Rural limited space intelligent supervision system and method based on multi-source data fusion

Abstract

This invention discloses an intelligent monitoring system and method for rural confined spaces based on multi-source data fusion, comprising the following steps: collecting multi-source environmental and personnel monitoring information and completing standardized processing; encrypting and reliably transmitting the monitoring data; constructing queryable multi-source historical time-series data; achieving risk assessment and level determination based on a dual-branch model and evidence fusion; generating tiered alarms according to rules and associating them with disposal work orders; executing emergency response and writing back the results to form a closed-loop monitoring system. This invention, through multi-source data fusion and intelligent risk trend analysis, achieves dynamic assessment and closed-loop handling of safety risks in rural confined spaces, possessing advantages such as accurate early warning, timely response, and high monitoring reliability.

Description

Technical Field

[0001] This invention relates to the field of rural intelligent supervision, and in particular to a rural confined space intelligent supervision system and method based on multi-source data fusion. Background Technology

[0002] In rural areas, numerous and scattered confined spaces such as biogas digesters, septic tanks, and sewage wells can easily generate toxic and harmful gases such as methane and hydrogen sulfide, posing significant safety risks. Current technologies typically monitor environmental parameters in these confined spaces using single or a small number of gas sensors and transmit the data wirelessly to a backend platform for alarms or manual inspections to assist management. Some systems incorporate video surveillance or remote fan control to detect intrusions or take simple action when gas levels exceed limits, but overall, they still rely primarily on threshold triggers and manual judgment, lacking the ability to systematically analyze multi-source monitoring data and understand risk evolution.

[0003] However, existing technologies for monitoring confined spaces in rural areas generally suffer from problems such as low utilization of multi-source data, rudimentary risk assessment, and incomplete closed-loop response. On the one hand, there is a lack of effective fusion mechanisms between data from different sensors and video recognition results, which can easily lead to false alarms or missed alarms due to instantaneous fluctuations or single-source anomalies, making it difficult to reflect the true trend of risks. On the other hand, most systems only operate at the alarm level, lacking a unified data link between alarms and response, feedback, and verification, making it difficult to form traceable closed-loop monitoring results, support tiered response and continuous optimization of monitoring strategies, and meet the development needs of refined and intelligent safety monitoring of confined spaces in rural areas.

[0004] Therefore, how to provide an intelligent monitoring system and method for rural confined spaces based on multi-source data fusion is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent monitoring system and method for rural confined spaces based on multi-source data fusion. This invention achieves dynamic assessment and closed-loop handling of safety risks in rural confined spaces through multi-source data fusion and intelligent risk trend analysis, and has the advantages of accurate early warning, timely response and high monitoring reliability.

[0006] The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to embodiments of the present invention includes the following steps: Collect data on methane and hydrogen sulfide concentrations, temperature, liquid level, and AI camera electronic fence monitoring and human recognition results to form a standardized multi-source monitoring dataset; The standardized multi-source monitoring dataset is encrypted with AES-256 using the TCP / IP+MQTT protocol stack and uploaded with breakpoint resume capability to form a secure data transmission stream. Securely transmit data streams to the InfluxDB time series database and build historical time series data to form a searchable multi-source historical dataset. Risk trend analysis is performed on multi-source historical datasets using a dual-branch long short-term memory network. Evidence credibility configuration is formed based on cross-branch consistency gating units. Multi-source data are fused using DS evidence theory to obtain fused risk assessment results. The risk trend analysis results are used as prior constraints to generate risk level adjudication results. Based on the three-level triggering conditions in the alarm rule base, the risk level adjudication results generate graded alarm events, and the workflow engine of the back-end management subsystem forms work order-instruction association results; The system executes remote start / stop of ventilation equipment, voice alarms, and a three-tiered response mechanism (village-town-district) and a tiered timeout escalation mechanism for alarm work orders and handling instructions. It collects handling feedback from the mobile application subsystem and intelligently verifies the handling results through the back-end management subsystem before writing them back to the InfluxDB time-series database and the electronic file management of hazard sources, forming a closed-loop regulatory result set.

