Chemical fire safety situation edge computing communication and accident tracing system

Abstract

This invention relates to the field of edge computing technology for chemical safety monitoring, specifically a chemical fire safety situation edge computing communication and accident tracing system, comprising: a central analysis platform and multiple edge computing nodes within a chemical industrial park. Each edge computing node deploys an environmental sensing sensor network to collect signals of chemical concentration, temperature, pressure, and flame characteristics. After synchronous processing, these signals generate structured sensing data blocks, independently complete local risk assessments, and generate edge risk reports, which are then uploaded to the central analysis platform via a time-sensitive network communication protocol. The central platform integrates all edge risk reports and, combined with information on the park's equipment layout and process flow, conducts a global safety situation fusion analysis to generate a global risk map. When an abnormal alarm is detected, an accident tracing report is generated based on this map. This invention can reduce network load, shorten risk response time, ensure stable data transmission, and achieve overall safety control and accurate accident tracing within the industrial park.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology for chemical safety monitoring, and in particular to an edge computing communication and accident tracing system for chemical fire safety situation. Background Technology

[0002] Currently, fire safety monitoring and risk assessment in chemical industrial parks primarily employ a centralized cloud processing model. Environmental sensors collect various monitoring signals, such as chemical concentrations, temperature, pressure, and flame characteristics, which are directly uploaded to cloud servers. The cloud then handles data analysis, risk assessment, and anomaly signal identification. Data transmission largely utilizes conventional network communication protocols, and accident tracing relies solely on fragmented monitoring data stored in the cloud, lacking systematic risk correlation analysis support. Under this processing model, the continuous uploading of massive amounts of raw sensing data consumes significant network bandwidth, easily leading to data transmission congestion and delays. Risk assessment in localized areas depends on cloud processing feedback, preventing real-time local response and resulting in a significant lag in risk identification.

[0003] Conventional network communication protocols are insufficient to meet the low-latency, high-reliability transmission requirements of chemical safety monitoring data, and cannot guarantee the stability of data interaction between the edge and the cloud. Meanwhile, cloud processing of uploaded data is mostly limited to simple aggregation and statistics, failing to integrate and analyze static information on equipment layout within the chemical industrial park with dynamic information on the process flow, making it difficult to form a comprehensive and accurate global safety situation awareness. Furthermore, existing technologies at the edge cannot synchronously process multi-source sensing signals and independently complete local safety risk assessments, cannot generate structured regional real-time sensing data blocks and edge risk reports, and lack coherent global risk data support for accident tracing, making it impossible to fully reconstruct the risk correlation process of abnormal events. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a chemical fire safety situation edge computing communication and accident tracing system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a chemical fire safety situation edge computing communication and accident tracing system, comprising: An environmental sensing sensor network covering its physical area is deployed for each edge computing node. The environmental sensing sensor network continuously collects chemical concentration signals, temperature signals, pressure signals, and flame characteristic signals of its physical area. Each edge computing node synchronously processes multiple signals from the connected environmental sensing sensor network to generate structured regional real-time sensing data blocks. Each edge computing node independently performs a preliminary local security risk assessment based on structured regional real-time perception data blocks, identifying potential risk points and risk types; Each edge computing node will package the assessed local risk points and risk types, along with key summary information from the structured regional real-time perception data blocks, into an edge risk report; The edge risk reports generated by each edge computing node are periodically uploaded to the central analysis platform located in the cloud via a time-sensitive network communication protocol. The central analysis platform receives and integrates edge risk reports from all edge computing nodes, combines static information on equipment layout and dynamic information on process flow in the chemical industrial park, performs global security situation fusion analysis, and calculates a global risk map. Upon detecting any form of abnormal alarm signal, an incident tracing report is generated based on the global risk map.

[0006] As a further aspect of the present invention, the environmental sensing sensor network continuously collects chemical concentration signals, temperature signals, pressure signals, and flame characteristic signals of the physical area to which it belongs, specifically including: The environmental sensing sensor network includes a variety of distributed sensor terminals, including gas concentration sensors, infrared thermal imagers, pressure transmitters, and ultraviolet flame detectors. The gas concentration sensor measures the concentration of a variety of key hazardous chemicals in the air at fixed sampling intervals and generates a chemical concentration signal that includes sensor identification, timestamp, and concentration value. An infrared thermal imager continuously scans a designated area to acquire a sequence of infrared images that reflect the temperature field distribution on the surface of the equipment and in space. The temperature value of a specific monitoring point is extracted from the infrared image sequence to generate a temperature signal. Pressure transmitters are installed on critical pipelines and reaction vessels to monitor internal pressure changes in real time and output pressure signals containing pressure readings and time information. The ultraviolet flame detector monitors the status of open flames within its field of view. When it detects ultraviolet light of a specific wavelength, it generates a flame characteristic signal that includes a fire confirmation mark and the detection time.

[0007] As a further aspect of the present invention, each edge computing node synchronously processes multiple signals from the connected environmental sensing sensor network to generate structured regional real-time sensing data blocks, specifically including: The processing includes threshold comparison of abnormal signals, signal drift calibration, and fusion and alignment of multi-source signals on timestamps; Edge computing nodes allocate an independent buffer queue for each type of signal received and record its original timestamp of arrival; Edge computing nodes initiate clock synchronization services, using the unified time signal issued by the central analysis platform as a reference, to calibrate the original timestamps of all signals and convert them to a unified time reference. For signals after the calibration timestamp, outlier detection is performed, signals that exceed the physical range of the sensor are marked as invalid, and signal spikes caused by transient interference are smoothed and filtered. Extract concentration change rate features from chemical concentration signals, temperature rise rate features from temperature signals, and pressure gradient features from pressure signals; All types of processed and time-calibrated signals are aligned and fused according to a unified time reference. Multiple signal feature values, signal quality identifiers, and sensor location codes within the same time window are packaged to form a structured regional real-time sensing data block.

[0008] As a further aspect of the present invention, each edge computing node independently performs a preliminary local security risk assessment based on structured regional real-time perception data blocks, identifying potential risk points and risk types, specifically including: The edge computing node stores multiple chemical concentration thresholds, temperature thresholds, and pressure thresholds preset based on regional attributes, and these thresholds are associated with risk levels; Read the chemical concentration characteristics, temperature characteristics, and pressure characteristics of the current time window from the structured regional real-time sensing data block; The chemical concentration characteristic values ​​read are compared with the stored chemical concentration thresholds one by one. Chemicals that exceed the thresholds are identified and their degree of exceedance is recorded as leakage risk points and risk levels. The read temperature feature values ​​are compared with the stored temperature thresholds. Monitoring points that exceed the thresholds are identified and their degree of exceeding the limits are recorded as over-temperature risk points and risk levels. The read pressure characteristic values ​​are compared with the stored pressure thresholds. Monitoring points that exceed the thresholds are identified and their degree of exceedance is recorded as overpressure risk points and risk levels.

