A distributed multi-source heterogeneous sensor data processing method and system

Abstract

The present application relates to the technical field of data processing, and more particularly to a distributed multi-source heterogeneous sensor data processing method and system. The method comprises the following steps: collecting environmental parameters of a leakage area in real time through a distributed sensor array; aligning the coordinate system and time stamp of the environmental parameters of the leakage area to generate a standardized leakage situation data set; extracting gas diffusion characteristics and thermal imaging characteristics of the standardized leakage situation data set to calculate gas cloud diffusion path probability; generating a risk decision instruction based on a pre-set leakage level classification neural network for the gas cloud diffusion path probability to obtain the risk decision instruction; calling a matched emergency plan based on the risk decision instruction; analyzing the implementation steps of the emergency plan and converting the instruction to generate an emergency plan instruction. The present application improves the timeliness of data processing through real-time data collection, intelligent risk assessment and automated emergency response.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for processing distributed multi-source heterogeneous sensor data. Background Technology

[0002] Initially, sensor data processing focused primarily on the acquisition and analysis of data from single-type sensors, with centralized architectures being the norm. This approach suffers from drawbacks such as susceptibility to single points of failure and issues with data compatibility and processing efficiency across different sensor types. With the rapid development of cloud computing and edge computing, distributed architectures can effectively handle sensor data from diverse sources and types. This process typically involves multiple stages, including data acquisition, transmission, fusion, and analysis, each requiring consideration of key factors such as the unification of heterogeneous data formats, real-time transmission, and efficient processing. Furthermore, the continuous development of artificial intelligence and big data technologies has led to the application of deep learning and machine learning algorithms in sensor data processing, improving the accuracy and real-time performance of data analysis. However, current predictions of gas diffusion paths often rely on a single data source, making it difficult to effectively integrate gas monitoring and thermal imaging data. Additionally, traditional leak risk decision-making often depends on manual judgment, resulting in low efficiency and susceptibility to human error, ultimately leading to low responsiveness in data processing. Summary of the Invention

[0003] Therefore, it is necessary to provide a distributed multi-source heterogeneous sensor data processing method and system to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a distributed multi-source heterogeneous sensor data processing method is provided, the method comprising the following steps:

[0005] Step S1: Collect environmental parameters of the leak area in real time using a distributed sensor array; align the environmental parameters of the leak area with coordinate system 1 and timestamps to generate a standardized leak status dataset;

[0006] Step S2: Extract gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion path; generate risk decision instructions based on the gas cloud diffusion path probability using a preset leak level classification neural network to obtain risk decision instructions.

[0007] Step S3: Invoke the matching emergency plan based on the risk decision instructions; analyze the implementation steps of the emergency plan and convert the instructions to generate emergency plan instructions; prioritize the emergency plan instructions and generate a time-sequenced execution instruction set;

[0008] Step S4: Use the IoT gateway to distribute and dynamically optimize the timing execution instruction set to the intelligent explosion-proof robot terminal to perform explosion-proof operations.

[0009] This invention uses a distributed sensor array to collect environmental parameters of the leak area in real time, and performs coordinate system unification and timestamp alignment on the data to generate a standardized leak situation dataset, improving data consistency and timeliness. Subsequently, gas diffusion characteristics and thermal imaging features are extracted from this dataset to calculate the diffusion path probability of the gas cloud, and a leak level classification neural network is used for risk assessment, automatically generating risk decision instructions to ensure the accuracy and reliability of the judgment. Based on this, the system automatically matches appropriate emergency plans according to the risk decision instructions, analyzes the implementation steps of the plans to convert instructions, generates emergency plan instructions, and prioritizes the instructions to ensure the rationality and efficiency of emergency actions. Finally, an optimized set of time-series execution instructions is distributed to an intelligent explosion-proof robot terminal through an IoT gateway. The robot performs explosion-proof operations according to a dynamic optimization strategy, achieving precise and efficient accident handling, reducing human intervention, and improving safety. This process integrates technologies such as artificial intelligence, IoT, and automated control, greatly improving the intelligent and automated response capabilities to leak accidents, and is suitable for high-risk scenarios such as petrochemicals and hazardous materials storage and transportation. Therefore, this invention improves the timeliness and responsiveness of data processing through real-time data acquisition, intelligent risk assessment, and automated emergency response.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: Simultaneously acquire concentration gradient data of hydrogen sulfide, chlorine, and benzene series compounds using a fixed multi-gas detector; scan the thermal radiation distribution of the leak source using an unmanned aerial vehicle (UAV)-borne infrared spectroscopy imaging device to obtain thermal radiation distribution data; integrate the concentration gradient data and thermal radiation distribution data using a distributed sensor array to obtain environmental parameters of the leak area.

[0012] Step S12: Perform coordinate system unification processing on the environmental parameters of the leak area to generate unified leak environment data;

[0013] Step S13: Timestamp align the unified data of the leaked environment to generate leaked environment aligned data;

[0014] Step S14: Standardize the leak environment alignment data to generate a standardized leak situation dataset.

[0015] This invention achieves accurate perception and efficient modeling of the leak area environment through multi-source data fusion and standardized processing. First, a fixed multi-gas detector simultaneously collects concentration gradient data of hydrogen sulfide, chlorine, and benzene compounds, and combines this with an UAV-borne infrared spectral imaging device to scan the thermal radiation distribution of the leak source, obtaining high-precision thermal radiation data. Subsequently, a distributed sensor array fuses these heterogeneous data to form complete environmental parameters of the leak area. Next, a coordinate system is used for unified processing to achieve spatial alignment of the data, ensuring the compatibility and comparability of the multi-source data. Based on this, the system timestamps the data to further improve its timeliness and synchronization. Finally, through standardized transformation, a standardized leak situation dataset with a unified format is generated. This process significantly improves the completeness, accuracy, and real-time performance of leak monitoring data, providing high-quality input data for subsequent gas diffusion modeling, risk assessment, and emergency response. Simultaneously, this method, combining fixed sensors, UAV inspection, and intelligent data fusion technology, greatly improves the detection accuracy and response speed of hazardous chemical leaks, and is widely applicable to high-risk scenarios such as petrochemicals and hazardous chemical storage and transportation.

[0016] Preferably, step S2, which involves extracting gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion paths, includes:

[0017] Gas concentration gradient analysis was performed on the standardized leakage situation dataset to generate gas diffusion gradient data.

[0018] Thermal imaging spectral analysis was performed on the standardized leak situation dataset to extract thermal distribution features and generate thermal imaging feature data;

[0019] Spatiotemporal correlation analysis is performed based on gas diffusion gradient data and thermal imaging feature data to generate gas cloud diffusion trend data.

[0020] Probabilistic modeling is performed on the gas cloud diffusion trend data to calculate the probability of gas cloud diffusion paths under different wind speeds and environmental conditions, and to generate gas diffusion path probability data.

[0021] This invention achieves accurate prediction of gas cloud diffusion paths through gas concentration gradient analysis, thermal imaging spectral analysis, spatiotemporal correlation analysis, and probabilistic modeling. Its advantage lies in combining multi-dimensional data to construct a diffusion trend model, ensuring high accuracy in diffusion path prediction. Simultaneously, it utilizes probabilistic modeling to dynamically simulate diffusion under different wind speeds and environmental conditions, enabling the prediction results to adapt to complex and ever-changing real-world operating conditions. Based on diffusion path probability data, the system can identify high-risk areas in advance, providing a scientific basis for emergency decision-making and reducing the risk of personal injury and environmental pollution. Furthermore, this method provides high-quality data support for subsequent risk assessment and emergency plan optimization, improving the speed and accuracy of emergency response and reducing accident losses. Overall, this method enhances the intelligence level of leak monitoring, strengthens safety protection capabilities in high-risk scenarios, and is widely applicable to fields such as petrochemicals and hazardous chemical storage and transportation.

[0022] Preferably, the thermal imaging spectral analysis of the standardized leak situation dataset includes:

[0023] The standardized leakage situation dataset was imported into FLIR ResearchIR for spectral analysis parameter settings: the spectral range was set to 8μm-14μm, the frame rate to 60Hz, the temperature range to -20°C-650°C, the emissivity to 0.95, and the background temperature compensation to the ambient temperature. The spectral analysis settings data were obtained.

[0024] The non-uniformity correction of the standardized leakage situation dataset was performed by setting the data through spectral analysis, and high-frequency noise was removed by using Gaussian 3×3 filtering to obtain spectral analysis filtered data.

[0025] Based on a preset high-temperature region threshold, hotspot analysis is performed on the spectral analytical filtering data to generate thermal imaging feature data.

[0026] This invention achieves high-precision thermal imaging feature extraction through parameter optimization, non-uniformity correction, and high-frequency noise removal using the professional spectral analysis software FLIR ResearchIR. First, for a standardized leak situation dataset, the spectral range, frame rate, temperature range, and emissivity are precisely set, and temperature compensation is performed based on the ambient temperature at the site, ensuring the accuracy and adaptability of the spectral analysis. Then, the spectral analysis settings are used to correct the non-uniformity of the thermal imaging data, eliminating sensor response differences, and a Gaussian 3×3 filter is used to remove high-frequency noise, improving the signal-to-noise ratio. Finally, hotspot analysis is performed using a preset high-temperature region threshold to accurately identify the thermal anomaly characteristics of the leak area, generating high-quality thermal imaging feature data. This method effectively improves the accuracy of thermal imaging detection of leak sources, ensures the reliability of gas cloud diffusion modeling, and provides precise thermal imaging support for subsequent risk assessment and emergency response. It is widely applicable to intelligent monitoring and early warning in high-risk scenarios such as petrochemicals and hazardous chemical storage and transportation.