[0007] Optionally, the generation of the standardized multi-source monitoring dataset specifically includes: Intelligent sensing units are deployed within the rural confined space monitoring area, and each sensing unit is synchronized in a unified time and calibrated in its installation location to form a sensing baseline state. The intelligent sensing unit includes a methane sensing unit, a hydrogen sulfide sensing unit, a temperature sensing unit, a liquid level sensing unit, and an AI camera sensing unit. The raw monitoring data of methane concentration and hydrogen sulfide concentration are obtained through the methane sensing unit and the hydrogen sulfide sensing unit, and range consistency correction and zero point offset compensation are performed to generate gas concentration monitoring results. The temperature of the internal environment of the confined space is periodically collected by the temperature sensing unit, and abnormal jitter suppression and time window smoothing are performed to form temperature monitoring results. The liquid level is detected in real time within a confined space by a liquid level sensing unit. The detection results are then processed to a uniform scale based on the installation reference height of the liquid level sensing unit to generate liquid level monitoring results. The AI ​​camera sensing unit collects video of the surrounding and internal areas of a limited space, sets up an electronic fence monitoring area, performs human recognition processing on targets entering the electronic fence monitoring area, and generates personnel intrusion recognition results. Perform unified timestamp alignment and data structure encapsulation on gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results to form a multi-source monitoring dataset containing multiple monitoring dimensions. Perform data integrity verification and validity marking on the multi-source monitoring dataset, remove missing or abnormal monitoring data items, and generate a standardized multi-source monitoring dataset.

[0008] Optionally, the generation of the secure transmission data stream specifically includes: encapsulating a standardized multi-source monitoring dataset based on the monitoring point identifier and data type, constructing a communication session using TCP / IP and MQTT protocols, adaptively selecting a transmission mode between fourth-generation mobile communication and narrowband IoT based on the current communication signal quality, enabling LoRa self-organizing network for data forwarding in areas without network coverage, performing AES-256 encryption on the data to be transmitted and attaching integrity verification information, recording and controlling the data transmission status through a breakpoint resume mechanism, updating the status of the transmitted data based on the confirmation receipt in the communication session, forming continuously confirmed upload results, and aggregating them into a secure transmission data stream in chronological order.

[0009] Optionally, the generation of the multi-source historical dataset specifically includes: performing decapsulation and integrity verification on the securely transmitted data stream, recovering the gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results contained therein, performing structured parsing and sequential organization of various monitoring results based on monitoring point identifiers and timestamps, and writing the parsed monitoring records into the InfluxDB time-series database. During the writing process, a time-series storage structure is established according to the monitoring point dimension and monitoring type to form continuously covered historical time-series data, and a multi-source historical dataset supporting time alignment and conditional retrieval is constructed based on the historical time-series data.

[0010] Optionally, the generation of the risk level determination result specifically includes: Extract gas concentration monitoring result sequences, temperature monitoring result sequences, liquid level monitoring result sequences, and personnel intrusion identification result sequences that are aligned to timestamps from multi-source historical datasets according to monitoring point identifiers. Slice and resample each sequence according to preset time windows to form a windowed time series sample set. The windowed time series sample set is divided into monitoring dimensions to construct the input of a dual-branch long short-term memory network. The gas concentration monitoring result sequence is used as the gas branch input and the temperature monitoring result sequence is used as the temperature branch input. Time series encoding and memory state update are performed to generate gas branch trend representation and temperature branch trend representation. In the cross-branch consistency gating unit, consistency calculation and gating mapping are performed on the trend representations of the gas branch and the trend representations of the temperature branch, the gating results are output and mapped to the evidence credibility configuration; Risk trend analysis results are generated based on gas branch trend characterization and temperature branch trend characterization. Based on the credibility of the evidence, evidence is constructed from gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, personnel intrusion identification results, and risk trend analysis results. Evidence allocation functions are generated for each, and the results are input into the DS evidence theory fusion process to obtain the fusion risk assessment results. The DS evidence theory fusion process calculates the support combination result of the two evidences on the risk level value set for any two evidence allocation functions, calculates the conflict degree formed by the incompatible parts of the two evidences, normalizes the conflict degree, normalizes and synthesizes each pair of support combination results to obtain the fusion allocation function of the two evidences, and iterates and fuses the fusion allocation function with the other evidence allocation functions one by one according to the same rules to obtain the final fusion allocation function covering all evidence allocation functions. The fusion risk assessment result is determined by the final fusion allocation function. The risk trend analysis results are used as a priori constraints to perform consistency adjudication and level correction on the integrated risk assessment results, thereby generating risk level adjudication results. The level correction includes: limiting the upward adjustment of the risk level of the fusion risk assessment result when the risk trend analysis result indicates a continuously rising state; limiting the maintenance of the risk level of the fusion risk assessment result when the risk trend analysis result indicates a stable state; and limiting the delayed confirmation of the risk level of the fusion risk assessment result when the risk trend analysis result indicates an abnormal fluctuation state.

