A distributed thunder and lightning monitoring and early warning system based on internet of things

Abstract

The present application relates to lightning monitoring and early warning technical field, concretely is a kind of distributed lightning monitoring and early warning system based on internet of things, solve the problem that the edge node of existing system can only independently carry out preliminary early warning, cannot utilize the data of adjacent node to carry out cross verification, early warning accuracy is limited, early warning advance time is shorter, cannot provide enough emergency preparation time.A kind of distributed lightning monitoring and early warning system based on internet of things, including distributed edge perception layer, regional collaborative decision-making layer and cloud intelligent management platform, and distributed edge perception layer, regional collaborative decision-making layer and cloud intelligent management platform are interconnected by enhanced internet of things communication network.The present application shortens the delay time of early warning response, can carry out lightning production monitoring in earlier time before lightning generation and issue early warning, reserves sufficient emergency preparation window for large-scale activities, high-risk industry.

Description

Technical Field

[0001] This invention relates to the field of lightning monitoring and early warning technology, specifically to a distributed lightning monitoring and early warning system based on the Internet of Things. Background Technology

[0002] Lightning is a serious natural disaster that causes numerous casualties and property losses every year. Therefore, it is necessary to monitor and provide timely warnings about lightning. Lightning monitoring and early warning refers to a technical system that uses multi-source observation data, such as lightning location, radar, satellite, and ground electric field, to identify, track, and issue graded warnings about lightning activity through multi-data fusion and algorithm analysis. Its monitoring methods include magnetic orientation method, time difference method, and narrowband interferometer location method. It determines discharge parameters by detecting the electromagnetic pulse radiation field of lightning. Typical application scenarios cover the protection of key areas such as petrochemical plants, power grids, and mountain scenic areas. Traditional lightning monitoring systems mainly adopt a centralized architecture, which has high data transmission latency, high risk of single point of failure, and relies on dedicated communication networks, resulting in high deployment costs and limited coverage.

[0003] Existing systems can only provide preliminary warnings independently at edge nodes, and cannot cross-validate data from adjacent nodes. This limits the accuracy of warnings and results in short lead times, failing to provide sufficient emergency preparedness time. Therefore, these systems do not meet current needs. To address this, we propose a distributed lightning monitoring and warning system based on the Internet of Things (IoT). Summary of the Invention

[0004] The purpose of this invention is to provide a distributed lightning monitoring and early warning system based on the Internet of Things to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a distributed lightning monitoring and early warning system based on the Internet of Things, comprising a distributed edge sensing layer, a regional collaborative decision-making layer, and a cloud-based intelligent management platform, wherein the distributed edge sensing layer, the regional collaborative decision-making layer, and the cloud-based intelligent management platform are interconnected through an enhanced Internet of Things communication network; The distributed edge sensing layer consists of multiple heterogeneous intelligent sensing nodes. Each node integrates a MEMS resonant differential multi-parameter sensor array, a RISC-V lightweight AI-accelerated edge computing unit, a radio frequency wake-up dual-mode communication module, and an environmental energy hybrid power supply module. The regional collaborative decision-making layer consists of multiple regional collaborative gateways. Each gateway is responsible for managing edge sensing nodes within its scope, receiving preprocessed data and cluster inference results uploaded by the nodes, performing multimodal data fusion based on spatiotemporal attention mechanism and improved conflict evidence fusion decision-making, calculating the probability, intensity, location and movement trajectory of lightning, and generating regional early warning information. The cloud-based intelligent management platform is used to receive aggregated data and early warning information uploaded by regional collaborative gateways, perform global-level data analysis, model training and system management, and issue early warning information to users and relevant departments.

[0006] Preferably, the MEMS resonant differential multi-parameter sensor array includes: The MEMS resonant differential three-dimensional atmospheric electric field sensor adopts a differential resonant structure; A graphene-based terahertz electromagnetic pulse sensor with an operating frequency range of 0.1THz to 10THz and a sensitivity of -120dBm; An optical lightning sensor employs a silicon photodiode array; The acoustic sensor employs a MEMS capacitive microphone array; Environmental auxiliary sensors include temperature and humidity sensors, air pressure sensors, and wind speed sensors.

[0007] Preferably, the RISC-V lightweight AI-accelerated edge computing unit can dynamically adjust the sampling frequency and data transmission priority according to the characteristics of different sensors.