[0009] As a further aspect of the present invention, each edge computing node packages the assessed local risk points and risk types, along with key summary information from the structured regional real-time perception data block, into an edge risk report, specifically including: Edge computing nodes aggregate all risk points identified within the same assessment period, generating a risk entry for each risk point that includes location code, risk type, risk level, first occurrence time, and duration. In the process of generating structured regional real-time sensing data blocks, feature statistics of various signals within the evaluation period are extracted, including maximum, minimum, average and variance, as key summary information; The current assessment period's timeframe, the unique identifier of this edge computing node, all summarized risk items, and extracted key summary information are organized according to a preset binary encoding format. Add a protocol header before the organized data content. The protocol header contains the report type, data length, sequence number and check code information to complete the data packet encapsulation of the edge risk report. The edge computing node stores the encapsulated edge risk report in the transmission buffer, waiting to be transmitted to the central analysis platform.

[0010] As a further aspect of the present invention, the step of periodically uploading the edge risk reports generated by each edge computing node to the central analysis platform located in the cloud via a time-sensitive network communication protocol specifically includes: The central analysis platform plans communication time slots for all edge computing nodes in the park and allocates different transmission priorities and bandwidth guarantees for data with different risk levels. When each edge computing node arrives at its assigned communication time slot, it checks the transmission buffer for any edge risk reports to be sent. If there are edge risk reports to be sent, the edge computing node loads the edge risk report data packets into the protocol data unit according to the frame format of the time-sensitive network communication protocol; Within the protocol data unit, a corresponding quality of service label and transmission priority identifier are set based on the highest risk level of the risk items contained in the report; Edge computing nodes will send protocol data units carrying quality of service tags and transmission priority identifiers to the uplink network link leading to the central analysis platform via industrial Ethernet switches.

[0011] As a further aspect of the present invention, the central analysis platform receives and integrates edge risk reports from all edge computing nodes, combines the static information of equipment layout and the dynamic information of process flow in the chemical industrial park, performs a global security situation fusion analysis, and calculates a global risk map, specifically including: The central analysis platform parses each received edge risk report and extracts the unique identifier of the edge computing node, risk items, and key summary information. The extracted risk items are mapped to the corresponding coordinates on the stored 3D digital map of the chemical industrial park according to their location codes. The operating status of the equipment around the risk point, the types of materials involved, and the current process operation are obtained from the dynamic information database of the process flow. A comprehensive analysis is conducted on the geographical location, risk type, risk level, and dynamic information of surrounding processes of the risk points to assess the spread trend of the risk points, the equipment affected, and the chain risks to adjacent areas. All risk points identified in the assessment, their current status, scope of impact, and cascading risk relationships are marked on the 3D digital map of the chemical industrial park using an overlay method, forming a dynamic and visualized global risk map.

[0012] As a further aspect of the present invention, the step of generating an accident tracing report based on the global risk map when any form of abnormal alarm signal is detected includes: The central analysis platform immediately triggered the accident tracing mode and issued a high-density data collection command to the edge computing node that issued the alarm signal; The edge computing node that receives the high-density data acquisition command instructs its subordinate environmental perception sensor network to switch to high-speed sampling mode and encapsulates the acquired raw signal stream with the complete timestamp to form an accident backtracking data packet; The edge computing nodes upload the accident backtracking data packets to the central analysis platform. The central analysis platform uses the global risk map and the real-time uploaded accident backtracking data packets to reconstruct the complete event sequence from before the accident to after the accident. The central analysis platform will compare the reconstructed complete event sequence with typical patterns in the pre-set chemical accident knowledge base to determine the direct and indirect cause chains of the accident, and finally generate a standardized accident tracing report. The central analysis platform immediately triggers the incident tracing mode, issuing high-density data collection commands to the edge computing nodes corresponding to the alarm signals, specifically including: An abnormal alarm signal originates from any of the following: a risk point determined by the central analysis platform to have reached a catastrophic risk level in the global security situation fusion analysis; or, an edge risk report uploaded by the edge computing node contains the highest level risk item that requires immediate response; or, an emergency alarm manually triggered by a person through the monitoring system. Once an abnormal alarm signal is detected, the central analysis platform immediately analyzes the abnormal alarm signal and locates the unique identifier of the edge computing node that issued the alarm and its associated physical area; The central analysis platform generates high-density data acquisition instructions, which include the unique identifier of the target edge computing node, the start time of high-density sampling, the sampling duration, and the required sensor signal type and sampling frequency. The central analysis platform sends high-density data acquisition commands to the target edge computing nodes through an independent control channel with the highest network priority; The central analysis platform also creates an incident tracing task container locally, named after the incident alarm time and location, to receive subsequent incident backtracking data.

[0013] As a further aspect of the present invention, the edge computing node receiving the high-density data acquisition command instructs its subordinate environmental perception sensor network to switch to high-speed sampling mode, and encapsulates the acquired raw signal stream with a complete timestamp to form an accident backtracking data packet, specifically including: Edge computing nodes receive and parse high-density data acquisition instructions, obtain sampling parameters, and send sampling mode switching commands to all connected sensor terminals of relevant types; In an environmental sensing sensor network, the sensor terminal immediately switches from the conventional sampling mode to the high-speed sampling mode upon receiving a command, and acquires the raw signal at the high frequency specified by the instruction. The edge computing node adds a local timestamp to each piece of raw signal data received in high-speed sampling mode and caches it. After the sampling duration specified in the instruction ends, the edge computing node commands the environmental awareness sensor network to switch back to the normal sampling mode. Edge computing nodes compress and encapsulate the cached high-frequency raw signal streams and their corresponding local timestamps in chronological order, and attach an incident tracing identifier to the header of the data packet to form a complete incident tracing data packet.

[0014] As a further aspect of the present invention, the central analysis platform utilizes a global risk map, combined with real-time uploaded accident retrospective data packets, to reverse-reconstruct the complete event sequence from before the accident to after the accident, specifically including: The central analysis platform decompresses and parses the received accident backtracking data packets to recover the high-frequency original signal stream and its timestamps; From the global risk map, historical risk evolution data centered on alarm points and traced back a certain period in time are extracted as background information for accident precursors. The high-frequency signal stream in the accident retrospective data packet is matched and aligned with the background information of the accident precursors on the time axis. Taking the mutation points in high-frequency signal streams as the core, and combining them with background information on accident precursors, we analyze and infer the equipment operation, material changes, or environmental condition changes that cause each signal mutation in reverse chronological order. The series of events deduced through analysis are arranged in logical and chronological order to form a complete event sequence from the initial anomaly to the outbreak of the accident. Each event in the sequence is associated with a specific time point, the equipment or materials involved, and signal evidence.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Edge computing nodes synchronously process chemical concentration, temperature, pressure, and flame characteristic signals collected by the environmental sensing sensor network, generating structured real-time regional sensing data blocks. Based on these data blocks, they independently complete local safety risk assessments, identifying potential risk points and types, and packaging them into an edge risk report. This structured data processing method allows for the orderly integration of multi-source sensing signals, avoiding the disorganization of scattered signals, reducing the volume of raw sensing data transmission, and mitigating the excessive network load caused by directly uploading large amounts of raw data. Local risk assessment is completed locally at the edge, eliminating the need to wait for cloud processing feedback, shortening the response time for risk identification, and ensuring that the risk assessment of local areas more closely matches the real-time sensing status of the corresponding physical area, reducing the lag in risk identification.