[0027] Preferably, the spatiotemporal correlation analysis based on gas diffusion gradient data and thermal imaging feature data includes:

[0028] Spatial interpolation analysis is performed on gas diffusion gradient data to generate continuous gas concentration field data;

[0029] Calculate the temperature gradient of the thermal imaging feature data to obtain the thermal convection effect data;

[0030] Multidimensional spatiotemporal matching of continuous gas concentration field data and thermal convection influence data is performed to generate a gas diffusion-thermal imaging correlation matrix.

[0031] Trend fitting is performed based on the gas diffusion-thermal imaging correlation matrix to generate gas cloud diffusion trend data.

[0032] This invention achieves precise correlation between gas diffusion characteristics and thermal imaging characteristics through spatial interpolation analysis, temperature gradient calculation, multidimensional spatiotemporal matching, and trend fitting, thereby improving the accuracy and reliability of diffusion trend prediction. First, continuous gas concentration field data is constructed through spatial interpolation analysis to eliminate data discreteness and improve the completeness of gas diffusion characteristics. Then, the temperature gradient of the thermal imaging characteristic data is calculated to generate thermal convection impact data, quantifying the effect of the thermal environment on gas diffusion. Next, through multidimensional spatiotemporal matching, the gas diffusion data and thermal convection data are fused to construct a gas diffusion-thermal imaging correlation matrix, achieving deep correlation analysis between gas diffusion and environmental thermal characteristics. Finally, trend fitting is performed based on the correlation matrix to generate gas cloud diffusion trend data, providing accurate predictive support for gas leak risk assessment and emergency response. This method effectively improves the precision of gas diffusion simulation, making the prediction results more consistent with actual environmental changes, and is suitable for leak monitoring and early warning systems in high-risk scenarios such as petrochemicals and hazardous chemical storage and transportation.

[0033] Preferably, the construction process of the preset leakage level classification neural network in step S2 includes:

[0034] Obtain meteorological data;

[0035] Extract the location coordinates of the leak source from the standardized leak situation dataset to obtain the leak source location data;

[0036] Meteorological data, leak source location data, and gas cloud diffusion path probability are merged to generate neural network input data;

[0037] Feedforward calculations are performed on the input data of the neural network, and the probability distribution of leakage risk level is generated by calculating through a multilayer perceptron. The calculation results are used as the output of the model to obtain the risk probability data of leakage risk level.

[0038] Post-processing of leakage risk level probability data generates final leakage risk level decision data;

[0039] A leakage level classification neural network is constructed based on the neural network input data, leakage risk level probability data, and final leakage risk level decision data to obtain the preset leakage level classification neural network.

[0040] This invention constructs a leak level classification neural network based on a multilayer sensor by fusing meteorological data, leak source location data, and the probability of gas cloud diffusion paths, achieving accurate determination and intelligent classification of leak risk levels. First, meteorological and leak source location data are used to enhance the model's perception of environmental factors, improving its adaptability to gas diffusion behavior under different operating conditions. Second, through data merging and feedforward calculation, the neural network can fully learn gas diffusion characteristics and generate a probability distribution of leak risk levels, providing high-precision data support for leak level assessment. Subsequently, post-processing optimization ensures the stability and reliability of the decision results, ultimately generating accurate leak risk level decision data. This method effectively improves the automation level and decision efficiency of leak level classification, making risk assessment more intelligent and refined, enabling rapid response to sudden leak events, providing a scientific basis for the execution of emergency plans, and is widely applicable to leak monitoring and early warning systems in high-risk fields such as petrochemicals and hazardous chemical storage and transportation.

[0041] Preferably, the emergency response plan invoked based on risk decision instructions in step S3 includes:

[0042] Analyze the leakage risk level in the risk decision instruction to obtain leakage risk level data;

[0043] Based on the leakage risk level data, emergency response requirements are extracted from the risk decision instructions to obtain emergency response requirement data.

[0044] The system queries the leakage risk level data, matches the corresponding emergency plan from the preset emergency plan database, and generates emergency plan template matching data.

[0045] A risk assessment is conducted on the selected emergency response plan template, and the steps are adjusted according to the risk assessment results to match the template with the data, resulting in a matching emergency response plan.

[0046] This invention achieves intelligent emergency response to leakage accidents by parsing risk decision-making instructions, matching them with an emergency plan database, and making dynamic adjustments, thereby improving the accuracy and adaptability of emergency plans. First, based on leakage risk level data, emergency response needs are accurately extracted to ensure the relevance of the emergency plan. Then, through an intelligent query and matching mechanism, the most suitable emergency plan is quickly located from a pre-set emergency plan database, improving response efficiency. Simultaneously, combined with risk assessment results, the selected emergency plan is optimized and adjusted to better suit the current leakage situation and environmental conditions, enhancing the feasibility and effectiveness of emergency response. Ultimately, this method can generate highly matched emergency plans, achieving rapid and accurate decision support, reducing accident losses, and improving the intelligence level of emergency response. It is widely applicable to emergency management systems in high-risk industries such as petrochemicals and hazardous chemical storage and transportation.

[0047] Preferably, step S4 includes the following steps:

[0048] Step S41: Initialize the configuration of the IoT gateway;

[0049] Step S42: Activate the explosion-proof mode by sending a set of timing execution instructions to the intelligent explosion-proof robot terminal through the configured IoT gateway, and collect execution process data of the intelligent explosion-proof robot to obtain robot execution data;

[0050] Step S43: Extract acoustic vibration features from the robot's execution data to obtain acoustic vibration feature data; construct a three-dimensional obstacle map based on the acoustic vibration feature data;

[0051] Step S44: Optimize the robot's execution data dynamically based on the 3D obstacle map to perform explosion-proof operations.

[0052] This invention achieves dynamic optimization and precise execution of intelligent explosion-proof operations through the collaborative work of an IoT gateway and an intelligent explosion-proof robot terminal. First, the initial configuration of the IoT gateway ensures the stability and reliability of the system, providing a foundation for subsequent command transmission. Then, the intelligent explosion-proof robot is activated by executing a time-sequenced instruction set, and the robot's execution process is monitored through real-time data acquisition, generating detailed robot execution data for subsequent analysis and optimization. Based on the acoustic vibration characteristics in the execution data, the system can construct a three-dimensional obstacle map, achieving environmental perception and obstacle recognition, thereby providing more efficient path planning and decision support for explosion-proof operations. Finally, through dynamic instruction optimization, the system can adjust the robot's operation in real time according to the actual execution situation, making explosion-proof operations more precise and safer, and avoiding potential risks. This method effectively improves the intelligence level of explosion-proof operations, ensuring safety and efficiency during the operation process, and is widely applicable to automated explosion-proof operations in high-risk fields such as petrochemicals and hazardous materials handling.

[0053] Preferably, step S41 includes the following steps:

[0054] Step S411: Configure the IoT gateway hardware parameters: the gateway power input range is 12V±5%DC, the gateway operating temperature range is set to -20℃ to +60℃, the gateway communication interface is configured to dual-band 2.4GHz and 5GHz, and the signal transmission rate is set to 150Mbps or 433Mbps.

[0055] Step S412: Configure the communication protocol for the IoT gateway: Set the data transmission protocol to MQTT and the message quality service level to QoS1;

[0056] Step S413: Configure the network connection and bandwidth management of the IoT gateway: limit the gateway's bandwidth usage to 10Mbps.

[0057] This invention ensures system stability, reliability, and efficient data transmission capabilities through fine-tuning of the hardware parameters, communication protocols, and network connections of an IoT gateway. First, hardware parameter configuration ensures the gateway operates normally within specific power input and temperature ranges, and optimizes communication frequency bands and signal transmission rates, providing stable wireless connectivity. Second, communication protocol configuration employs the MQTT protocol and an appropriate Quality of Service (QoS) level 1, making data transmission more reliable, reducing the risk of data loss, and improving communication efficiency. Finally, network connection and bandwidth management configuration ensures the gateway operates stably under a specified static IP address, while limiting bandwidth usage, optimizing network resource allocation, and avoiding the impact of bandwidth overload on system performance. Overall, this configuration method provides efficient hardware support and communication management for IoT systems, ensuring the stable operation and real-time data transmission of explosion-proof operating systems, and is widely applicable to intelligent monitoring and emergency response systems in fields such as petrochemicals and hazardous materials storage and transportation.

[0058] This specification provides a distributed multi-source heterogeneous sensor data processing system for executing the above-described distributed multi-source heterogeneous sensor data processing method. The distributed multi-source heterogeneous sensor data processing system includes:

[0059] The data acquisition module is used to collect environmental parameters of the leak area in real time through a distributed sensor array; the environmental parameters of the leak area are aligned with coordinate system and timestamp to generate a standardized leak status dataset;

[0060] The risk decision module is used to extract gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion path; based on the preset leak level classification neural network, risk decision instructions are generated for the probability of gas cloud diffusion path to obtain risk decision instructions.

[0061] The instruction generation module is used to invoke the matching emergency plan based on the risk decision instruction; analyze the implementation steps of the emergency plan and convert them into instructions to generate emergency plan instructions; prioritize the emergency plan instructions and generate a time-sequenced execution instruction set.

[0062] The instruction distribution module is used to distribute and dynamically optimize the timing execution instruction set to the intelligent explosion-proof robot terminal using the IoT gateway to perform explosion-proof operations.