[0008] Preferably, the enhanced IoT communication network adopts a three-tier hybrid communication architecture: The edge sensing nodes and the regional collaborative gateway adopt dual-mode communication of LoRaWAN Class B and NB-IoT. NB-IoT is used in areas with good network coverage, and LoRaWAN is used in remote areas. TSN time-sensitive network nodes are deployed in key protection areas (airports, chemical plants, mines), and TSN communication is used between them and the regional collaborative gateway; Point-to-point self-organizing network communication is supported between adjacent edge sensing nodes; The regional collaborative gateway and the cloud-based intelligent management platform use 5G SA and fiber optic dual-mode communication.

[0009] Preferably, the cloud-based intelligent management platform also integrates a lightweight edge blockchain evidence storage module, which uses a simplified PBFT consensus algorithm to complete the evidence storage of key monitoring data and early warning information at the regional gateway level, with an evidence storage latency of less than 100ms.

[0010] Preferably, the multimodal data fusion algorithm based on spatiotemporal attention mechanism includes a temporal attention layer, a spatial attention layer, and a modal attention layer connected in sequence; The temporal attention layer is used to learn the weights of data at different time steps and capture the temporal evolution of lightning precursor signals. The spatial attention layer employs a graph attention network structure to learn the weights of node data from different geographical locations, thereby capturing the spatial distribution characteristics of thunderstorm clouds. The modal attention layer is used to learn the weights of different sensor modal data and automatically adjust the contribution of each sensor data in the fusion process.

[0011] Preferably, the improved conflict evidence fusion decision adopts a weighted DS evidence theory based on evidence credibility and conflict degree, and its evidence combination rule is as follows: in, For the proposition The comprehensive basic probability allocation, For the first The credibility weight of each piece of evidence The conflict coefficient, For the first The basic probability assignment function for each piece of evidence. The number of pieces of evidence.

[0012] Preferably, the evidence credibility weight The calculation formula is: in, For the first The evidence and the first The distance between each piece of evidence is calculated using cosine similarity. To identify the number of propositions in the framework.

[0013] Preferably, the heterogeneous intelligent sensing node also integrates a device self-diagnosis module, which can monitor the operating status of the sensors, communication modules and power modules in real time.

[0014] Preferably, the distributed lightning monitoring and early warning system adopts a four-level early warning mechanism: Level 0 Warning (White): Widespread lightning activity may occur within the next 1-3 hours, issued by the cloud-based lightning precursor early detection and analysis module; Level 1 Warning (Blue): Thunderstorms are possible within the next 30 minutes, with a probability of 30%-60%. Level II Warning (Yellow): There is a high probability of thunderstorms within the next 15 minutes, with a 60%-90% chance of occurrence. Level 3 Warning (Red): There is an extremely high probability of lightning activity within the next 5 minutes, or lightning activity has already occurred and may continue, with a probability of lightning occurrence greater than 90%.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention shortens the delay time of early warning response by using a multi-node collaborative edge inference cluster mechanism and a radio frequency wake-up early warning broadcast module, and issues lightning warnings in a shorter time. Relying on a graphene-based terahertz electromagnetic pulse sensor and a lightning precursor advanced perception and analysis module, it can issue early warnings 1-3 hours before lightning occurs, reserving sufficient emergency preparation windows for large-scale events and high-risk industries. 2. This invention reduces data storage latency by using a lightweight edge blockchain storage module, thereby alleviating storage pressure and bandwidth consumption in the cloud. At the same time, it improves early warning accuracy and reduces false alarm rate by combining MEMS resonant differential sensors with a multi-node collaborative inference mechanism, and ensures timely delivery of early warning information even under network congestion. Attached Figure Description

[0016] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a flowchart illustrating the multi-node collaborative edge inference cluster mechanism of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] A distributed lightning monitoring and early warning system based on the Internet of Things (IoT) includes a distributed edge sensing layer, a regional collaborative decision-making layer, and a cloud-based intelligent management platform. The distributed edge sensing layer, the regional collaborative decision-making layer, and the cloud-based intelligent management platform are interconnected through an enhanced IoT communication network.

[0019] The enhanced IoT communication network adopts a three-tier hybrid communication architecture: The edge sensing nodes and the regional collaborative gateway adopt dual-mode communication of LoRaWAN Class B and NB-IoT. NB-IoT is used in areas with good network coverage, and LoRaWAN is used in remote areas. TSN time-sensitive network nodes are deployed in key protection areas (airports, chemical plants, mines), and TSN communication is used with the regional collaborative gateway. The latency of critical data transmission is less than 1μs and the jitter is less than 100ns. Point-to-point self-organizing network communication is supported between adjacent edge sensing nodes. When communication between a node and the gateway is interrupted, data can be forwarded through neighboring nodes. The regional collaborative gateway and the cloud-based intelligent management platform use 5G SA and fiber optic dual-mode communication to ensure the reliability and real-time performance of data transmission.