[0016] Edge computing nodes periodically upload edge risk reports to the cloud-based central analysis platform via a time-sensitive network communication protocol. This protocol ensures the real-time performance and stability of data interaction between the edge and the cloud, avoiding transmission delays and congestion issues common in conventional communication protocols. The central analysis platform integrates edge risk reports from all edge nodes, combining static information on equipment layout and dynamic information on process flows within the chemical industrial park to conduct a global safety situation fusion analysis. This generates a global risk map, enabling comprehensive control over the park's safety status. This overcomes the limitations of simply aggregating data in the cloud, making global safety analysis more closely aligned with the actual production and operation scenarios of the park. When an abnormal alarm signal is triggered, an accident tracing report is generated based on the global risk map. This report integrates risk data from various regions, forming coherent tracing content and fully reconstructing the risk correlation process corresponding to the abnormal event, thus addressing the lack of systematic risk correlation support in existing accident tracing technologies. Attached Figure Description

[0017] Figure 1 This is a flowchart of the chemical fire safety situation edge computing communication and accident tracing system described in this invention; Figure 2 A flowchart for signal acquisition in an environmental sensing sensor network; Figure 3 A flowchart for a local security risk assessment; Figure 4 Thermal diagram of multi-parameter anomalies for ECU-C-05 node; Figure 5 This is a comparison chart of hydrogen concentration change trends and warning thresholds. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 The system architecture includes a central analysis platform located in the cloud and multiple edge computing nodes selected within the chemical industrial park based on regional divisions and risk levels. Each edge computing node is equipped with an environmental sensing sensor network covering its assigned physical area. This network operates continuously, collecting signals such as chemical concentration, temperature, pressure, and flame characteristics. Each edge computing node synchronously processes the various signals received from its sensor network, generating structured real-time regional sensing data blocks. Based on these local structured real-time regional sensing data blocks, each edge computing node independently runs a risk assessment algorithm to perform a preliminary local safety risk assessment, identifying potential risk points and determining their risk types. After the assessment, each edge computing node packages the identified local risk points and risk types, along with key summary information extracted from the structured real-time regional sensing data blocks, into a standardized edge risk report. Each node then uploads the generated edge risk report to the central analysis platform in the cloud via a time-sensitive network communication protocol at preset intervals. The central analysis platform receives and integrates reports from all edge computing nodes, combining static information on equipment layout and dynamic information on process flow within the chemical industrial park to perform a global safety situation fusion analysis, calculating and updating the global risk map. When the system detects any form of abnormal alarm signal, it immediately initiates the analysis process based on the latest global risk map, generating a detailed incident tracing report.

[0021] See Figure 2In one embodiment of the present invention, the environmental sensing sensor network includes a variety of distributed sensor terminals, including gas concentration sensors, infrared thermal imagers, pressure transmitters, and ultraviolet flame detectors. In the tank area of ​​a chemical industrial park, the gas concentration sensor measures the concentration of hydrogen and chlorine in the air at fixed sampling intervals for preset key hazardous chemicals. The generated chemical concentration signal includes a sensor identifier, a timestamp, and a concentration value. The sampling interval of the gas concentration sensor is one second, and the output signal format is a triplet. In some embodiments, the infrared thermal imager continuously scans a designated area on the surface of the tank to acquire an infrared image sequence reflecting the temperature field distribution. The frame rate of the infrared image sequence is ten frames per second. The temperature value of a specific monitoring point on the surface of the tank is extracted from the infrared image sequence to generate a temperature signal, which includes the coordinates of the monitoring point and the temperature value. The pressure transmitter is installed on a key pipeline of the tank to monitor internal pressure changes in real time and outputs a pressure signal containing pressure readings and time information. The output frequency of the pressure signal is five times per second. The ultraviolet flame detector monitors the field of view of the storage tank area. When it detects ultraviolet light of a specific wavelength, it generates a flame characteristic signal that includes a fire confirmation mark and the detection time. The flame characteristic signal uses a Boolean value to represent the fire status.

[0022] Each edge computing node synchronously processes multiple signals from the connected environmental sensing sensor network. This processing includes anomaly threshold comparison, signal drift calibration, and fusion and alignment of multi-source signals on timestamps. Each edge computing node allocates an independent buffer queue for each type of received signal; gas concentration signals, temperature signals, pressure signals, and flame characteristic signals are stored in different queues. The edge computing node records the original timestamp of each signal's arrival. The edge computing node initiates a clock synchronization service, using the unified time signal issued by the central analysis platform as a reference, calibrating the original timestamps of all signals and converting them to a unified time base (Coordinated Universal Time). Anomaly detection is performed on the calibrated signals. Signals exceeding the sensor's physical range are marked as invalid. For gas concentration sensors with a range of 0 to 100 ppm, concentration values ​​exceeding this range are marked as invalid. Signal spikes caused by transient interference are smoothed using a filtering process. Optionally, a moving average filtering method is used, expressed by the formula: in: This represents the smoothed signal value. Represents the original signal value. This indicates the size of the moving average window. Concentration change rate features are extracted from chemical concentration signals, calculated as the ratio of the concentration difference between adjacent sampling points to the time interval. Temperature rise rate features are extracted from temperature signals, calculated as the ratio of the temperature difference between adjacent sampling points to the time interval. Pressure gradient features are extracted from pressure signals, calculated as the ratio of the pressure difference between adjacent sampling points to the time interval. Essentially, all types of processed and time-calibrated signals are aligned and fused according to a unified time reference. The alignment operation matches signal feature values ​​to the same time window at the smallest time granularity. Multiple signal feature values, signal quality identifiers, and sensor location codes within the same time window are packaged to form a structured regional real-time sensing data block, which is stored in binary format.

[0023] In some embodiments, the concentration signal from the gas concentration sensor has a millisecond-level deviation before time calibration. After calibration, all signal timestamps are aligned to Coordinated Universal Time (UTC), ensuring that data collected at the same time have a consistent time reference. When edge computing nodes perform outlier detection on signals, the temperature signal from the infrared thermal imager experiences instantaneous spikes due to environmental thermal radiation interference. Smoothing and filtering processes eliminate these spikes, preserving the true temperature trend. In a specific implementation, the fusion and alignment of multi-source signals on timestamps uses a 100-millisecond time window. Feature values ​​from all signals within the window are aggregated to generate a structured regional real-time sensing data block. This structured data block includes the start and end times of the time window and the sensor data array. Optionally, signal drift calibration employs a linear compensation method, calculating a correction based on the deviation between the unified timing signal and the sensor's local clock, and applying the correction to all original timestamps. When extracting pressure gradient features from pressure signals, these features reflect the rate of pressure change and are used for subsequent risk assessment. It is understood that flame characteristic signals do not require rate-of-change feature extraction and can directly participate in fusion and alignment as Boolean values. After extracting the concentration change rate feature from the chemical concentration signal of the gas concentration sensor, the concentration change rate feature and the original concentration value are stored together in a structured regional real-time sensing data block.