[0063] The beneficial effects of this invention lie in the coordinated operation of modules such as data acquisition, risk decision-making, instruction generation, and instruction distribution, which achieves accurate assessment of leakage risks, intelligent emergency response, and efficient explosion-proof operations. First, the data acquisition module monitors environmental parameters in the leakage area in real time using a distributed sensor array and aligns them with timestamps through a unified coordinate system to generate a standardized leakage situation dataset, providing basic data support for subsequent decision-making. Next, the risk decision-making module extracts gas diffusion characteristics and thermal imaging features, and combines them with a pre-defined leakage level classification neural network to calculate path probability, thereby generating accurate risk decision-making instructions. These instructions provide a scientific basis for the invocation and execution of emergency plans. Subsequently, the instruction generation module matches emergency plans, performs risk assessments and adjusts steps to ensure the relevance and operability of the emergency plan; simultaneously, it prioritizes emergency plan instructions to ensure that the most urgent tasks are handled first. Finally, the instruction distribution module, through the collaborative operation of an IoT gateway and an intelligent explosion-proof robot terminal, distributes and optimizes execution instructions in real time, achieving dynamic adjustment and precise execution of explosion-proof operations. Overall, this system improves the efficiency and accuracy of emergency response to leaks, ensures the safety and intelligence of explosion-proof operations, and is suitable for automated monitoring and emergency management in high-risk industries such as petrochemicals and hazardous chemical storage and transportation. Therefore, this invention improves the timeliness and responsiveness of data processing through real-time data acquisition, intelligent risk assessment, and automated emergency response. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the steps of a distributed multi-source heterogeneous sensor data processing method.

[0065] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S1.

[0066] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S4.

[0067] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0068] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0069] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0070] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0071] To achieve the above objectives, please refer to Figures 1 to 3 A distributed multi-source heterogeneous sensor data processing method, the method comprising the following steps:

[0072] Step S1: Collect environmental parameters of the leak area in real time using a distributed sensor array; align the environmental parameters of the leak area with coordinate system 1 and timestamps to generate a standardized leak status dataset;

[0073] Step S2: Extract gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion path; generate risk decision instructions based on the gas cloud diffusion path probability using a preset leak level classification neural network to obtain risk decision instructions.

[0074] Step S3: Invoke the matching emergency plan based on the risk decision instructions; analyze the implementation steps of the emergency plan and convert the instructions to generate emergency plan instructions; prioritize the emergency plan instructions and generate a time-sequenced execution instruction set;

[0075] Step S4: Use the IoT gateway to distribute and dynamically optimize the timing execution instruction set to the intelligent explosion-proof robot terminal to perform explosion-proof operations.

[0076] This invention uses a distributed sensor array to collect environmental parameters of the leak area in real time, and performs coordinate system unification and timestamp alignment on the data to generate a standardized leak situation dataset, improving data consistency and timeliness. Subsequently, gas diffusion characteristics and thermal imaging features are extracted from this dataset to calculate the diffusion path probability of the gas cloud, and a leak level classification neural network is used for risk assessment, automatically generating risk decision instructions to ensure the accuracy and reliability of the judgment. Based on this, the system automatically matches appropriate emergency plans according to the risk decision instructions, analyzes the implementation steps of the plans to convert instructions, generates emergency plan instructions, and prioritizes the instructions to ensure the rationality and efficiency of emergency actions. Finally, an optimized set of time-series execution instructions is distributed to an intelligent explosion-proof robot terminal through an IoT gateway. The robot performs explosion-proof operations according to a dynamic optimization strategy, achieving precise and efficient accident handling, reducing human intervention, and improving safety. This process integrates technologies such as artificial intelligence, IoT, and automated control, greatly improving the intelligent and automated response capabilities to leak accidents, and is suitable for high-risk scenarios such as petrochemicals and hazardous materials storage and transportation. Therefore, this invention improves the timeliness and responsiveness of data processing through real-time data acquisition, intelligent risk assessment, and automated emergency response.

[0077] In this embodiment of the invention, reference is made to Figure 1 The diagram shown is a flowchart illustrating the steps of a distributed multi-source heterogeneous sensor data processing method according to the present invention. In this example, the distributed multi-source heterogeneous sensor data processing method includes the following steps:

[0078] Step S1: Collect environmental parameters of the leak area in real time using a distributed sensor array; align the environmental parameters of the leak area with coordinate system 1 and timestamps to generate a standardized leak status dataset;

[0079] In this embodiment of the invention, a distributed sensor array, including gas concentration sensors (such as...), is deployed in the leak risk area of ​​a chemical plant. Sensors include PID VOC sensors, temperature and humidity sensors, wind speed and direction sensors, and air pressure sensors. Each sensor collects data at set time intervals (e.g., 1 second or less) and transmits the data to the edge computing unit. Sensor nodes perform local data cleaning, such as removing sensor misreads, filling in short-term lost data, and reducing noise through low-pass filtering. All sensors in the leak area are pre-calibrated with geographical coordinates (e.g., GPS coordinates or the factory's local coordinate system) during installation, and the collected data is uniformly converted to the same coordinate system. Due to the asynchronous timestamps of the sensor data, Network Time Protocol (NTP) or GPS timing technology is used to unify the timestamps, and data is aligned using interpolation. All collected data is stored in standard JSON or CSV format, including time, coordinates, gas concentration, temperature, humidity, wind speed, and direction. Standardized data is stored in a local database or cloud data center to support subsequent modeling and analysis.

[0080] Step S2: Extract gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion path; generate risk decision instructions based on the gas cloud diffusion path probability using a preset leak level classification neural network to obtain risk decision instructions.

[0081] In this embodiment of the invention, gas concentration data at different times and locations (e.g., [data missing]) are extracted from a standardized leakage situation dataset. (etc.). Calculate the concentration change rate (gradient) to form a gas diffusion trend curve: ;in The rate of change of gas concentration. For position In time The gas concentration at that location, For time. The propagation range of the gas is estimated using a Gaussian diffusion model: ;in For the strength of the leakage source, and Where is the diffusion coefficient. , These are the location coordinates. Combined with wind speed and direction data, the gas diffusion model is corrected to form a more accurate wind field-corrected diffusion model: ;in For wind speed, To correct for the temperature fluctuations, a thermal infrared camera (such as a FLIR thermal imager) is used to photograph the leak area and acquire thermal imaging data. Pseudo-color mapping is then performed to extract areas of temperature anomalies (where the leak point temperature differs significantly from the background temperature). A background temperature baseline is established, and the temperature difference is calculated. ,like Areas with high risk are marked as such. Neural networks such as YOLOv5 and ResNet are used to extract features from thermal imaging images, detect the distribution of hot spots in the leak area, and extract feature vectors: ,in To input thermal imaging data, This is the output feature vector. A gas diffusion probability model is constructed by combining gas concentration distribution, wind speed and direction, and thermal imaging features: ;in , and These are weighting coefficients. Using temporal neural networks such as LSTM and Transformer, and inputting gas diffusion data and historical leakage data, the future diffusion path is predicted. The system calculates the probability of high-risk areas and generates a heat map of the gas cloud's diffusion path. A classification neural network (such as the Transformer classifier) ​​is used to classify the gas cloud's diffusion path into three risk categories: low risk (0-30%), medium risk (30-60%), and high risk (above 60%). Based on the classification results, corresponding emergency commands are generated: Low risk: Close monitoring and early warning; Medium risk: Activate local ventilation and alarms; High risk: Immediate evacuation of personnel and activation of emergency response.

[0082] Step S3: Invoke the matching emergency plan based on the risk decision instructions; analyze the implementation steps of the emergency plan and convert the instructions to generate emergency plan instructions; prioritize the emergency plan instructions and generate a time-sequenced execution instruction set;

[0083] In this embodiment of the invention, an emergency response plan database is constructed using knowledge graph technology, and the correlation between different leakage scenarios (e.g., chemical plants, underground pipelines) and the plans is established through an entity relationship network. This includes a relationship chain of "leakage type → gas properties → affected area → plan type → key action steps". Natural Language Processing (NLP) is used to parse risk decision instructions, and the best emergency response plan is automatically matched through graph database queries (e.g., Neo4j). Cosine similarity or Jaccard similarity is used to calculate the optimal matching plan. Dependency Parsing is used to decompose the instructions in the emergency response plan. For example, "start the gas neutralization system" is broken down into "operation object = gas neutralization system, operation = start". Key actions (e.g., "blockade", "evacuation", "detection") are extracted based on Named Entity Recognition (NER) and categorized into standardized instruction templates. The parsed instructions are converted into a structured format suitable for subsequent sorting and execution. Key indicators of emergency response (e.g., personnel safety, leak control, environmental impact) are defined. The relative weight of each indicator is calculated using the Analytic Hierarchy Process (AHP) method. For example: personnel safety (weight 0.5), leak control (weight 0.3), and equipment protection (weight 0.2). Calculate a comprehensive score based on the indicators involved in the instructions and rank them. Define state nodes, with each emergency action as a state, such as [Leak Detection] → [Blockade] → [Personnel Evacuation] → [Gas Neutralization] → [Environmental Monitoring]. Use Boolean logic or conditional triggers (e.g., "Neutralization system can only be activated after blockade is completed") to ensure the order of instructions meets emergency requirements. Adjust the execution order of timing instructions based on real-time feedback (e.g., sensor data). Use Petri nets to model the dependencies between instructions, ensuring that subsequent steps are executed only after the preceding tasks are completed.

[0084] Step S4: Use the IoT gateway to distribute and dynamically optimize the timing execution instruction set to the intelligent explosion-proof robot terminal to perform explosion-proof operations.