[0020] The distributed edge sensing layer consists of multiple heterogeneous intelligent sensing nodes. Each node integrates a MEMS resonant differential multi-parameter sensor array, a RISC-V lightweight AI-accelerated edge computing unit, a radio frequency wake-up dual-mode communication module, and an environmental energy hybrid power supply module.

[0021] The heterogeneous intelligent sensing node also integrates a device self-diagnosis module, which can monitor the operating status of sensors, communication modules and power modules in real time. When a device fault is detected, it can automatically locate the fault and perform preliminary repairs, and report the fault information to the cloud intelligent management platform.

[0022] MEMS resonant differential multiparameter sensor arrays include: The MEMS resonant differential three-dimensional atmospheric electric field sensor adopts a differential resonant structure, with a measurement range of -50kV / m to +50kV / m, a resolution of 1V / m, a response time of less than 1ms, and no moving parts. The graphene-based terahertz electromagnetic pulse sensor operates in the frequency range of 0.1THz to 10THz and has a sensitivity of -120dBm. It can detect terahertz precursor radiation generated by thunderstorm clouds 10-15 minutes in advance. The optical lightning sensor employs a silicon photodiode array, with a response wavelength range of 300nm to 1100nm and a response time of less than 0.1μs. The acoustic sensor uses a MEMS capacitive microphone array with a frequency response range of 20Hz to 20kHz and a sensitivity of -35dB. Environmental auxiliary sensors include temperature and humidity sensors, air pressure sensors, and wind speed sensors.

[0023] The RISC-V lightweight AI-accelerated edge computing unit can dynamically adjust the sampling frequency and data transmission priority according to the characteristics of different sensors. When the atmospheric electric field strength exceeds the first threshold (±10kV / m), the sampling frequency of the graphene terahertz electromagnetic pulse sensor and the optical sensor is increased from 100Hz to 1000Hz; when the atmospheric electric field strength exceeds the second threshold (±30kV / m), the sampling frequency of all sensors is further increased to 2000Hz.

[0024] The edge computing unit is deployed with spatiotemporal feature extraction and preliminary early warning models, and supports a multi-node collaborative edge inference cluster mechanism. When a single node detects that the lightning risk exceeds the first threshold (30%), it immediately wakes up 3-5 neighboring nodes to form a temporary inference cluster, and completes data fusion and risk assessment in a distributed and parallel manner. The assessment results are synchronized within the cluster.

[0025] The workflow of the multi-node collaborative edge inference cluster mechanism is as follows: Once the triggering node detects that the lightning risk exceeds the first threshold (30%), it sends a cluster formation request to 3-5 neighboring nodes. Upon receiving the request, the neighboring node immediately joins the cluster and sends its local feature data to the triggering node; The triggering node acts as the cluster master node, and uses a distributed parallel inference algorithm to fuse the feature data of all nodes to calculate the comprehensive lightning risk value. The master node synchronizes the inference results to all nodes in the cluster and uploads them to the regional collaboration gateway; Once the lightning risk has subsided, the cluster will automatically disband, and each node will resume independent operation.

[0026] During the cluster formation process, the triggering node sends a cluster formation request to the neighboring nodes. After receiving the request, the neighboring nodes join the cluster and send local feature data. When the detected lightning risk exceeds the second threshold (60%), a local audible and visual alarm is immediately triggered and an emergency data transmission mode is activated, prioritizing the uploading of high-priority feature data to the regional collaborative gateway.

[0027] The regional collaborative decision-making layer consists of multiple regional collaborative gateways. Each gateway is responsible for managing edge sensing nodes within a range (radius of 20km), receiving preprocessed data and cluster inference results uploaded by the nodes, performing multimodal data fusion based on spatiotemporal attention mechanism and improved conflict evidence fusion decision-making, calculating the probability, intensity, location and movement trajectory of lightning, and generating regional early warning information.

[0028] The management range radius is determined based on the LoRaWAN communication distance and lightning positioning accuracy requirements. Nodes within a 20km radius can meet the requirement of lightning positioning accuracy of less than 50m.