[0011] Optionally, the generation of the work order-instruction association result specifically includes: The risk level determination results are collected based on the monitoring point identification to form a risk level determination record set for a single monitoring point; In the alarm rule base, load the risk level adjudication record set with three-level triggering conditions corresponding to the rural confined space type. The three-level triggering conditions include threshold over-limit triggering conditions, trend abnormality triggering conditions, and multi-source collaborative alarm triggering conditions. Encapsulate the three-level triggering conditions into a set of executable rule entries to form an alarm rule entry set. Based on the alarm rule set, the risk level adjudication record set is matched with the rules one by one. The threshold over-limit judgment result, the trend abnormality judgment result, and the multi-source collaborative alarm judgment result are triggered and aggregated to generate a trigger summary. Based on the trigger summary and the risk level adjudication result, the alarm level, alarm type and alarm time window of the graded alarm event are determined to form a graded alarm event. The workflow engine of the background management subsystem receives hierarchical alarm events, generates alarm work orders for hierarchical alarm events based on monitoring point identifiers and alarm types, and writes them into the pending processing queue to form an alarm work order set. Based on the alarm work order set, a handling instruction is generated, including control instructions for remote start and stop of ventilation equipment, handling task instructions from the mobile application subsystem, and flow instructions for the three-level response of village-town-district. The handling instructions are then bound to the corresponding alarm work orders to form a work order-instruction association result.

[0012] Optionally, the generation of the closed-loop monitoring result set specifically includes: based on the work order-instruction association result, the handling instructions corresponding to the alarm work orders are uniformly scheduled and issued in the background management subsystem, remote start and stop control of ventilation equipment, voice alarm triggering, and three-level response flow from village to town to district are executed, and handling status and response time are established at each response level. When any level fails to complete the response confirmation within the preset time window, the hierarchical timeout escalation mechanism is triggered, and the alarm work order is automatically escalated and flowed in hierarchical order. During the handling process, the gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results are collected through the mobile application subsystem to confirm the arrival of the handling personnel and verify the on-site verification. The handling feedback is generated, and the background management subsystem performs consistency verification on the handling feedback and the multi-source historical dataset, ventilation equipment execution receipts, and voice alarm receipts within the alarm time window, generates handling verification results, and writes them back to the InfluxDB time series database and the electronic file management of hazardous sources to form a closed-loop monitoring result set.