[0024] See Figure 3In one embodiment of the present invention, each edge computing node independently performs a preliminary local security risk assessment based on structured regional real-time sensing data blocks. Taking an edge computing node located in the alkylation reactor area as an example, the edge computing node stores various chemical concentration thresholds, temperature thresholds, and pressure thresholds preset based on the regional attributes of the alkylation reactor. The primary warning threshold for hydrogen concentration is 10 ppm, and the secondary alarm threshold is 25 ppm. The primary warning threshold for reactor outer wall temperature is 150°C, and the secondary alarm threshold is 180°C. The primary warning threshold for reactor internal pressure is 1.5 MPa, and the secondary alarm threshold is 2.0 MPa. These thresholds are associated with risk levels. During the assessment, the edge computing node reads the chemical concentration characteristics, temperature characteristics, and pressure characteristics for the current time window from the structured regional real-time sensing data blocks. The hydrogen concentration characteristic value for the current time window is 30 ppm, the reactor outer wall temperature characteristic value is 170°C, and the reactor internal pressure characteristic value is 1.8 MPa. The edge computing node compares the read hydrogen concentration characteristic value of 30 ppm with the stored chemical concentration thresholds item by item. The hydrogen concentration characteristic value of 30 ppm exceeds the first-level warning threshold of 10 ppm and the second-level alarm threshold of 25 ppm. The hydrogen is identified and its exceedance level is recorded as a leak-related risk point and the corresponding second-level risk level. The edge computing node compares the read reactor outer wall temperature characteristic value of 170℃ with the stored temperature threshold. The temperature characteristic value of 170℃ exceeds the first-level warning threshold of 150℃ but does not exceed the second-level alarm threshold of 180℃. The reactor outer wall monitoring point is identified and its exceedance level is recorded as an over-temperature risk point and the corresponding first-level risk level. The edge computing node compares the read reactor internal pressure characteristic value of 1.8 MPa with the stored pressure threshold. The pressure characteristic value of 1.8 MPa exceeds the first-level warning threshold of 1.5 MPa but does not exceed the second-level alarm threshold of 2.0 MPa. The reactor internal pressure monitoring point is identified and its exceedance level is recorded as an over-pressure risk point and the corresponding first-level risk level.

[0025] In some embodiments, the risk assessment process involves a comprehensive judgment of multiple characteristic values, and the correlation between thresholds stored within the edge computing node and risk levels can be represented by a mapping table. Optionally, the calculation of the risk level can introduce a comprehensive scoring formula, expressed as: in: This represents the overall risk score. Indicates the percentage of concentration exceeding the limit. Indicates the percentage of temperature exceeding the limit. Indicates the percentage of pressure exceeding the limit. , , These are weighting coefficients set for concentration, temperature, and pressure, respectively. Edge computing nodes are based on a comprehensive risk score. The numerical range determines the final risk level. Edge computing nodes package the assessed local risk points and risk types, along with key summary information from structured regional real-time sensing data blocks, into an edge risk report. Edge computing nodes summarize all risk points identified within the same assessment period, generating a risk entry for each risk point containing location code, risk type, risk level, first occurrence time, and duration. During the generation of structured regional real-time sensing data blocks, edge computing nodes extract characteristic statistical values ​​of various signals within the assessment period, including the hydrogen concentration (maximum 35 ppm, minimum 28 ppm, average 31 ppm, variance 2.1), reactor outer wall temperature (maximum 172℃, minimum 168℃, average 170℃, variance 1.0), and reactor internal pressure (maximum 1.82 MPa, minimum 1.78 MPa, average 1.80 MPa, variance 0.01). These characteristic statistical values ​​serve as key summary information.

[0026] In practice, edge computing nodes organize the current assessment period's time range, their unique identifier, all summarized risk entries, and extracted key summary information according to a preset binary encoding format. Risk entries use fixed-length field encoding: location code occupies 4 bytes, risk type occupies 1 byte, risk level occupies 1 byte, and time information occupies 8 bytes each. A protocol header is added before the organized data content, containing report type, data length, sequence number, and checksum information, completing the data packet encapsulation of the edge risk report. The edge computing node stores the encapsulated edge risk report in a transmission buffer, awaiting transmission to the central analysis platform.

[0027] It is understood that the generation of risk entries is based on real-time analysis of structured regional real-time sensing data blocks, and the process from the identification of each risk point to the generation of an entry is continuous. In some embodiments, for monitoring points that do not exceed a threshold, the edge computing node will not generate risk entries for them, but the characteristic statistical values ​​of the relevant signals will still be extracted and packaged as part of the key summary information. Optionally, the binary encoding format adopts a TLV structure, with each data segment consisting of three parts: type, length, and value, to ensure the flexibility and scalability of data parsing. The variance value extracted from the key summary information reflects the fluctuation of the signal data within the evaluation period, and the magnitude of the variance value is related to the stability of the signal. The unique identifier of the edge computing node is written to a fixed starting position in the data packet when organizing the data, which is used by the central analysis platform to quickly identify the data source. The checksum information in the protocol header is calculated and generated using a cyclic redundancy check algorithm, which is used to verify the data integrity of the edge risk report during data transmission. It is understood that the sending buffer serves as a temporary storage area for temporarily storing the encapsulated edge risk report data packets waiting for network transmission, and the capacity configuration of the sending buffer needs to consider the maximum report size and upload cycle.

[0028] In one embodiment of the present invention, edge risk reports generated by each edge computing node are periodically uploaded to a central analysis platform located in the cloud via a time-sensitive network communication protocol. Taking a chemical plant area as an example, the central analysis platform plans communication time slots for all edge computing nodes in the park. The communication time slot for edge computing node A is the first 10-millisecond window every five minutes, and the communication time slot for edge computing node B is the second 10-millisecond window every five minutes. The central analysis platform allocates different transmission priorities and bandwidth guarantees for data with different risk levels. Level 2 risk level reports are assigned the highest priority identifier 0x07 and a guaranteed bandwidth of 10Mbps, Level 1 risk level reports are assigned the medium priority identifier 0x05 and a guaranteed bandwidth of 5Mbps, and no-risk reports are assigned the low priority identifier 0x01 and a guaranteed bandwidth of 1Mbps. Each edge computing node checks whether there is an edge risk report to be sent in its transmission buffer when its allocated communication time slot arrives. Edge computing node A checks its transmission buffer at 10:00:00 on March 5, 2026 and finds an edge risk report to be sent. If an edge risk report to be sent exists, edge computing node A loads the edge risk report data packet into a protocol data unit according to the frame format of the Time-Sensitive Networking Protocol (TSP). The protocol data unit includes a preamble, start-of-frame delimiter, destination MAC address, source MAC address, VLAN tag, Ethernet type, payload, and frame check sequence. Within the protocol data unit, a corresponding Quality of Service (QoS) tag and transmission priority identifier are set based on the highest risk level of the risk entry contained in the report. Since the report to be sent by edge computing node A contains a risk entry at a level 2 risk, edge computing node A sets the QoS tag value to 0x07 in the VLAN tag field of the protocol data unit and sets the transmission priority identifier field to 0x07 in the payload header. Edge computing node A then sends the protocol data unit carrying the QoS tag and transmission priority identifier to the uplink network link leading to the central analysis platform via an industrial Ethernet switch. The industrial Ethernet switch performs priority queuing and forwarding of the data frame based on the QoS tag value in the VLAN tag.