[0085] In this embodiment of the invention, environmental parameters such as temperature, humidity, and flammable gas concentration are analyzed based on operational requirements. The task type of the explosion-proof robot is determined, such as inspection, fire extinguishing, leak detection, and hazard mitigation. The task is broken down into multiple execution steps, identifying key nodes. Specific actions to be performed by the robot are determined, including forward movement, rotation, scanning, detection, and grasping. The instruction set includes task number, execution order, action type, execution parameters, and priority. A time window for task execution is set to ensure that critical tasks are completed first. The explosion-proof robot is ensured to have a normal connection with the IoT gateway, supporting multiple communication protocols such as 5G, WiFi, and LoRa. Data encryption and authentication mechanisms are configured to prevent instruction tampering. The IoT gateway parses the task plan and distributes task instructions to the explosion-proof robot according to priority. After receiving the instructions, the robot verifies them and returns confirmation information. According to the received timing instructions, the robot executes each action sequentially, such as moving, detecting, obstacle avoidance, and operating the robotic arm. During execution, the robot monitors environmental data in real time; if any abnormality occurs, the task can be paused and an alarm sent. The robot periodically uploads its execution status to the IoT gateway, including task progress, anomalies, and real-time monitoring data. Based on the feedback, the IoT gateway adjusts the command issuance rhythm to adapt to the actual working environment. The IoT gateway analyzes the data returned by the robot to determine if the task progress meets expectations. It optimizes the task execution sequence using AI algorithms, such as adjusting inspection routes in high-temperature areas to reduce risk exposure. If the robot encounters obstacles or detects abnormal gases, the IoT gateway can immediately adjust commands, modify the movement path, or execute a new task. Upon task completion, the system automatically generates a work report and stores it in the cloud for subsequent analysis and optimization. After completing its task, the robot sends a task completion signal to the IoT gateway. The IoT gateway records all execution data and performs closed-loop management of the task.

[0086] As an example of the present invention, reference is made to Figure 2 As shown, in this example, step S1 includes:

[0087] Step S11: Simultaneously acquire concentration gradient data of hydrogen sulfide, chlorine, and benzene series compounds using a fixed multi-gas detector; scan the thermal radiation distribution of the leak source using an unmanned aerial vehicle (UAV)-borne infrared spectroscopy imaging device to obtain thermal radiation distribution data; integrate the concentration gradient data and thermal radiation distribution data using a distributed sensor array to obtain environmental parameters of the leak area.

[0088] Step S12: Perform coordinate system unification processing on the environmental parameters of the leak area to generate unified leak environment data;

[0089] Step S13: Timestamp align the unified data of the leaked environment to generate leaked environment aligned data;

[0090] Step S14: Standardize the leak environment alignment data to generate a standardized leak situation dataset.

[0091] In this embodiment of the invention, fixed multi-gas detectors are installed in and around the potential leak area to ensure coverage of key locations such as emission outlets, downstream areas, and low-lying areas. Concentration gradients of hydrogen sulfide, chlorine, and benzene compounds are simultaneously detected to obtain real-time concentration data and spatial distribution. The collected data includes time, concentration values, and location information, forming a gas concentration gradient data table. A drone flight path is set to cover the leak source area, and an infrared spectral imaging device is used for scanning. Infrared imaging technology is used to detect the leak point and surrounding areas with abnormal temperatures, generating thermal radiation distribution data. The infrared image data includes time, spatial coordinates, and a temperature distribution matrix. A distributed sensor network is used to transmit the gas concentration gradient data and thermal radiation distribution data to a central processing unit. A heterogeneous data integration algorithm is used, based on spatial location matching and environmental feature association, to fuse the multi-gas concentration data and infrared spectral data to generate environmental parameters for the leak area. These parameters include complete environmental parameters such as gas concentration, temperature distribution, sensor location information, and timestamps. Spatial coordinates from different data sources (such as GPS coordinates, relative position coordinates, and drone image coordinates) are uniformly converted to a unified geographic coordinate system (WGS84 or UTM coordinates). Based on spatial interpolation algorithms, location matching is performed on sensor and UAV data to ensure all data use the same coordinate reference. A spatial parameter matrix of the leak area is constructed, enabling all data points to be analyzed within the same coordinate framework. A unified time reference (e.g., UTC time) is adopted to ensure time consistency across all data sources (gas sensors, UAV imagery, sensor arrays). Time interpolation algorithms are used to perform time correction on asynchronously acquired data, ensuring that each data point is matched with the same time step. The aligned data is stored as a time series, forming a time-series leak environment dataset. Timestamp anomalies or missing data are detected, and interpolation or predictive completion techniques are used to ensure time continuity, ultimately generating aligned leak environment data. Concentration values, temperature values, and time series data are normalized to ensure data values ​​fall within a uniform range (e.g., 0-1). Statistical analysis methods (e.g., mean deviation detection) are used to remove outliers, preventing erroneous data from affecting the analysis results and generating a standardized leak situation dataset.

[0092] Preferably, step S2, which involves extracting gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion paths, includes:

[0093] Gas concentration gradient analysis was performed on the standardized leakage situation dataset to generate gas diffusion gradient data.

[0094] Thermal imaging spectral analysis was performed on the standardized leak situation dataset to extract thermal distribution features and generate thermal imaging feature data;

[0095] Spatiotemporal correlation analysis is performed based on gas diffusion gradient data and thermal imaging feature data to generate gas cloud diffusion trend data.

[0096] Probabilistic modeling is performed on the gas cloud diffusion trend data to calculate the probability of gas cloud diffusion paths under different wind speeds and environmental conditions, and to generate gas diffusion path probability data.

[0097] In this embodiment of the invention, a standardized leak situation dataset is collected, including concentration data from gas sensors (such as gas detectors and infrared sensors) at different locations. The data from all measurement points are spatially discretized to generate grid-like data (e.g., each grid is 1m × 1m). The measured gas concentration is assigned to the corresponding grid locations. Using the finite difference method or gradient calculation, the concentration difference between adjacent grid points is calculated to obtain the gas concentration gradient (unit: ppm / m³ or...). ). formula: ,in and These represent the gas concentrations at two adjacent grid points. This is the distance between the two grid points. The calculated gas diffusion gradient data is saved as a data matrix, representing the gas concentration variation trend at different locations. Thermal imaging of the leak area is acquired using a thermal imaging instrument (such as an infrared thermal imager). Ensure that the thermal imaging data and gas concentration data are synchronized in time and space. Image processing is performed on the thermal imaging data to extract thermal distribution features. Thermal image analysis methods are typically used, focusing on areas with significant temperature changes, especially temperature fluctuations near the leak source. The average temperature of each grid area is calculated, identifying temperature anomalies associated with the gas leak (such as changes in the surrounding heat distribution due to heat absorption or release from a gas cloud). A threshold-based detection method is used, setting a threshold temperature and identifying areas in the thermal image exceeding this temperature as potential leak areas, generating thermal imaging feature data representing the areas of temperature anomalies and their characteristic values ​​(such as temperature, range, duration, etc.). The gas diffusion gradient data and thermal imaging feature data are fused. Spatiotemporal correlation analysis is performed, considering changes in gas concentration and the spatial distribution of temperature anomaly areas. A Kalman filter or mutual information method is used to combine these two data sources to more accurately predict the diffusion path and trend of gas clouds. Based on the correlation between gas diffusion gradients and thermal imaging features, the particle tracking method is used to simulate the diffusion trend of gas clouds. The spatial location of the gas cloud at different time steps is calculated according to the direction, velocity, and temperature anomalies of gas diffusion. The diffusion path of the gas cloud is simulated, taking into account environmental factors such as wind speed and direction, to estimate the diffusion area of ​​the gas cloud and generate gas cloud diffusion trend data, including diffusion path, velocity, and direction information, representing the diffusion dynamics of the gas cloud. Based on the gas cloud diffusion trend data, considering different environmental factors such as wind speed, wind direction, temperature, and humidity, the probability of the gas cloud diffusion path is calculated using Monte Carlo simulation or a Bayesian network. A stochastic process model (such as a random walk model) is used to simulate and calculate the diffusion probability of the gas cloud under different conditions. The probability of each diffusion path is evaluated, and the gas diffusion model is trained and optimized based on historical data. For each diffusion path, the probability of its occurrence at different time points and under different environmental conditions is calculated. For example, factors such as wind speed, wind direction, temperature, and humidity all affect path selection. Maximum likelihood estimation (MLE) or least squares is used to fit the model to ensure that the probabilistic model accurately reflects the diffusion characteristics of the gas cloud, generating gas diffusion path probability data for subsequent risk assessment, emergency response planning, and leak source identification.

[0098] Preferably, the thermal imaging spectral analysis of the standardized leak situation dataset includes:

[0099] The standardized leakage situation dataset was imported into FLIR ResearchIR for spectral analysis parameter settings: the spectral range was set to 8μm-14μm, the frame rate to 60Hz, the temperature range to -20°C-650°C, the emissivity to 0.95, and the background temperature compensation to the ambient temperature. The spectral analysis settings data were obtained.

[0100] The non-uniformity correction of the standardized leakage situation dataset was performed by setting the data through spectral analysis, and high-frequency noise was removed by using Gaussian 3×3 filtering to obtain spectral analysis filtered data.

[0101] Based on a preset high-temperature region threshold, hotspot analysis is performed on the spectral analytical filtering data to generate thermal imaging feature data.