[0029] The algorithm for multimodal data fusion based on spatiotemporal attention mechanism includes a temporal attention layer, a spatial attention layer and a modal attention layer connected in sequence; The temporal attention layer is used to learn the weights of data at different time steps and capture the temporal evolution of lightning precursor signals. The temporal attention layer employs an LSTM + attention mechanism structure. LSTM is used to learn the long-term dependencies of the time series, while the attention mechanism is used to calculate the weights at different time steps. The formula for calculating the temporal attention weights is: in, For the first Attention weights at each time step For LSTM in the first The hidden state at each time step and The weight matrix is ​​a learnable matrix. For bias vectors, This represents the number of time steps.

[0030] The spatial attention layer employs a graph attention network structure to learn the weights of node data from different geographical locations, thereby capturing the spatial distribution characteristics of thunderstorm clouds. The spatial attention layer adopts a graph attention network (GAT) structure, treating each edge-aware node as a node in a graph, and the edge weights between nodes are determined by the geographical distance between the nodes. The formula for calculating spatial attention weights is: in, For nodes For nodes Attention weights and They are nodes and nodes eigenvectors, and The weight matrix is ​​a learnable matrix. For nodes The set of neighboring nodes, This indicates a splicing operation.

[0031] The modal attention layer is used to learn the weights of different sensor modal data and automatically adjust the contribution of each sensor data in the fusion process. The modal attention layer employs a fully connected layer structure to calculate the weights for different sensor modes. The formula for calculating the modal attention weights is: in, For the first Attention weights for each sensor modality, For the first The fused feature vector of each modality and The weight matrix is ​​a learnable matrix. For bias vectors, This represents the number of sensor modes.

[0032] The improved conflict evidence fusion decision-making adopts a weighted DS evidence theory based on evidence credibility and conflict degree, and its evidence combination rule is as follows: in, For the proposition The comprehensive basic probability allocation, For the first The credibility weight of each piece of evidence The conflict coefficient, For the first The basic probability assignment function for each piece of evidence. The number of pieces of evidence; Weight of evidence credibility The calculation formula is: in, For the first The evidence and the first The distance between each piece of evidence is calculated using cosine similarity. To identify the number of propositions in the framework.

[0033] The cloud-based intelligent management platform is used to receive aggregated data and early warning information uploaded by regional collaborative gateways, perform global-level data analysis, model training and system management, and issue early warning information to users and relevant departments.

[0034] The cloud-based intelligent management platform integrates a lightning precursor early sensing and analysis module, which can predict large-scale lightning activity 1-3 hours in advance based on very low frequency ionospheric disturbance data and issue a level zero warning; the system adopts a four-level warning mechanism.

[0035] The cloud-based intelligent management platform also integrates a lightweight edge blockchain notarization module, which uses a simplified PBFT consensus algorithm to notarize key monitoring data and early warning information at the regional gateway level, with a notarization latency of less than 100ms.

[0036] The system also includes a regional linkage sound and light early warning relay network, which consists of multiple sound and light early warning relay nodes deployed in key protection areas; Each relay node in the regional coordinated audible and visual early warning relay network integrates a radio frequency wake-up module, a high-decibel audible and visual alarm, and an LED display screen. It can receive the radio frequency wake-up signal sent by the regional collaborative gateway, immediately trigger the audible and visual alarm, and display the early warning information, with a coverage radius of 500m.

[0037] The regional collaborative gateway integrates a radio frequency wake-up ultra-low latency early warning broadcast module, which can send radio frequency wake-up signals to all edge sensing nodes and audible and visual early warning relay nodes within its management range, triggering a synchronized early warning across the entire network.

[0038] The regional collaborative gateway also supports a cross-regional collaborative early warning relay mechanism: when a thunderstorm cloud cluster is detected to be about to move to an adjacent area, it can predict the time and intensity of the thunderstorm cloud cluster arriving in the adjacent area and actively send early warning information to the collaborative gateway of the adjacent area. After receiving the early warning information, the collaborative gateway of the adjacent area immediately activates the enhanced monitoring mode of its own area, increases the sampling frequency of all edge sensing nodes to 1000Hz, and sends early warning signals to all nodes in its own area through the radio frequency wake-up module.

[0039] The distributed lightning monitoring and early warning system adopts a four-level early warning mechanism: Level 0 Warning (White): Widespread lightning activity may occur within the next 1-3 hours, issued by the cloud-based lightning precursor early detection and analysis module; Level 1 Warning (Blue): Thunderstorms are possible within the next 30 minutes, with a probability of 30%-60%. Level II Warning (Yellow): There is a high probability of thunderstorms within the next 15 minutes, with a 60%-90% chance of occurrence. Level 3 Warning (Red): There is an extremely high probability of lightning activity within the next 5 minutes, or lightning activity has already occurred and may continue, with a probability of lightning occurrence greater than 90%.