[0013] The intelligent monitoring system for rural confined spaces based on multi-source data fusion according to an embodiment of the present invention includes: The intelligent sensing module is used to collect gas concentration data, temperature data, liquid level height, and electronic fence monitoring and human recognition results in limited spaces in rural areas, forming a standardized multi-source monitoring dataset; The data transmission module is used to encrypt and encapsulate standardized multi-source monitoring datasets based on the TCP / IP and MQTT protocol stack, and upload them with breakpoint resumption to form a secure data transmission stream; The data platform module is used to receive securely transmitted data streams and write them into the InfluxDB time series database to establish continuously covered historical time series data and form a searchable multi-source historical dataset. The risk analysis module is used to construct a dual-branch long short-term memory network based on multi-source historical datasets, perform risk trend analysis, form evidence credibility configuration through cross-branch consistency gating units, fuse multi-source data using DS evidence theory, and generate fused risk assessment results and risk level adjudication results. The alarm and work order management module is used to generate graded alarm events based on the three-level trigger conditions of the alarm rule base, and form work order-instruction association results through the workflow engine of the background management subsystem. The emergency response and closed-loop monitoring module is used to execute remote start / stop of ventilation equipment, voice alarms, village-town-district three-level response and graded timeout escalation mechanism, collect the response feedback from the mobile application subsystem and perform intelligent verification of the response results, write back to the time-series database and the electronic file management of hazard sources to form a closed-loop monitoring result set.

[0014] The beneficial effects of this invention are: This invention achieves continuous, dynamic, and intelligent monitoring of safety risks in confined rural spaces by constructing a complete technical chain covering perception, transmission, analysis, alarm, and response. Compared to traditional methods relying on single sensor threshold judgments or manual inspections, this invention introduces unified acquisition and time-series management of multi-source monitoring data. It also uses a dual-branch long short-term memory network to model trends in key risk factors such as gas and temperature, effectively depicting the evolution of risks and avoiding false alarms and missed alarms caused by triggering alarms based solely on instantaneous anomalies. Simultaneously, through cross-branch consistency gating units and evidence credibility configuration, different monitoring sources are constrained and weighted at the trend consistency level, making the fused risk assessment results more consistent with the objective laws of risk formation under real-world conditions.

[0015] Furthermore, this invention incorporates risk trend analysis results as a priori constraints into the multi-source fusion and risk level adjudication process, enabling risk assessment to not only reflect the current state but also the directionality and sustainability of risk development, thereby significantly improving the foresight and stability of alarm decisions. Based on this, through a rule-driven hierarchical alarm system, a work order and disposal instruction linkage mechanism, and a closed-loop disposal process involving ventilation control, voice alarms, and a three-tiered response system (village-township-district), a closed-loop supervision system covering the entire process from risk identification to on-site intervention and result verification is achieved. This effectively reduces the probability of delayed response and disposal failures in rural confined space accidents, significantly improving the overall intelligence level of safety supervision and the actual prevention and control effectiveness. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows the intelligent monitoring method for rural confined spaces based on multi-source data fusion proposed in this invention. Figure 2 This is a schematic diagram of the risk trend analysis and evidence fusion structure based on a dual-branch long short-term memory network for the intelligent monitoring method of rural confined space based on multi-source data fusion proposed in this invention. Figure 3 This is a schematic diagram of the hierarchical alarm and closed-loop monitoring process of the intelligent monitoring method for rural confined spaces based on multi-source data fusion proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A method for intelligent monitoring of confined spaces in rural areas based on multi-source data fusion includes the following steps: Collect data on methane and hydrogen sulfide concentrations, temperature, liquid level, and AI camera electronic fence monitoring and human recognition results to form a standardized multi-source monitoring dataset; The standardized multi-source monitoring dataset is encrypted with AES-256 using the TCP / IP+MQTT protocol stack and uploaded with breakpoint resume capability to form a secure data transmission stream. Securely transmit data streams to the InfluxDB time series database and build historical time series data to form a searchable multi-source historical dataset. Risk trend analysis is performed on multi-source historical datasets using a dual-branch long short-term memory network. Evidence credibility configuration is formed based on cross-branch consistency gating units. Multi-source data are fused using DS evidence theory to obtain fused risk assessment results. The risk trend analysis results are used as prior constraints to generate risk level adjudication results. Based on the three-level triggering conditions in the alarm rule base, the risk level adjudication results generate graded alarm events, and the workflow engine of the back-end management subsystem forms work order-instruction association results; The system executes remote start / stop of ventilation equipment, voice alarms, and a three-tiered response mechanism (village-town-district) and a tiered timeout escalation mechanism for alarm work orders and handling instructions. It collects handling feedback from the mobile application subsystem and intelligently verifies the handling results through the back-end management subsystem before writing them back to the InfluxDB time-series database and the electronic file management of hazard sources, forming a closed-loop regulatory result set.