[0102] In this embodiment of the invention, a standardized leak situation dataset is imported into FLIR ResearchIR software. This dataset should include thermal imaging data and related time and spatial coordinate information. The spectral analysis band range is set to 8μm-14μm, which falls within the long-wave infrared (LWIR) band and is suitable for detecting thermal imaging changes caused by gas leaks. The frame rate is set to 60Hz to ensure sufficiently high temporal resolution for data acquisition, suitable for monitoring rapidly changing heat distribution. The temperature range is set to -20°C to 650°C to accommodate temperature changes caused by different gas leaks, covering a range from low to high temperatures. The emissivity is set to 0.95, which is a common emissivity for most object surfaces and is suitable for most leaking objects. The Beijing temperature compensation is set to the average ambient temperature, and compensation is performed based on the ambient temperature to ensure the accuracy of the thermal imaging data. The software generates spectral analysis setting data based on the above settings, including spectral range, frame rate, temperature range, etc., to prepare for subsequent spectral analysis. The spectral analysis setting data is used to perform non-uniformity correction on the standardized leak situation dataset. This correction step is to correct the spatial non-uniform response of the detector in the thermal imaging device, thereby improving the accuracy of the thermal image. In FLIR ResearchIR, selecting the appropriate correction function will automatically correct the non-uniformity of the thermal imaging image. The Gaussian 3×3 filtering algorithm is selected and applied to the spectral analysis data for noise removal. Gaussian filtering reduces high-frequency noise by smoothing the image, ensuring a clearer thermal image and avoiding unnecessary image interference. In FLIR ResearchIR, the filter is set to Gaussian 3×3 and applied to the dataset for smoothing. The noise in the filtered image is effectively removed, resulting in spectral analysis filtered data. At this point, the thermal imaging data has undergone non-uniformity correction and noise removal, presenting a more realistic and clear thermal distribution. A threshold for high-temperature areas is set according to the actual situation. This threshold should be able to distinguish temperature change areas near the leak source. For example, setting the threshold to a certain temperature point, such as 50°C, will treat all areas exceeding this temperature value as heat source areas. In FLIR ResearchIR, this temperature point is set using the threshold tool. Based on the spectral analysis filtered data, hotspot analysis is performed using the software to identify areas in the thermal image that exceed the set threshold, i.e., hotspot areas. These areas typically represent the location of gas leak sources or gas clouds. In FLIR ResearchIR, the hotspot analysis tool automatically identifies areas of temperature anomalies and generates heat maps. Based on the hotspot analysis results, thermal imaging feature data is generated, including the spatial distribution of hotspot areas, temperature values, and duration. The feature values ​​of the hotspot areas (such as temperature, location, and duration) are extracted to provide a basis for subsequent analysis and decision-making.

[0103] Preferably, the spatiotemporal correlation analysis based on gas diffusion gradient data and thermal imaging feature data includes:

[0104] Spatial interpolation analysis is performed on gas diffusion gradient data to generate continuous gas concentration field data;

[0105] Calculate the temperature gradient of the thermal imaging feature data to obtain the thermal convection effect data;

[0106] Multidimensional spatiotemporal matching of continuous gas concentration field data and thermal convection influence data is performed to generate a gas diffusion-thermal imaging correlation matrix.

[0107] Trend fitting is performed based on the gas diffusion-thermal imaging correlation matrix to generate gas cloud diffusion trend data.

[0108] In this embodiment of the invention, gas diffusion gradient data is ensured to be ready, containing gas concentration gradient information at multiple spatial locations. This data is typically collected by gas sensors at different locations and times. To generate continuous gas concentration field data, spatial interpolation techniques such as Kriging Interpolation or Inverse Distance Weighted (IDW) are used to interpolate the gas diffusion gradient data. Kriging Interpolation estimates the gas concentration at each location based on the spatial variogram. First, the spatial correlation between each location is calculated, and then these correlations are used to estimate the concentration value in the blank area. Inverse Distance Weighted (IDW) generates the gas concentration in the blank area by weighted averaging of data from surrounding known locations. Using interpolation algorithms, a continuous gas concentration field dataset is calculated and generated, representing the gas concentration value at each spatial location. Temperature gradient analysis is performed on the thermal imaging feature data to calculate the rate of temperature change at each spatial location, thereby obtaining the thermal convection influence data. The thermal imaging image is processed using the finite difference method or a gradient operator. ;in For temperature, For temperature gradient, , , These represent the image's position on the spatial coordinate axes. The calculated thermal convection impact data is saved as a data matrix, representing the temperature change trend of the thermal imaging area. Continuous gas concentration field data and thermal convection impact data are integrated into a unified multidimensional data structure. Each data point contains information on gas concentration and temperature changes, possessing a spatiotemporal dimension. The data includes a spatial dimension (location coordinates) and a temporal dimension (gas concentration and temperature data at different time points). Spatiotemporal matching of the continuous gas concentration field data and thermal convection impact data is performed, using techniques such as cross-correlation analysis or dynamic time warping (DTW) to find the correlation between them. Cross-correlation analysis assesses the relationship between gas concentration changes and temperature gradients by calculating the correlation coefficient between the two. Dynamic time warping (DTW), when processing time series data, assesses their spatiotemporal correlation by comparing the degree of matching between gas concentration and temperature gradients at different time points. The spatiotemporal matching results are stored as a gas diffusion-thermal imaging correlation matrix, which reflects the correlation between the gas diffusion process and the thermal imaging data. Each element in the matrix represents the correlation between gas concentration and temperature change at a specific spatiotemporal location. Based on the gas diffusion-thermal imaging correlation matrix, a suitable fitting algorithm is selected for trend modeling. This matrix provides relevant information between gas diffusion and thermal imaging and can be used as input data. Methods such as polynomial fitting, exponential fitting, or support vector regression (SVR) are used to model the diffusion trend of gas clouds. The least squares method is employed to fit the gas diffusion-thermal imaging correlation matrix to find the optimal trend curve or trend function. For nonlinear relationships, curve fitting methods can be used for more accurate trend modeling, as shown in the formula: ;in, For time or space variables, These are the fitting parameters. The fitted trend model generates gas cloud diffusion trend data, representing the diffusion trend of the gas cloud over time and space. This data is used to predict future gas cloud behavior and its potential risks.

[0109] Of particular importance, the multidimensional spatiotemporal matching of continuous gas concentration field data and thermal convection influence data also includes:

[0110] The continuous gas concentration field data is discretized to generate gridded concentration data; temporal features are extracted from the gridded concentration data to generate concentration change feature data.

[0111] Convection feature extraction is performed on the thermal convection impact data to generate thermal convection feature data; spatial distribution analysis is performed on the concentration change feature data to generate diffusion pattern data;

[0112] Flow field structure analysis is performed on thermal convection characteristic data to generate flow field structure data; time synchronization is performed on diffusion mode data to generate time-aligned data; spatial registration is performed on flow field structure data to generate spatial mapping data.

[0113] The spatiotemporal correlation between time-series aligned data and spatially mapped data is analyzed to generate multidimensional correlation data; matrix operations are performed on the multidimensional correlation data to generate a gas diffusion-thermal imaging correlation matrix.

[0114] In this embodiment of the invention, continuous gas concentration field data is acquired, which includes gas concentration values ​​at different times and spatial locations. The gas concentration field is then gridded, discretizing the entire field into multiple grid cells. Each grid cell represents the gas concentration at a spatial location. Regular or adaptive gridding methods can be used: Regular gridding divides the gas concentration field into regular grids based on a predetermined grid resolution (e.g., each grid unit is 1m × 1m). Adaptive gridding automatically adjusts the grid resolution according to the drastic changes in gas concentration, using smaller grid cells in areas with significant concentration variations. The generated gridded concentration data includes the gas concentration value within each grid cell, providing foundational data for subsequent analysis. Concentration values ​​at multiple time points for each grid cell are extracted from the gridded concentration data to form time series data. The time series data for each grid cell is analyzed to extract the following temporal features: Mean: Represents the average concentration, reflecting the overall gas concentration over a certain period. Standard Deviation: Represents the degree of concentration fluctuation, reflecting the variability of gas concentration. Rate of Change: Calculates the rate of change of concentration between adjacent time points, reflecting the speed at which the concentration changes over time. Periodic analysis: This involves analyzing the periodic changes in concentration over time to determine whether gas diffusion exhibits periodic fluctuations. Time-series analysis generates characteristic data on concentration changes, describing the trend and fluctuations of gas concentration over time. Data on the impact of thermal convection is acquired, typically through thermal imaging or temperature gradient data. Thermal convection characteristics are calculated, primarily including: calculating the temperature gradient at each location to understand how heat diffuses from high-temperature to low-temperature regions; inferring the direction and velocity of heat flow based on the temperature gradient to determine the flow pattern of thermal convection; assessing the intensity of heat flow, usually expressed by calculating the rate of temperature change, generating characteristic data on thermal convection describing the temperature gradient, flow direction, and intensity of the convection region; acquiring characteristic data on concentration changes, derived from the time-series feature extraction process of gridded concentration data; and using spatial interpolation methods (such as Kriging interpolation and IDW) to spatially interpolate the concentration change characteristic data, generating a continuous concentration change distribution map. The spatial distribution pattern of concentration changes is analyzed, including the distribution of high-concentration and low-concentration regions and the spatial range of gas diffusion. Based on spatial analysis, diffusion pattern data is generated to describe the spatial distribution trend of gas diffusion and identify diffusion paths and hotspot regions. Thermal convection characteristic data is acquired, including information such as the direction and intensity of heat flow. Fluid dynamics analysis methods (such as flow field simulation or particle tracking) are used to analyze the flow field structure in the thermal convection region. Analysis objectives include: flow field stability: determining whether the flow field is stable and whether eddies or unstable regions exist; flow field extensibility: predicting whether thermal convection will extend outwards, generating flow field structure data to describe the flow structure and patterns of thermal convection, providing fundamental data for subsequent spatiotemporal matching.Diffusion model data is synchronized in time using interpolation or Dynamic Time Warping (DTW) to ensure comparison within the same time dimension. Interpolation of the diffusion model data based on time intervals generates continuous time series data. Irregular time series are aligned to ensure synchronization between different time series. Time-aligned data is output, synchronizing the diffusion models with time and providing the correct time dimension for spatiotemporal analysis. Spatial registration algorithms (such as rigid transformation, affine transformation, and non-rigid transformation) are used to spatially register the flow field structure data with the gas concentration field. The flow field structure data is translated, rotated, and scaled to align with the gas concentration field data. For complex flow fields, more complex transformation models are used for registration. Spatial mapping data is output, describing the spatial relationship between thermal convection structure and gas concentration distribution. Spatiotemporal correlation between time-aligned data and spatial mapping data is calculated using correlation analysis (such as Pearson correlation and mutual information). Methods such as weighted averaging or principal component analysis (PCA) are used to fuse these two types of data and analyze their spatiotemporal dynamic characteristics. The output is multidimensional correlation data, which describes the spatiotemporal relationship between gas diffusion and thermal convection, providing a basis for predicting gas diffusion paths. Matrix operations, such as matrix multiplication, eigenvalue decomposition, or singular value decomposition (SVD), are performed on the multidimensional correlation data to extract key features. These features are then used to generate a gas diffusion-thermal imaging correlation matrix, reflecting the complex relationship between gas diffusion and thermal imaging data.