[0040] The workflow of the lightning monitoring and early warning system is as follows: S1, Advanced Warning Stage: The cloud-based lightning precursor early sensing and analysis module continuously receives data from very low frequency signal monitoring stations and analyzes changes in ionospheric electron density. When abnormal changes are detected, a Level 0 warning (white) is issued, and relevant departments are notified 1-3 hours in advance to prepare for emergency response.

[0041] S2, Normal Monitoring Mode: The multi-parameter sensor array of the edge sensing node collects data at the default sampling frequency, and the edge computing unit preprocesses the data and performs preliminary risk assessment. When the probability of lightning occurrence is less than 30%, the node uploads average data to the regional collaborative gateway every 5 minutes.

[0042] S3, Enhanced Monitoring Mode: When the edge computing unit detects that the atmospheric electric field strength exceeds the first threshold (±10kV / m) or the probability of lightning occurrence exceeds 30%, the node automatically enters the enhanced monitoring mode. The sampling frequency of the graphene terahertz electromagnetic pulse sensor and optical sensor has been increased to 1000Hz, and the node uploads data to the regional collaborative gateway once every minute; The node initiates a multi-node collaborative edge inference cluster mechanism to conduct risk assessments together with neighboring nodes.

[0043] S4, Emergency Warning Mode: When the edge computing unit or cluster inference results show that the probability of lightning occurrence exceeds 60%, the node immediately triggers a local audible and visual alarm and starts the emergency data transmission mode. At this point, the sampling frequency of all sensors is increased to 2000Hz, and the nodes upload high-priority feature data to the regional collaborative gateway with the highest priority.

[0044] S5, Regional-level Early Warning Decision-Making: After receiving data uploaded by multiple edge sensing nodes and cluster inference results, the regional collaborative gateway first uses a multimodal data fusion algorithm based on spatiotemporal attention mechanism to perform feature fusion, and then uses an improved conflict evidence fusion decision model to calculate the comprehensive lightning risk value. When the overall lightning risk value exceeds 60%, a level 2 warning is generated; when it exceeds 90%, a level 3 warning is generated.

[0045] S6, Network-wide Synchronous Early Warning: After generating an early warning message, the regional collaborative gateway immediately sends an RF wake-up signal to all nodes and relay devices within its management range through the RF wake-up ultra-low latency early warning broadcast module. Upon receiving the signal, the nodes and relay devices immediately trigger audible and visual alarms and display warning information; The regional linkage sound and light early warning relay network triggered a wide range of sound and light alarms in key protection areas.

[0046] S7, Cross-regional early warning relay: The regional collaborative gateway determines whether a thunderstorm cloud cluster is about to move to an adjacent area based on the lightning location results and trajectory prediction results. If so, predict the time and intensity of the thunderstorm cloud cluster arriving in the adjacent area, and proactively send early warning information to the collaborative gateway of the adjacent area; Upon receiving the early warning information, the collaborative gateway in the adjacent area immediately activates the enhanced monitoring mode for its own area and sends the early warning signal to all nodes in the area via the radio frequency wake-up module.

[0047] S8, Cloud-based early warning release: The regional collaborative gateway uploads regional-level early warning information to the cloud-based intelligent management platform; The cloud-based intelligent management platform performs global analysis and processing of early warning information, and then releases the early warning information to users and relevant departments through various channels such as mobile APP, SMS, WeChat official account, broadcast, and electronic display screen; The warning information is precisely pushed based on the user's geographical location, and is only sent to users in the affected area.