[0019] In this embodiment, the generation of the standardized multi-source monitoring dataset specifically includes: Intelligent sensing units are deployed within the rural confined space monitoring area, and each sensing unit is synchronized in a unified time and calibrated in its installation location to form a sensing baseline state. The intelligent sensing unit includes a methane sensing unit, a hydrogen sulfide sensing unit, a temperature sensing unit, a liquid level sensing unit, and an AI camera sensing unit. The raw monitoring data of methane concentration and hydrogen sulfide concentration are obtained through the methane sensing unit and the hydrogen sulfide sensing unit, and range consistency correction and zero point offset compensation are performed to generate gas concentration monitoring results. The temperature of the internal environment of the confined space is periodically collected by the temperature sensing unit, and abnormal jitter suppression and time window smoothing are performed to form temperature monitoring results. The liquid level is detected in real time within a confined space by a liquid level sensing unit. The detection results are then processed to a uniform scale based on the installation reference height of the liquid level sensing unit to generate liquid level monitoring results. The AI ​​camera sensing unit collects video of the surrounding and internal areas of a limited space, sets up an electronic fence monitoring area, performs human recognition processing on targets entering the electronic fence monitoring area, and generates personnel intrusion recognition results. Perform unified timestamp alignment and data structure encapsulation on gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results to form a multi-source monitoring dataset containing multiple monitoring dimensions. Perform data integrity verification and validity marking on the multi-source monitoring dataset, remove missing or abnormal monitoring data items, and generate a standardized multi-source monitoring dataset.

[0020] In this embodiment, the generation of the secure transmission data stream specifically includes: encapsulating a standardized multi-source monitoring dataset based on the monitoring point identifier and data type; constructing a communication session using TCP / IP and MQTT protocols; adaptively selecting a transmission mode between fourth-generation mobile communication and narrowband IoT based on the current communication signal quality; enabling LoRa self-organizing network for data forwarding in areas without network coverage; performing AES-256 encryption on the data to be transmitted and attaching integrity verification information; recording and controlling the data transmission status through a breakpoint resume mechanism; updating the status of the transmitted data based on the confirmation receipt in the communication session; forming continuously confirmed upload results; and aggregating them into a secure transmission data stream in chronological order.

[0021] In this embodiment, the generation of the multi-source historical dataset specifically includes: performing decapsulation and integrity verification on the secure transmission data stream, restoring the gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results contained therein, performing structured parsing and sequential organization of various monitoring results based on monitoring point identifiers and timestamps, and writing the parsed monitoring records into the InfluxDB time-series database. During the writing process, a time-series storage structure is established according to the monitoring point dimension and monitoring type to form continuously covered historical time-series data, and a multi-source historical dataset supporting time alignment and conditional retrieval is constructed based on the historical time-series data.