[0115] Preferably, the construction process of the preset leakage level classification neural network in step S2 includes:

[0116] Obtain meteorological data;

[0117] Extract the location coordinates of the leak source from the standardized leak situation dataset to obtain the leak source location data;

[0118] Meteorological data, leak source location data, and gas cloud diffusion path probability are merged to generate neural network input data;

[0119] Feedforward calculations are performed on the input data of the neural network, and the probability distribution of leakage risk level is generated by calculating through a multilayer perceptron. The calculation results are used as the output of the model to obtain the risk probability data of leakage risk level.

[0120] Post-processing of leakage risk level probability data generates final leakage risk level decision data;

[0121] A leakage level classification neural network is constructed based on the neural network input data, leakage risk level probability data, and final leakage risk level decision data to obtain the preset leakage level classification neural network.

[0122] In this embodiment of the invention, real-time or historical meteorological data, including but not limited to temperature, wind speed, wind direction, humidity, and atmospheric pressure, is acquired. This data can be obtained from meteorological stations, satellite data, weather forecasting systems, or meteorological APIs. Meteorological data can be stored in CSV, JSON, or API response formats, ensuring that the data includes timestamps and various meteorological parameters. The acquired meteorological data is cleaned, missing and outlier values ​​are processed, and the timestamps are ensured to be consistent with other data for easy subsequent fusion. A standardized leak situation dataset is acquired, which includes information about the leak source (such as location, leakage amount, and leak type). Leak source location coordinates are extracted from the leak situation dataset; these locations are typically values ​​in longitude, latitude, or spatial coordinate systems. If the data contains multiple leak sources, coordinate extraction is performed for each leak source to generate corresponding leak source location data. The extracted leak source location data is standardized, mapping it to a format suitable for neural network input (e.g., normalizing coordinate values ​​to the 0-1 range). This data is aligned according to timestamps or spatial coordinates: the meteorological data and leak source location data are merged with gas diffusion path probability data. The merged dataset can use time series data or spatial coordinate systems as the basis for matching. If the data formats are inconsistent, interpolation or padding methods can be used to align data from different sources at a unified time point or location, generating merged neural network input data, including meteorological information, leak source location, and diffusion path probability, ready as input to the neural network. A multilayer perceptron (MLP) neural network architecture is designed, typically including: an input layer: receiving the merged input data; hidden layers: containing at least one or more neurons, adjusting the number of layers and neurons according to task complexity and training data volume; and an output layer: outputting the probability distribution of leak risk levels. A suitable activation function (such as ReLU, Sigmoid, or Softmax) is selected to increase non-linearity. The neural network input data generated in step S23 is input into the neural network model. The neural network performs calculations through a feedforward process: data enters from the input layer, is weighted and summed and activated by neurons in the hidden layer, and then passed to the output layer. The output layer generates the probability distribution of leak risk levels, i.e., the probability of occurrence for each risk level. The neural network outputs the probability distribution of leak risk levels, providing a corresponding probability value for each risk level. Post-processing of the output leakage risk level probabilities commonly includes: determining the final leakage risk level based on a set threshold; selecting the risk level with the highest probability as the final classification result; and, if multiple risk levels have high probability values, using a weighted average method for comprehensive decision-making. This involves selecting probability thresholds or making multi-class decisions based on the leakage risk level probability data.To ensure the final result meets practical needs, for example, if the risk level is divided into low, medium, and high, then the risk level is determined when the probability exceeds a certain threshold, generating the final leakage risk level decision data, i.e., the risk level selected based on the model output and the threshold. Using the input data, output data, and post-processing results from steps S23 and S24, train the neural network to construct a leakage level classification neural network. Select an appropriate loss function (such as the cross-entropy loss function) for model optimization to ensure the neural network can effectively classify different leakage risk levels. Train the model using the backpropagation algorithm, continuously adjusting the weight parameters of the neural network until the loss function reaches its minimum. Use historical data for training and evaluate the model's performance using cross-validation. Evaluate the model's precision, recall, F1 score, and other metrics based on the validation data during training, and select the best model. The completed leakage level classification neural network can perform leakage risk assessment in real-time or offline.

[0123] Preferably, the emergency response plan invoked based on risk decision instructions in step S3 includes:

[0124] Analyze the leakage risk level in the risk decision instruction to obtain leakage risk level data;

[0125] Based on the leakage risk level data, emergency response requirements are extracted from the risk decision instructions to obtain emergency response requirement data.

[0126] The system queries the leakage risk level data, matches the corresponding emergency plan from the preset emergency plan database, and generates emergency plan template matching data.

[0127] A risk assessment is conducted on the selected emergency response plan template, and the steps are adjusted according to the risk assessment results to match the template with the data, resulting in a matching emergency response plan.

[0128] In this embodiment of the invention, the risk decision instruction is a leakage risk level derived from real-time or historical data using a neural network model. This instruction contains relevant information about the leakage event, including the location of the leakage source, the gas diffusion path, meteorological conditions, and the leakage risk level. The leakage risk level data is extracted by parsing the risk decision instruction. Leakage risk levels typically include three levels: low, medium, and high, or more detailed classifications. Data can be parsed using natural language processing techniques or simple data extraction methods to extract leakage risk level information and generate leakage risk level data, i.e., the specific leakage risk level contained in the risk decision instruction. Based on the leakage risk level data, corresponding emergency response requirement data is extracted. Emergency response requirements are directly related to the leakage risk level: Low risk level: requires minimal intervention, such as monitoring or minor handling. Medium risk level: requires the deployment of more emergency resources, such as personnel, equipment, and material preparation. High risk level: requires immediate and full-scale response, including comprehensive evacuation, area lockdown, and gas detection. Using a preset response requirement template or rule engine, different risk levels are mapped to corresponding emergency response requirements. For example, in low-risk situations, response requirements only include local environmental monitoring and recording; in high-risk situations, response requirements include full evacuation, evacuation route planning, and air quality monitoring, generating emergency response requirement data that includes specific response tasks, resource requirements, and time requirements. The pre-set emergency plan library is a database containing emergency response plans for various risk levels, built based on the analysis results of historical leak events and simulated cases. The emergency plan library can be stored according to categories such as leak risk level, leak type, and environmental conditions. Based on the leak risk level data, a query operation is performed in the emergency plan library. Through the query interface, the most suitable emergency plan for that risk level is matched. If the risk level is low, a low-level emergency plan can be selected, involving relatively simple operations. If the risk level is high, the corresponding high-risk level emergency plan in the library is queried, involving more complex operations and requiring coordination of multiple resources. Through a set query algorithm or rule engine, the leak risk level data is matched with the plan templates in the emergency plan library. The matching process typically includes: filtering emergency plan templates based on information such as risk level, leak type, and meteorological conditions. The content of the matching emergency response plan needs to be further adjusted according to specific scenarios (such as chemical leaks, oil and gas leaks, etc.) to generate emergency response plan template matching data, that is, emergency response plan templates obtained from the emergency response plan database based on query results. A risk assessment is then conducted on the selected emergency response plan templates to ensure the effectiveness and adaptability of the plan. The assessment focuses on: whether the plan matches the risk level of the current leak event; considering the impact of environmental changes on the implementation of the plan, such as the impact of changes in meteorological conditions (wind speed, wind direction, etc.) on gas diffusion paths; and assessing whether the required resources (personnel, equipment, materials, etc.) are sufficient and whether there are potential resource shortage risks.Based on the risk assessment results, the emergency response plan is adjusted: if the assessment finds that certain steps in the plan are not suitable for the current situation, adjustments and optimizations are made. Based on the assessment feedback, modifications need to be made to the emergency response sequence, the deployment locations of response personnel, and the scheduling of response times. Simulation drills are conducted to verify the smooth execution of the emergency response plan, and the plan steps are further optimized based on the drill results to generate a matching emergency response plan. That is, based on the risk assessment and the adjusted emergency response plan, it is ensured to be applicable to the handling of the current leak incident. An actual emergency response is conducted according to the matching emergency response plan, executing all tasks in the plan. Specific tasks include: area containment and personnel evacuation based on the leak risk level and environmental conditions. Appropriate fire extinguishing and leak control equipment is selected based on the leak type. During the emergency response, the progress of the leak incident and environmental changes are monitored in real time, and the emergency plan is adjusted promptly to ensure the effectiveness of the emergency response. After the incident, a post-event assessment is conducted to check the implementation of the emergency response plan, summarize lessons learned, and improve the emergency response plan for future similar incidents. The emergency response is completed, and the results are reported back.