[0029] The central analysis platform receives and integrates edge risk reports from all edge computing nodes, combining static information on equipment layout and dynamic information on process flow within the chemical industrial park to conduct a global security situation fusion analysis. In practice, the central analysis platform parses each received edge risk report, extracting the unique identifier of the edge computing node, risk entries, and key summary information. For example, from the report of edge computing node A, the node identifier "ECU-A-01" is extracted, the risk entry is "Location P-102, hydrogen leak, Level 2 risk, first occurrence at 10:00:00," and the key summary information includes statistical values ​​for hydrogen concentration. The central analysis platform maps the extracted risk items to the corresponding coordinates on the stored 3D digital map of the chemical industrial park according to their location codes. The location code "P-102" corresponds to the coordinates (X=150.3, Y=89.7, Z=5.2) on the 3D digital map. The central analysis platform obtains the operating status of the equipment around the risk point, the types of materials involved, and the current process operation from the process flow dynamic information database. It finds that there is a hydrogen compressor and a hydrogen pipeline within three meters of the coordinates (X=150.3, Y=89.7, Z=5.2). The compressor is currently running, the material in the pipeline is hydrogen, and the current process operation is called "hydrogen pressurization and transportation". The central analysis platform comprehensively analyzes the geographical location, risk type, risk level, and surrounding process dynamics of the risk point. It assesses the risk's spread trend, affected equipment, and cascading risks to adjacent areas. The analysis concludes that the hydrogen leak is located at the compressor outlet pipe flange. The leaking hydrogen, being less dense than air, may diffuse upwards, affecting the electrical control cabinet three meters above. Simultaneously, the pressure in the nearby hydrogen pipeline is 2.5 MPa, posing a risk of a jet fire due to the leak's expansion. A quantitative impact model can be used to assess the cascading risks, expressed by the following formula: in: Indicates the comprehensive impact index. This indicates the risk level value of the main risk point. This represents the risk level value of the j-th cascade risk point. and These are weighting coefficients determined based on distance and process correlation. This represents the number of cascading risk points. The central analysis platform will overlay all assessed risk points, their current status, impact range, and cascading risk relationships onto a 3D digital map of the chemical industrial park using layers. Hydrogen leak points are marked with a flashing red icon, their impact range is represented by a semi-transparent red sphere, affected electrical control cabinets are marked with a yellow warning icon, and cascading jet fire risk areas are marked with an orange dashed line area. These markings together form a dynamic and visualized global risk map.

[0030] Optionally, during the edge risk report upload process, if multiple reports are to be sent in the sending buffer of an edge computing node, the node sorts them according to the highest risk level of the risk entries in the reports, and higher-level reports are packaged and sent first. When performing global security situation fusion analysis, the central analysis platform performs time alignment on reports from different edge computing nodes to ensure all data is on the same analysis time base. The equipment operating status information obtained from the process flow dynamic information database is updated once per second to ensure real-time analysis. It can be understood that the dynamic visualization update frequency of the global risk map is consistent with the cycle of the central analysis platform receiving and processing edge risk reports, for example, updating the risk status and impact range on the map every five minutes. In some embodiments, the risk point icon style and color on the 3D digital map have a strict mapping relationship according to the risk type and risk level; for example, red represents leakage, orange represents overheating, and purple represents overpressure, and the icon flashing frequency is positively correlated with the risk level. Optionally, when assessing risk diffusion trends, the central analysis platform will invoke its built-in gas diffusion model, combined with real-time wind direction and speed data, to predict the movement path and concentration distribution of hazardous gas clouds over a future period. The prediction results will then be overlaid onto the global risk map as dynamic isosurfaces. The process of parsing edge risk reports by the central analysis platform includes verifying the checksum information in the protocol header; reports that fail verification will be discarded and logged.

[0031] In one embodiment of the present invention, the central analysis platform immediately triggers the accident tracing mode upon detecting any form of abnormal alarm signal. The abnormal alarm signal originates from any of the following: the central analysis platform determines a risk point that reaches a catastrophic risk level in the global security situation fusion analysis, for example, the global risk map shows that the concentration of ethylene oxide at a certain location rises sharply from 5 ppm to 80 ppm within 30 seconds, exceeding 50% of the lower explosive limit, and the central analysis platform determines it as a catastrophic leak risk and generates an abnormal alarm signal; or, the edge risk report uploaded by the edge computing node contains the highest level risk item that requires immediate response, for example, the report uploaded by edge computing node C contains a risk item showing "Location P-205, chlorine leak, Level 3 risk (highest level)", and the central analysis platform generates an abnormal alarm signal after parsing the report; or, an emergency alarm manually triggered by a person through the monitoring system, for example, the monitor discovers unidentified smoke in the video footage and manually clicks the emergency alarm button on the control panel to generate a manually triggered abnormal alarm signal. Once an abnormal alarm signal is detected, the central analysis platform immediately analyzes the abnormal alarm signal, locates the unique identifier of the edge computing node that issued the alarm and its associated physical area, analyzes the alarm signal generated by the report from edge computing node C, and locates the edge computing node with the unique identifier "ECU-C-05", whose associated physical area is "Hydrogen Purification Unit Area B".

[0032] The central analysis platform generates high-density data acquisition instructions. These instructions include the unique identifier of the target edge computing node, the start time of high-density sampling, the sampling duration, and the required sensor signal types and sampling frequency. In specific implementation, the central analysis platform interprets the alarm time as 14:25:30.500 on March 5, 2026, and the unique identifier of the target edge computing node is "ECU-C-05". The parameters of the high-density data acquisition instructions generated by the platform are as follows: the start time of high-density sampling is the alarm time (14:25:30.500), the sampling duration is 60 seconds, the required sensor signal types include hydrogen concentration signal, temperature signal, pressure signal, and flame characteristic signal, and the sampling frequency is uniformly required to be 100Hz. The generation logic of the high-density data acquisition instructions can be represented by the following function: in: This indicates the generated high-density data acquisition instruction data packet. This indicates the time point of the abnormal alarm obtained from the parsing. This indicates the sampling start time offset set in the instruction (usually set to 0, indicating immediate start). This indicates the sampling duration set in the instruction. This indicates a list of sensor signal types required in the instruction. This indicates the sampling frequency required in the instruction. The central analysis platform issues high-density data acquisition instructions to the target edge computing nodes through a dedicated control channel with the highest network priority. This control channel is a virtual local area network (VLAN) dedicated to control instruction transmission, with a network priority set to 7 (highest level), higher than the data stream priority used for transmitting edge risk reports in the time-sensitive network. Referring to Table 1, the central analysis platform simultaneously creates an incident tracing task container locally, named after the incident alarm time and location, to receive subsequent incident retrospective data.

