal temperature and humidity basic static parameters, and completion of data standardization decoding and caching; 2. Execute a multi-source data dynamic time-series registration and correction process, perform segmented interpolation resampling and time delay offset compensation for heterogeneous monitoring data, and unify the time reference and spatial coordinate system of all monitoring data; 3. Simultaneously extract the continuous temporal fluctuation characteristics of the atmospheric electric field and the three-dimensional echo intensity and vertical structure characteristics of radar clouds, and carry out ground-air dual-dimensional joint cross-discrimination to filter out false lightning signals generated by local environmental electrostatic interference; 4. By pre-training an AI model to fit the spatiotemporal evolution of thunderclouds, calculate the movement trajectory, expansion and dissipation speed, and cloud charge accumulation rate of thunderclouds, and predict the potential spatiotemporal range of lightning occurrence in the next 10–30 minutes; 5. Retrieve terrain and seasonal parameters of the target area to match adaptive dynamic discrimination thresholds, and classify multiple levels of lightning risk warning based on the strength of the risk.

6. Generate a full-domain gridded short-term lightning warning message and a visual map of the impact range, and push it synchronously to various on-site control terminals to complete the complete short-term lightning warning business process.

4. The lightning short-term early warning algorithm based on the fusion of atmospheric electric field and meteorological radar multi-source data according to claim 3, characterized in that, The dual-dimensional fusion discrimination mechanism described in step 3 simultaneously uses two physical dimensions—the continuous electric field change characteristics near the ground surface and the three-dimensional structural characteristics of high-altitude thunderclouds—to cross-verify the authenticity of thunderstorms, thereby reducing the false alarm rate of early warnings caused by a single monitoring device from the source.

5. The lightning short-term early warning algorithm based on the fusion of atmospheric electric field and meteorological radar multi-source data according to claim 3, characterized in that, The adaptive hierarchical early warning thresholds described in step 5 are independently adapted and adjusted for various geological surface environments such as mountains, permafrost, plains, and ancient cultural relics and buildings, as well as for the meteorological conditions of spring, summer, autumn and winter. This solves the defects of traditional global fixed thresholds, such as missed lightning warnings and meaningless over-warnings.

Images