[0129] As an example of the present invention, reference is made to Figure 3 As shown, step S4 in this example includes:

[0130] Step S41: Initialize the configuration of the IoT gateway;

[0131] Step S42: Activate the explosion-proof mode by sending a set of timing execution instructions to the intelligent explosion-proof robot terminal through the configured IoT gateway, and collect execution process data of the intelligent explosion-proof robot to obtain robot execution data;

[0132] Step S43: Extract acoustic vibration features from the robot's execution data to obtain acoustic vibration feature data; construct a three-dimensional obstacle map based on the acoustic vibration feature data;

[0133] Step S44: Optimize the robot's execution data dynamically based on the 3D obstacle map to perform explosion-proof operations.

[0134] In this embodiment of the invention, the hardware system of the IoT gateway is ensured to operate normally, connect to the network, and have the ability to communicate with the intelligent explosion-proof robot terminal. The IoT gateway supports multiple protocols (such as MQTT, HTTP, CoAP, etc.) to transmit data. Network connections between the gateway and local or remote servers are configured, including configuring Wi-Fi or cellular network connections. Depending on the specific situation, data encryption and communication security policies can be selected to ensure data security during communication. Device authentication is performed on the IoT gateway to verify its identity and ensure connection security. The gateway is bound to the intelligent explosion-proof robot terminal, and corresponding device parameters are set, such as device ID, communication frequency, and data transmission protocol. After initialization configuration is completed, the gateway establishes communication with each device in the system (including the intelligent explosion-proof robot terminal), ensuring the stability and reliability of data transmission. Based on the explosion-proof operation requirements, a time-series execution instruction set is designed. This instruction set includes a series of operation tasks (such as movement, positioning, sensor scanning, explosion-proof operation initiation, etc.), and the execution sequence and conditions for each task are set. Each instruction in the instruction set includes operation content, parameter settings, task time, and task triggering conditions. The designed time-series execution instruction set is sent to the intelligent explosion-proof robot terminal through the IoT gateway. The gateway is responsible for transmitting instructions to the robot's terminal device, ensuring accurate transmission via wireless or wired means. Each task in the instruction set will be executed according to a predetermined sequence after the robot terminal receives the instruction, and the robot will activate the explosion-proof mode based on the instruction set. During explosion-proof operations, the robot will collect execution process data, including but not limited to: robot position and attitude data (via the built-in positioning system and inertial measurement unit IMU), various sensor data (such as temperature, gas concentration, vibration, etc.), and operation process status (such as whether the operation was successful, whether any abnormalities occurred, etc.). This data is transmitted in real time to the monitoring system or server via the IoT gateway for processing and analysis, generating robot execution data, including various robot operation statuses and sensor feedback data, providing a basis for subsequent analysis. Acoustic vibration data is extracted from the execution process data collected by the robot. This data is collected by vibration sensors or acoustic sensors equipped on the robot. The vibration signals monitored by these sensors during the robot's explosion-proof operations can reveal the presence of obstacles or the robot's interaction with the environment. The collected acoustic vibration data undergoes signal processing, including: denoising the data to remove environmental noise and sensor errors; converting the signal from the time domain to the frequency domain using methods such as Fourier Transform (FFT) or wavelet transform; extracting vibration characteristics across different frequency ranges; and calculating the amplitude, frequency, and other characteristics of the acoustic vibration signal to understand the vibration intensity and propagation mode. These processing results help identify obstacle characteristics (such as hard, soft, fixed, or moving obstacles). The extracted acoustic vibration feature data will be used to construct a 3D obstacle map and subsequent dynamic optimization commands.Using acoustic vibration feature data, algorithms (such as threshold-based detection and machine learning models) are employed to identify the presence and attributes of obstacles. Based on changes in vibration signals, the type, location, size, and shape of obstacles are determined. This acoustic vibration feature data is then transformed into a 3D obstacle map. This requires modeling the location and spatial distribution of each obstacle, typically using point cloud data or meshed data to represent obstacles in space. Through sensor feedback and the robot's positioning information, the obstacle map is updated in real time, forming a 3D spatial model of the robot's surroundings and generating a 3D obstacle map that displays the distribution of obstacles around the robot, providing data support for subsequent dynamic optimization commands. Based on the 3D obstacle map, the robot's current movement path and the positions of surrounding obstacles are analyzed. The optimal path is determined using path planning algorithms (such as Dijkstra's algorithm). It is then determined whether certain obstacles need to be avoided or a safer path should be chosen to perform the explosion-proof task. Based on real-time environmental information and the robot's execution status, the robot's operational commands are dynamically adjusted. This includes modifying the robot's speed, angle, or direction to ensure obstacle avoidance. The robot's explosion-proof operation procedures are adjusted to cope with changes in the on-site environment or unexpected events. The optimized instructions will be sent to the robot terminal through the IoT gateway to generate dynamically optimized execution instructions, ensuring that the robot can successfully complete the explosion-proof operation task while avoiding obstacles and dealing with potential dangers.

[0135] Of particular importance, step S44 also includes the following steps:

[0136] Step S441: Confirm the location of the obstacle based on the 3D obstacle map; calculate the obstacle blocking height based on the robot's execution data according to the obstacle location to obtain the obstacle blocking height data;

[0137] Step S442: Perform obstacle avoidance on the robot's execution data using the obstacle blocking height data to obtain obstacle avoidance path data and obstacle avoidance command;

[0138] Step S443: Use an obstacle avoidance optimization command to perform horizontal distance matching on the robot execution data and obstacle avoidance path data. When the horizontal distance matches, calculate the obstacle coverage area on the 3D obstacle map to obtain the obstacle coverage area data.

[0139] Step S444: Perform secondary obstacle avoidance on the robot's execution data based on the obstacle coverage area data to obtain secondary obstacle avoidance path optimization data and secondary obstacle avoidance instructions;

[0140] Step S445: Optimize the robot's execution data by performing steering based on the first obstacle avoidance command and the second obstacle avoidance command to obtain robot steering optimization data; optimize the robot's execution data by performing movement based on the first obstacle avoidance path data and the second obstacle avoidance path data to generate robot movement optimization data;

[0141] Step S446: Optimize the robot's dynamic pose command based on the robot's steering optimization data and robot movement optimization data to perform explosion-proof operations.

[0142] In this embodiment of the invention, the spatial location and shape of obstacles are identified by extracting the position coordinates of all obstacles from a 3D obstacle map. The precise location and 3D coordinates of obstacles are obtained using LiDAR or a stereo vision system, ensuring high accuracy of obstacle information. The locations of obstacles are marked on the map, and the volume, shape, and height of each obstacle are determined. The blocking height of each obstacle is calculated based on its location and shape, i.e., the height occupied by the obstacle on the robot's path. Based on the obstacle's geometry and mesh model, the height range of the obstacle in a specific direction is calculated. Through simulation modeling or actual measurement, the height of obstacles that the robot cannot cross is determined, preventing the robot from touching or being obstructed, generating obstacle blocking height data to provide data support for subsequent obstacle avoidance path planning. Based on the obstacle blocking height data, an obstacle avoidance calculation is performed on the robot's execution data to generate a preliminary path to avoid obstacles. In the path planning algorithm, classic path planning methods such as Dijkstra's algorithm are used to calculate the obstacle avoidance path based on the obstacle's location. The obstacle avoidance path should be the shortest possible while avoiding excessive proximity to obstacles. The calculated obstacle avoidance path generates an initial obstacle avoidance command, including the target direction, travel path, and necessary turning and acceleration / deceleration during obstacle avoidance. This command is transmitted to the robot's execution system, initiating the obstacle avoidance process and generating initial obstacle avoidance path data and the initial obstacle avoidance command, guiding the robot's next operation. Using the initial obstacle avoidance path data and robot execution data, the horizontal distance between the robot and the obstacle is matched. Sensors (such as LiDAR and ultrasonic sensors) measure the horizontal distance between the robot and the obstacle in real time, adjusting the robot's forward direction. The path planning optimizes the robot's route around the obstacle, ensuring obstacle avoidance accuracy. When the horizontal distance matches, the obstacle's coverage area is calculated to determine whether the robot can pass or needs to avoid it again. The obstacle's influence range is determined based on the intersection area of ​​the obstacle's actual footprint and the robot's path. This calculation considers whether the robot's path overlaps with the obstacle's coverage area to avoid inappropriate path selection, generating obstacle coverage area data to provide a basis for generating the next obstacle avoidance command. Based on the obstacle coverage area data, a second obstacle avoidance process is performed on the robot's execution data, generating a more optimized obstacle avoidance path. A more refined assessment of the space surrounding obstacles is performed to avoid misjudgments, and the obstacle avoidance route is replanned. The Bellman-Ford algorithm or fast travel algorithm is used for path optimization, making the obstacle avoidance path smoother and safer. Based on the optimized path, secondary obstacle avoidance commands are generated, including adjusted travel direction, speed, and detour strategy. These commands are transmitted to the robot wirelessly, triggering the robot's execution system to adjust and generate secondary obstacle avoidance path optimization data and secondary obstacle avoidance commands, enabling the robot to avoid obstacles more accurately. Based on the primary and secondary obstacle avoidance commands, the robot's steering is optimized.By combining path data and employing PID controllers or fuzzy control algorithms, the robot's steering angle and path smoothness are optimized. This ensures precise robot direction, avoiding obstacles while preventing inefficiencies caused by sharp turns. Based on primary and secondary obstacle avoidance path data, the robot's movement is optimized. The robot's speed and acceleration are adjusted to ensure smoothness and efficiency during obstacle avoidance. Velocity-position-time curves are used to optimize the robot's motion near obstacles, avoiding abrupt acceleration and deceleration, generating optimized robot steering and movement data to guide subsequent execution. Based on the optimized robot steering and movement data, dynamic pose commands are optimized for the robot's execution data. Through dynamic optimization algorithms, the robot's current position and attitude (including roll, pitch, and yaw angles) are adjusted in real time to ensure precise implementation of explosion-proof operations. Based on the robot's real-time position feedback and target position, the robot's attitude is adjusted to complete complex tasks, generating dynamic pose command optimization data to ensure the robot can perform tasks efficiently and safely during explosion-proof operations.