[0033] Table 1: Example of parameters for high-density data acquisition commands In some embodiments, the data structure of the high-density data acquisition command adopts binary encoding, and the command header includes a version number, command type code, and total length field. Optionally, the sampling start time can be set to a very short future time point, such as 100 milliseconds after the current time, to allow for delay time for command transmission and node processing. The process of generating commands by the central analysis platform is completed in milliseconds. The platform completes command encapsulation and sends it to the distribution queue within 100 milliseconds after parsing the alarm signal. It can be understood that the independent control channel and the setting of the highest network priority ensure that the high-density data acquisition command can be delivered to the target edge computing node in real time, regardless of the congestion of the conventional data network. In some embodiments, the required list of sensor signal types can be dynamically adjusted according to the risk type that triggers the alarm. For example, for an alarm triggered by flame characteristics, the signal type list in its command will preferentially include ultraviolet flame detector signals and infrared thermal imager signals. The accident tracing task container created locally by the central analysis platform enters a state of waiting to receive data after creation, and the container index information is recorded in the platform's accident tracing task list. Optionally, the name of the accident tracing task container can include more precise geographic coordinate information.

[0034] See Figure 4 In the multi-parameter anomaly thermal map of the ECU-C-05 node, the standardized values ​​(0-1) of each sensor signal exhibit significant temporal evolution characteristics with sampling time (0-60 seconds), intuitively reflecting the complete process from normal operating conditions to an abnormal outbreak. Specifically: Hydrogen concentration: It remained at an extremely low level (close to 0) within 0-30 seconds, then jumped to 1.0 after 30 seconds and remained at a high level, indicating that a sudden hydrogen leak occurred around 30 seconds, and the leak volume quickly reached the danger threshold. Temperature and pressure: They were in the normal range (0-0.3) within 0-15 seconds, then began to gradually increase after 15 seconds, and stabilized in the medium-to-high risk range of 0.6-0.8 after 30 seconds, reflecting the continuous abnormal increase in ambient temperature and pipeline pressure caused by the leak. Flame signal: It was at a low level (0-0.5) within 0-25 seconds, then rapidly climbed to above 0.8 after 25 seconds, and stabilized at a high level close to 1.0 after 30 seconds, indicating that open flame combustion was triggered shortly after the leak occurred, and the fire continued to expand. From a temporal perspective, the abnormal evolution of each parameter exhibits a clear causal chain: the abrupt change in hydrogen concentration within 30 seconds is the core trigger of the accident; subsequently, the continuous increase in temperature and pressure, along with the rapid response of the flame signal, together constitute the complete accident chain from leak to combustion. This heatmap, through the synchronous visualization of multi-dimensional signals, provides crucial temporal evidence for accident tracing and can be used to reverse-engineer the event sequence from the initial leak to the fire outbreak, supporting precise analysis of the accident's causes.

[0035] In one embodiment of the present invention, an edge computing node receiving a high-density data acquisition command instructs its subordinate environmental sensing sensor network to switch to a high-speed sampling mode. The edge computing node receives and parses the high-density data acquisition command to obtain sampling parameters. Taking a specific scenario as an example, the high-density data acquisition command requires continuous high-speed sampling of hydrogen concentration signals, temperature signals, pressure signals, and flame characteristic signals at a frequency of 100Hz for 60 seconds. The edge computing node sends a sampling mode switching command to all connected sensor terminals of relevant types. The sensor terminals in the environmental sensing sensor network immediately switch from the conventional sampling mode to the high-speed sampling mode according to the received command. The hydrogen concentration sensor switches from sampling once per second to sampling 100 times per second, the infrared thermal imager switches from outputting 10 frames of images per second to outputting 100 frames of temperature data per second, the pressure transmitter switches from sampling five times per second to sampling 100 times per second, and the ultraviolet flame detector switches from status monitoring mode to a 100Hz ultraviolet light intensity high-speed acquisition mode. The sensor terminals acquire raw signals at the high frequency specified in the command. Edge computing nodes add a local timestamp to each raw signal data received in high-speed sampling mode and cache it. Internally, each edge computing node maintains a high-precision clock, appending a timestamp in the format "year-month-day hour:minute:second.millisecond" to each arriving 100Hz signal data, such as "2026-03-05 14:25:30.512", and storing the timestamped raw signal data in a dedicated high-speed cache. After the sampling duration specified by the instruction ends, the edge computing node commands the environmental perception sensor network to switch back to normal sampling mode. Sixty seconds after sampling begins, the edge computing node sends a mode switching command to all relevant sensors, and the hydrogen concentration sensor, infrared thermal imager, pressure transmitter, and ultraviolet flame detector revert to their original normal sampling frequencies. Edge computing nodes compress and encapsulate the cached high-frequency raw signal streams and their corresponding local timestamps in chronological order. The compression process uses a lossless compression algorithm to reduce data volume. During encapsulation, an incident tracing identifier, such as the identifier "ACCIDENT_TRACE_001", is appended to the header of the data packet to form a complete incident tracing data packet. The total data volume of the incident tracing data packet is... It can be estimated using the formula: in: This indicates the estimated size of the incident backtracking data packet. This indicates the number of sensor types involved in high-speed sampling. This represents the sampling frequency of the i-th type of sensor in high-speed mode. Indicates the sampling duration. This represents the average size of a single sample of data from the i-th type of sensor. This represents the average compression ratio of the compression algorithm.

[0036] The central analysis platform uses a global risk map combined with real-time uploaded accident retrospective data packets to reconstruct the complete event sequence from before to after the accident. In practice, the platform decompresses and parses the received accident retrospective data packets to recover high-frequency raw signal streams and their timestamps. After decompressing the data packets, the platform obtains four signal streams collected at 100Hz within 60 seconds, each data point containing a millisecond-level timestamp and raw reading. Historical risk evolution data, centered on the alarm point and traced back a certain period in time, is extracted from the global risk map as background information for the accident precursors. Historical risk data for alarm point "P-205" within ten minutes before the accident is extracted, including a record of a slow upward trend in hydrogen concentration and two brief low-level over-temperature alarm records. The high-frequency signal streams in the accident retrospective data packets are aligned with the accident precursor background information on the timeline, precisely connecting the starting timestamp of the 100Hz hydrogen concentration signal stream "14:25:30.500" with the 1Hz concentration data of the last second in the precursor background information on the timeline. By focusing on abrupt changes in high-frequency signal streams and combining this with background information about potential accidents, the analysis was conducted in reverse chronological order to identify the equipment operation, material changes, or environmental condition alterations that caused each signal change. The analysis revealed a sharp drop in the 100Hz pressure signal stream at "14:25:45.200", which, combined with the preceding information, was deduced to be due to the upstream hydrogen compressor inlet valve being mistakenly closed at that moment. At "14:25:40.100", the 100Hz temperature signal stream showed its first abnormal temperature rise, which, combined with the preceding information, was deduced to be due to localized overheating caused by friction from the compressor running idling due to the valve being closed.