[0048] S9. Model Update and Optimization: The cloud-based intelligent management platform regularly uses newly collected lightning data to train and update the lightning early warning model, and distributes the updated model to edge sensing nodes and regional collaborative gateways via OTA to continuously improve the system's early warning performance.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A distributed lightning monitoring and early warning system based on the Internet of Things, characterized in that: It includes a distributed edge sensing layer, a regional collaborative decision-making layer, and a cloud-based intelligent management platform. The distributed edge sensing layer, the regional collaborative decision-making layer, and the cloud-based intelligent management platform are interconnected through an enhanced Internet of Things (IoT) communication network. The distributed edge sensing layer consists of multiple heterogeneous intelligent sensing nodes. Each node integrates a MEMS resonant differential multi-parameter sensor array, a RISC-V lightweight AI-accelerated edge computing unit, a radio frequency wake-up dual-mode communication module, and an environmental energy hybrid power supply module. The regional collaborative decision-making layer consists of multiple regional collaborative gateways. Each gateway is responsible for managing edge sensing nodes within its scope, receiving preprocessed data and cluster inference results uploaded by the nodes, performing multimodal data fusion based on spatiotemporal attention mechanism and improved conflict evidence fusion decision-making, calculating the probability, intensity, location and movement trajectory of lightning, and generating regional early warning information. The cloud-based intelligent management platform is used to receive aggregated data and early warning information uploaded by regional collaborative gateways, perform global-level data analysis, model training and system management, and issue early warning information to users and relevant departments.

2. The distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The MEMS resonant differential multi-parameter sensor array includes: The MEMS resonant differential three-dimensional atmospheric electric field sensor adopts a differential resonant structure; A graphene-based terahertz electromagnetic pulse sensor with an operating frequency range of 0.1THz to 10THz and a sensitivity of -120dBm; An optical lightning sensor employs a silicon photodiode array; The acoustic sensor employs a MEMS capacitive microphone array; Environmental auxiliary sensors include temperature and humidity sensors, air pressure sensors, and wind speed sensors.

3. The distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The RISC-V lightweight AI-accelerated edge computing unit can dynamically adjust the sampling frequency and data transmission priority according to the characteristics of different sensors.

4. The distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The enhanced IoT communication network adopts a three-level hybrid communication architecture: The edge sensing nodes and the regional collaborative gateway adopt dual-mode communication of LoRaWAN Class B and NB-IoT. NB-IoT is used in areas with good network coverage, and LoRaWAN is used in remote areas. TSN time-sensitive network nodes are deployed in key protection areas (airports, chemical plants, mines), and TSN communication is used between them and the regional collaborative gateway; Point-to-point self-organizing network communication is supported between adjacent edge sensing nodes; The regional collaborative gateway and the cloud-based intelligent management platform use 5G SA and fiber optic dual-mode communication.

5. A distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The cloud-based intelligent management platform also integrates a lightweight edge blockchain evidence storage module, which uses a simplified PBFT consensus algorithm to complete the evidence storage of key monitoring data and early warning information at the regional gateway level, with an evidence storage latency of less than 100ms.

6. A distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The multimodal data fusion algorithm based on spatiotemporal attention mechanism includes a temporal attention layer, a spatial attention layer, and a modal attention layer connected in sequence. The temporal attention layer is used to learn the weights of data at different time steps and capture the temporal evolution of lightning precursor signals. The spatial attention layer employs a graph attention network structure to learn the weights of node data from different geographical locations, thereby capturing the spatial distribution characteristics of thunderstorm clouds. The modal attention layer is used to learn the weights of different sensor modal data and automatically adjust the contribution of each sensor data in the fusion process.

7. A distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The improved conflict evidence fusion decision adopts a weighted DS evidence theory based on evidence credibility and conflict degree, and its evidence combination rule is as follows: in, For the proposition The comprehensive basic probability allocation, For the first The credibility weight of each piece of evidence The conflict coefficient, For the first The basic probability assignment function for each piece of evidence. The number of pieces of evidence.

8. A distributed lightning monitoring and early warning system based on the Internet of Things according to claim 7, characterized in that: The weight of the credibility of the evidence The calculation formula is: in, For the first The evidence and the first The distance between each piece of evidence is calculated using cosine similarity. To identify the number of propositions in the framework.

9. A distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The heterogeneous intelligent sensing node also integrates a device self-diagnosis module, which can monitor the operating status of sensors, communication modules and power modules in real time.

10. A distributed lightning monitoring and early warning system based on the Internet of Things according to claim 1, characterized in that: The distributed lightning monitoring and early warning system adopts a four-level early warning mechanism: Level 0 Warning (White): Widespread lightning activity may occur within the next 1-3 hours, issued by the cloud-based lightning precursor early detection and analysis module; Level 1 Warning (Blue): Thunderstorms are possible within the next 30 minutes, with a probability of 30%-60%. Level II Warning (Yellow): There is a high probability of thunderstorms within the next 15 minutes, with a 60%-90% chance of occurrence. Level 3 Warning (Red): There is an extremely high probability of lightning activity within the next 5 minutes, or lightning activity has already occurred and may continue, with a probability of lightning occurrence greater than 90%.

Images