[0143] Preferably, step S41 includes the following steps:

[0144] Step S411: Configure the IoT gateway hardware parameters: the gateway power input range is 12V±5%DC, the gateway operating temperature range is set to -20℃ to +60℃, the gateway communication interface is configured to dual-band 2.4GHz and 5GHz, and the signal transmission rate is set to 150Mbps or 433Mbps.

[0145] Step S412: Configure the communication protocol for the IoT gateway: Set the data transmission protocol to MQTT and the message quality service level to QoS 1;

[0146] Step S413: Configure the network connection and bandwidth management of the IoT gateway: limit the gateway's bandwidth usage to 10Mbps.

[0147] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, 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 the equivalents of the application are intended to be included within the invention.

[0148] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for processing distributed multi-source heterogeneous sensor data, characterized in that, Applied to emergency scenarios involving toxic gas leaks in chemical industrial parks, and deployed on fixed multi-gas detectors, UAV-borne infrared spectral imaging devices, and intelligent explosion-proof robot terminals, the process includes the following steps: Step S1: Collect environmental parameters of the leak area in real time using a distributed sensor array; align the environmental parameters of the leak area with coordinate system 1 and timestamps to generate a standardized leak status dataset; Step S2: Extract gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion paths; generate risk decision instructions based on the gas cloud diffusion path probability using a preset leak level classification neural network, obtaining risk decision instructions. Step S2, which involves extracting gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion paths, includes: Gas concentration gradient analysis was performed on the standardized leakage situation dataset to generate gas diffusion gradient data. Thermal imaging spectral analysis is performed on the standardized leak situation dataset to extract heat distribution features and generate thermal imaging feature data. The thermal imaging spectral analysis of the standardized leak situation dataset includes: The standardized leakage situation dataset was imported into FLIR ResearchIR for spectral analysis parameter settings: the spectral range was set to 8μm-14μm, the frame rate to 60Hz, the temperature range to -20°C-650°C, the emissivity to 0.95, and the background temperature compensation to the ambient temperature. The spectral analysis settings data were obtained. The non-uniformity correction of the standardized leakage situation dataset was performed by setting the data through spectral analysis, and high-frequency noise was removed by using Gaussian 3×3 filtering to obtain spectral analysis filtered data. Hotspot analysis is performed on the spectral analytical filtering data based on a preset high-temperature region threshold to generate thermal imaging feature data; Spatiotemporal correlation analysis based on gas diffusion gradient data and thermal imaging feature data is performed to generate gas cloud diffusion trend data. This spatiotemporal correlation analysis includes: Spatial interpolation analysis is performed on gas diffusion gradient data to generate continuous gas concentration field data; Calculate the temperature gradient of the thermal imaging feature data to obtain the thermal convection effect data; Multidimensional spatiotemporal matching of continuous gas concentration field data and thermal convection influence data is performed to generate a gas diffusion-thermal imaging correlation matrix. Trend fitting is performed based on the gas diffusion-thermal imaging correlation matrix to generate gas cloud diffusion trend data; Probabilistic modeling is performed on the gas cloud diffusion trend data to calculate the probability of gas cloud diffusion paths under different wind speeds and environmental conditions, and to generate gas diffusion path probability data. Step S3: Invoke the matching emergency plan based on the risk decision instructions; analyze the implementation steps of the emergency plan and convert the instructions to generate emergency plan instructions; prioritize the emergency plan instructions and generate a time-sequenced execution instruction set; Step S4: Use the IoT gateway to distribute and dynamically optimize the timing execution instruction set to the intelligent explosion-proof robot terminal to perform explosion-proof operations.

2. The distributed multi-source heterogeneous sensor data processing method according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Simultaneously acquire concentration gradient data of hydrogen sulfide, chlorine, and benzene series compounds using a fixed multi-gas detector; scan the thermal radiation distribution of the leak source using an unmanned aerial vehicle (UAV)-borne infrared spectroscopy imaging device to obtain thermal radiation distribution data; integrate the concentration gradient data and thermal radiation distribution data using a distributed sensor array to obtain environmental parameters of the leak area. Step S12: Perform coordinate system unification processing on the environmental parameters of the leak area to generate unified leak environment data; Step S13: Timestamp align the unified data of the leaked environment to generate leaked environment aligned data; Step S14: Standardize the leak environment alignment data to generate a standardized leak situation dataset.

3. The distributed multi-source heterogeneous sensor data processing method according to claim 1, characterized in that, The construction process of the preset leakage level classification neural network in step S2 includes: Obtain meteorological data; Extract the location coordinates of the leak source from the standardized leak situation dataset to obtain the leak source location data; Meteorological data, leak source location data, and gas cloud diffusion path probability are merged to generate neural network input data; Feedforward calculations are performed on the input data of the neural network, and the probability distribution of leakage risk level is generated by calculating through a multilayer perceptron. The calculation results are used as the output of the model to obtain the risk probability data of leakage risk level. Post-processing of leakage risk level probability data generates final leakage risk level decision data; A leakage level classification neural network is constructed based on the neural network input data, leakage risk level probability data, and final leakage risk level decision data to obtain the preset leakage level classification neural network.

4. The distributed multi-source heterogeneous sensor data processing method according to claim 1, characterized in that, The emergency response plan that is invoked based on risk decision instructions in step S3 includes: Analyze the leakage risk level in the risk decision instruction to obtain leakage risk level data; Based on the leakage risk level data, emergency response requirements are extracted from the risk decision instructions to obtain emergency response requirement data. The system queries the leakage risk level data, matches the corresponding emergency plan from the preset emergency plan database, and generates emergency plan template matching data. A risk assessment is conducted on the selected emergency response plan template, and the steps are adjusted according to the risk assessment results to match the template with the data, resulting in a matching emergency response plan.

5. The distributed multi-source heterogeneous sensor data processing method according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Initialize the configuration of the IoT gateway; Step S42: Activate the explosion-proof mode by sending a set of timing execution instructions to the intelligent explosion-proof robot terminal through the configured IoT gateway, and collect execution process data of the intelligent explosion-proof robot to obtain robot execution data; Step S43: Extract acoustic vibration features from the robot's execution data to obtain acoustic vibration feature data; construct a three-dimensional obstacle map based on the acoustic vibration feature data; Step S44: Optimize the robot's execution data dynamically based on the 3D obstacle map to perform explosion-proof operations.

6. The distributed multi-source heterogeneous sensor data processing method according to claim 5, characterized in that, Step S41 includes the following steps: Step S411: Configure the IoT gateway hardware parameters: the gateway power input range is 12V±5%DC, the gateway operating temperature range is set to -20℃ to +60℃, the gateway communication interface is configured to dual-band 2.4GHz and 5GHz, and the signal transmission rate is set to 150Mbps or 433Mbps. Step S412: Configure the communication protocol for the IoT gateway: Set the data transmission protocol to MQTT and the message quality service level to QoS1; Step S413: Configure the network connection and bandwidth management of the IoT gateway: limit the gateway's bandwidth usage to 10Mbps.

7. A distributed multi-source heterogeneous sensor data processing system, characterized in that, For performing the distributed multi-source heterogeneous sensor data processing method as described in claim 1, the distributed multi-source heterogeneous sensor data processing system includes: The data acquisition module is used to collect environmental parameters of the leak area in real time through a distributed sensor array; the environmental parameters of the leak area are aligned with coordinate system and timestamp to generate a standardized leak status dataset; The risk decision module is used to extract gas diffusion features and thermal imaging features from the standardized leak situation dataset to calculate the probability of gas cloud diffusion path; based on the preset leak level classification neural network, risk decision instructions are generated for the probability of gas cloud diffusion path to obtain risk decision instructions. The instruction generation module is used to invoke the matching emergency plan based on the risk decision instruction; analyze the implementation steps of the emergency plan and convert them into instructions to generate emergency plan instructions; prioritize the emergency plan instructions and generate a time-sequenced execution instruction set. The instruction distribution module is used to distribute and dynamically optimize the timing execution instruction set to the intelligent explosion-proof robot terminal using the IoT gateway to perform explosion-proof operations.

Images