[0037] Optionally, when compressing high-frequency raw signal streams, edge computing nodes can employ a block compression strategy, compressing 60 seconds of data separately according to sensor type and in 10-second blocks before encapsulation. During reverse time analysis, the central analysis platform can establish associations between signal abrupt changes and typical fault modes in the knowledge base. It is understood that background information about accident precursors provides the necessary process context for understanding abrupt changes in high-frequency signal streams, making reverse inference more accurate. In some embodiments, the reconstructed event sequence is visualized as a timeline, and each event node can be expanded to view associated signal evidence curve segments. Optionally, when inferring the cause of an event, in addition to analyzing signal abrupt changes, cross-validation is performed using equipment operation logs and process flow dynamics information recorded in the global risk map. The accident tracing identifier in the accident retrospective data packet header is used by the central analysis platform to quickly identify the data packet type and route it to the corresponding accident tracing task container. It is understood that the entire process from receiving instructions to generating a complete event sequence constitutes a closed analysis loop, the purpose of which is to accurately reproduce the accident evolution process to clarify responsibility and improvement measures.

[0038] See Figure 5 In the accident precursor analysis phase of the edge computing communication and accident tracing system for chemical fire safety, the visualization of hydrogen concentration trends provides crucial procedural evidence for accident tracing. Specifically, the horizontal axis represents the time before the accident (in seconds), the vertical axis represents the hydrogen concentration (in ppm), the curve represents the time-series data of hydrogen concentration acquired through high-speed sampling, and the dashed line represents the preset 50 ppm warning threshold. Background information on accident precursors extracted from the global risk map shows that within the 10-minute time window before the accident (from -600 seconds to 0 seconds), the hydrogen concentration generally shows a gradual upward trend, and the fluctuation amplitude increases significantly over time. Approximately 400 seconds before the accident, the concentration briefly touches the warning threshold for the first time; in the interval from approximately 200 seconds to 0 seconds before the accident, the concentration repeatedly exceeds 50 ppm, with a peak value approaching 56 ppm, and both the frequency and amplitude of fluctuations significantly increase, indicating that the leakage risk continues to accumulate and enters a high-risk stage. When the central analysis platform reverse-engineered the event sequence, it cross-validated the high-resolution concentration curve with the equipment operation logs and process flow dynamics information in the global risk map: the phased rise and increased fluctuation of the concentration were highly consistent with the intermittent start-up and shutdown of the upstream hydrogen compressor and the local throttling operation of the pipeline valves in terms of time; while the continuous high-level oscillation of the concentration in the last 100 seconds before the accident directly corresponded to the continuous leakage event caused by the failure of the key seal.

[0039] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A chemical fire safety situation edge computing communication and accident tracing system, characterized in that, The system includes a central analysis platform and multiple selected edge computing nodes within the chemical industrial park; An environmental sensing sensor network covering its physical area is deployed for each edge computing node. The environmental sensing sensor network continuously collects chemical concentration signals, temperature signals, pressure signals, and flame characteristic signals of its physical area. Each edge computing node synchronously processes multiple signals from the connected environmental sensing sensor network to generate structured regional real-time sensing data blocks. Each edge computing node independently performs a preliminary local security risk assessment based on structured regional real-time perception data blocks, identifying potential risk points and risk types; Each edge computing node will package the assessed local risk points and risk types, along with key summary information from the structured regional real-time perception data blocks, into an edge risk report; The edge risk reports generated by each edge computing node are periodically uploaded to the central analysis platform located in the cloud via a time-sensitive network communication protocol. The central analysis platform receives and integrates edge risk reports from all edge computing nodes, combines static information on equipment layout and dynamic information on process flow in the chemical industrial park, performs global security situation fusion analysis, and calculates a global risk map. Upon detecting any form of abnormal alarm signal, an incident tracing report is generated based on the global risk map.

2. The chemical fire safety situation edge computing communication and accident tracing system according to claim 1, characterized in that, The environmental sensing sensor network continuously collects chemical concentration signals, temperature signals, pressure signals, and flame characteristic signals of the physical area it belongs to, specifically including: The environmental sensing sensor network includes a variety of distributed sensor terminals, including gas concentration sensors, infrared thermal imagers, pressure transmitters, and ultraviolet flame detectors. The gas concentration sensor measures the concentration of a variety of key hazardous chemicals in the air at fixed sampling intervals and generates a chemical concentration signal that includes sensor identification, timestamp, and concentration value. An infrared thermal imager continuously scans a designated area to acquire a sequence of infrared images that reflect the temperature field distribution on the surface of the equipment and in space. The temperature value of a specific monitoring point is extracted from the infrared image sequence to generate a temperature signal. Pressure transmitters are installed on critical pipelines and reaction vessels to monitor internal pressure changes in real time and output pressure signals containing pressure readings and time information. The ultraviolet flame detector monitors the status of open flames within its field of view. When it detects ultraviolet light of a specific wavelength, it generates a flame characteristic signal that includes a fire confirmation mark and the detection time.

3. The chemical fire safety situation edge computing communication and accident tracing system according to claim 2, characterized in that, Each edge computing node synchronously processes multiple signals from the connected environmental sensing sensor network to generate structured real-time regional sensing data blocks, specifically including: The processing includes threshold comparison of abnormal signals, signal drift calibration, and fusion and alignment of multi-source signals on timestamps; Edge computing nodes allocate an independent buffer queue for each type of signal received and record its original timestamp of arrival; Edge computing nodes initiate clock synchronization services, using the unified time signal issued by the central analysis platform as a reference, to calibrate the original timestamps of all signals and convert them to a unified time reference. For signals after the calibration timestamp, outlier detection is performed, signals that exceed the physical range of the sensor are marked as invalid, and signal spikes caused by transient interference are smoothed and filtered. Extract concentration change rate features from chemical concentration signals, temperature rise rate features from temperature signals, and pressure gradient features from pressure signals; All types of processed and time-calibrated signals are aligned and fused according to a unified time reference. Multiple signal feature values, signal quality identifiers, and sensor location codes within the same time window are packaged to form a structured regional real-time sensing data block.

4. The chemical fire safety situation edge computing communication and accident tracing system according to claim 3, characterized in that, Each edge computing node independently performs a preliminary local security risk assessment based on structured regional real-time perception data blocks, identifying potential risk points and risk types, specifically including: The edge computing node stores multiple chemical concentration thresholds, temperature thresholds, and pressure thresholds preset based on regional attributes, and these thresholds are associated with risk levels; Read the chemical concentration characteristics, temperature characteristics, and pressure characteristics of the current time window from the structured regional real-time sensing data block; The chemical concentration characteristic values ​​read are compared with the stored chemical concentration thresholds one by one. Chemicals that exceed the thresholds are identified and their degree of exceedance is recorded as leakage risk points and risk levels. The read temperature feature values ​​are compared with the stored temperature thresholds. Monitoring points that exceed the thresholds are identified and their degree of exceeding the limits are recorded as over-temperature risk points and risk levels. The read pressure characteristic values ​​are compared with the stored pressure thresholds. Monitoring points that exceed the thresholds are identified and their degree of exceedance is recorded as overpressure risk points and risk levels.

5. The chemical fire safety situation edge computing communication and accident tracing system according to claim 4, characterized in that, Each edge computing node will package the assessed local risk points and risk types, along with key summary information from the structured regional real-time perception data block, into an edge risk report, specifically including: Edge computing nodes aggregate all risk points identified within the same assessment period, generating a risk entry for each risk point that includes location code, risk type, risk level, first occurrence time, and duration. In the process of generating structured regional real-time sensing data blocks, feature statistics of various signals within the evaluation period are extracted, including maximum, minimum, average and variance, as key summary information; The current assessment period's timeframe, the unique identifier of this edge computing node, all summarized risk items, and extracted key summary information are organized according to a preset binary encoding format. Add a protocol header before the organized data content. The protocol header contains the report type, data length, sequence number and check code information to complete the data packet encapsulation of the edge risk report. The edge computing node stores the encapsulated edge risk report in the transmission buffer, waiting to be transmitted to the central analysis platform.

6. The chemical fire safety situation edge computing communication and accident tracing system according to claim 5, characterized in that, The process of periodically uploading edge risk reports generated by each edge computing node to a central analysis platform located in the cloud via a time-sensitive network communication protocol specifically includes: The central analysis platform plans communication time slots for all edge computing nodes in the park and allocates different transmission priorities and bandwidth guarantees for data with different risk levels. When each edge computing node arrives at its assigned communication time slot, it checks the transmission buffer for any edge risk reports to be sent. If there are edge risk reports to be sent, the edge computing node loads the edge risk report data packets into the protocol data unit according to the frame format of the time-sensitive network communication protocol; Within the protocol data unit, a corresponding quality of service label and transmission priority identifier are set based on the highest risk level of the risk items contained in the report; Edge computing nodes will send protocol data units carrying quality of service tags and transmission priority identifiers to the uplink network link leading to the central analysis platform via industrial Ethernet switches.

7. The chemical fire safety situation edge computing communication and accident tracing system according to claim 6, characterized in that, The central analysis platform receives and integrates edge risk reports from all edge computing nodes, combines static information on equipment layout and dynamic information on process flow within the chemical industrial park, performs a global security situation fusion analysis, and calculates a global risk map, specifically including: The central analysis platform parses each received edge risk report and extracts the unique identifier of the edge computing node, risk items, and key summary information. The extracted risk items are mapped to the corresponding coordinates on the stored 3D digital map of the chemical industrial park according to their location codes. The operating status of the equipment around the risk point, the types of materials involved, and the current process operation are obtained from the dynamic information database of the process flow. A comprehensive analysis is conducted on the geographical location, risk type, risk level, and dynamic information of surrounding processes of the risk points to assess the spread trend of the risk points, the equipment affected, and the chain risks to adjacent areas. All risk points identified in the assessment, their current status, scope of impact, and cascading risk relationships are marked on the 3D digital map of the chemical industrial park using an overlay method, forming a dynamic and visualized global risk map.

8. The chemical fire safety situation edge computing communication and accident tracing system according to claim 7, characterized in that, The step of generating an incident tracing report based on the global risk map upon detecting any form of abnormal alarm signal includes: The central analysis platform immediately triggered the accident tracing mode and issued a high-density data collection command to the edge computing node that issued the alarm signal; The edge computing node that receives the high-density data acquisition command instructs its subordinate environmental perception sensor network to switch to high-speed sampling mode and encapsulates the acquired raw signal stream with the complete timestamp to form an accident backtracking data packet; The edge computing nodes upload the accident backtracking data packets to the central analysis platform. The central analysis platform uses the global risk map and the real-time uploaded accident backtracking data packets to reconstruct the complete event sequence from before the accident to after the accident. The central analysis platform will compare the reconstructed complete event sequence with typical patterns in the pre-set chemical accident knowledge base to determine the direct and indirect cause chains of the accident, and finally generate a standardized accident tracing report. The central analysis platform immediately triggers the incident tracing mode, issuing high-density data collection commands to the edge computing nodes corresponding to the alarm signals, specifically including: An abnormal alarm signal originates from any of the following: a risk point determined by the central analysis platform to have reached a catastrophic risk level in the global security situation fusion analysis; or, an edge risk report uploaded by the edge computing node contains the highest level risk item that requires immediate response; or, an emergency alarm manually triggered by a person through the monitoring system. Once an abnormal alarm signal is detected, the central analysis platform immediately analyzes the abnormal alarm signal and locates the unique identifier of the edge computing node that issued the alarm and its associated physical area; The central analysis platform generates high-density data acquisition instructions, which include the unique identifier of the target edge computing node, the start time of high-density sampling, the sampling duration, and the required sensor signal type and sampling frequency. The central analysis platform sends high-density data acquisition commands to the target edge computing nodes through an independent control channel with the highest network priority; The central analysis platform also creates an incident tracing task container locally, named after the incident alarm time and location, to receive subsequent incident backtracking data.

9. The chemical fire safety situation edge computing communication and accident tracing system according to claim 8, characterized in that, The edge computing node that receives the high-density data acquisition command instructs its subordinate environmental perception sensor network to switch to high-speed sampling mode, and encapsulates the acquired raw signal stream with a complete timestamp to form an accident backtracking data packet, specifically including: Edge computing nodes receive and parse high-density data acquisition instructions, obtain sampling parameters, and send sampling mode switching commands to all connected sensor terminals of relevant types; In an environmental sensing sensor network, the sensor terminal immediately switches from the conventional sampling mode to the high-speed sampling mode upon receiving a command, and acquires the raw signal at the high frequency specified by the instruction. The edge computing node adds a local timestamp to each piece of raw signal data received in high-speed sampling mode and caches it. After the sampling duration specified in the instruction ends, the edge computing node commands the environmental awareness sensor network to switch back to the normal sampling mode. Edge computing nodes compress and encapsulate the cached high-frequency raw signal streams and their corresponding local timestamps in chronological order, and attach an incident tracing identifier to the header of the data packet to form a complete incident tracing data packet.

10. The chemical fire safety situation edge computing communication and accident tracing system according to claim 9, characterized in that, The central analysis platform utilizes a global risk map, combined with real-time uploaded accident retrospective data packets, to reconstruct the complete event sequence from before to after the accident, specifically including: The central analysis platform decompresses and parses the received accident backtracking data packets to recover the high-frequency original signal stream and its timestamps; From the global risk map, historical risk evolution data centered on alarm points and traced back a certain period in time are extracted as background information for accident precursors. The high-frequency signal stream in the accident retrospective data packet is matched and aligned with the background information of the accident precursors on the time axis. Taking the abrupt change points in the high-frequency signal stream as the core, combined with the background information of the accident precursors, the equipment operation, material change or environmental condition change that caused each signal abrupt change was analyzed and inferred in reverse chronological order. The series of events deduced through analysis are arranged in logical and chronological order to form a complete event sequence from the initial anomaly to the outbreak of the accident. Each event in the sequence is associated with a specific time point, the equipment or materials involved, and signal evidence.

Images