[0022] In this embodiment, the generation of the risk level determination result specifically includes: Extract gas concentration monitoring result sequences, temperature monitoring result sequences, liquid level monitoring result sequences, and personnel intrusion identification result sequences that are aligned to timestamps from multi-source historical datasets according to monitoring point identifiers. Slice and resample each sequence according to preset time windows to form a windowed time series sample set. The windowed time series sample set is divided into monitoring dimensions to construct the input of a dual-branch long short-term memory network. The gas concentration monitoring result sequence is used as the gas branch input and the temperature monitoring result sequence is used as the temperature branch input. Time series encoding and memory state update are performed to generate gas branch trend representation and temperature branch trend representation. In the cross-branch consistency gating unit, consistency calculation and gating mapping are performed on the trend representations of the gas branch and the trend representations of the temperature branch, the gating results are output and mapped to the evidence credibility configuration; The consistency calculation targets the gas branch trend representation and the temperature branch trend representation within the same time window, and sequentially performs trend direction consistency judgment, change amplitude coordination judgment, and trend persistence consistency judgment. Trend direction consistency judgment is completed by reading the risk change direction indicated by the two branch trend representations respectively and determining whether they simultaneously indicate risk increase, risk stability, or risk decrease. Change amplitude coordination judgment is completed by comparing whether the trend intensity difference reflecting the strength of change in the two branch trend representations meets the threshold, provided that the trend direction is consistent. Trend persistence consistency judgment is completed by comparing the maintenance of the risk change direction of the two branch trend representations within multiple adjacent time windows, which together constitute the cross-branch trend consistency judgment result. The gating mapping generates discrete gating states based on the cross-branch trend consistency determination results. When the three determination results of trend direction consistency, change amplitude coordination, and trend persistence consistency are all satisfied, a high consistency gating state is generated. When trend direction consistency is satisfied, but only one of change amplitude coordination or trend persistence consistency is not satisfied, a medium consistency gating state is generated. When trend direction consistency is not satisfied, or change amplitude coordination and trend persistence consistency are not satisfied at the same time, a low consistency gating state is generated. Risk trend analysis results are generated based on gas branch trend characterization and temperature branch trend characterization. Specifically, when the trend directions of two branches are consistent, the common trend direction is taken as the window-level fusion trend direction, and the window-level fusion trend intensity is determined based on the branch with higher change intensity. When the trend directions of two branches are inconsistent, the fusion trend is adjudicated in conjunction with the gating state. When the gating state is high consistency or medium consistency, the gas branch trend direction is taken as the fusion trend direction and the fusion trend intensity is limited. When the gating state is low consistency, the window-level fusion trend direction is determined as risk stable and marked as an abnormal fluctuation candidate window. The window-level fusion trend results of multiple consecutive time windows are subjected to consistency solidification processing. When the fusion trend direction continuously points to risk increase within a continuous time window, the risk trend analysis result is output as a continuously rising state. When the fusion trend direction remains stable for a long time or the change intensity is low, the risk trend analysis result is output as a stable state. When the fusion trend direction frequently switches within a continuous time window or the proportion of abnormal fluctuation candidate windows is too high, the risk trend analysis result is output as an abnormal fluctuation state. Based on the credibility of the evidence, evidence is constructed from gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, personnel intrusion identification results, and risk trend analysis results. Evidence allocation functions are generated for each, and the results are input into the DS evidence theory fusion process to obtain the fusion risk assessment results. The DS evidence theory fusion process calculates the support combination result of the two evidences on the risk level value set for any two evidence allocation functions, calculates the conflict degree formed by the incompatible parts of the two evidences, normalizes the conflict degree, normalizes and synthesizes each pair of support combination results to obtain the fusion allocation function of the two evidences, and iterates and fuses the fusion allocation function with the other evidence allocation functions one by one according to the same rules to obtain the final fusion allocation function covering all evidence allocation functions. The fusion risk assessment result is determined by the final fusion allocation function. The risk trend analysis results are used as a priori constraints to perform consistency adjudication and level correction on the integrated risk assessment results, thereby generating risk level adjudication results. The level correction includes: limiting the upward adjustment of the risk level of the fusion risk assessment result when the risk trend analysis result indicates a continuously rising state; limiting the maintenance of the risk level of the fusion risk assessment result when the risk trend analysis result indicates a stable state; and limiting the delayed confirmation of the risk level of the fusion risk assessment result when the risk trend analysis result indicates an abnormal fluctuation state.

[0023] In this embodiment, the generation of the work order-instruction association result specifically includes: The risk level determination results are collected based on the monitoring point identification to form a risk level determination record set for a single monitoring point; In the alarm rule base, load the risk level adjudication record set with three-level triggering conditions corresponding to the rural confined space type. The three-level triggering conditions include threshold over-limit triggering conditions, trend abnormality triggering conditions, and multi-source collaborative alarm triggering conditions. Encapsulate the three-level triggering conditions into a set of executable rule entries to form an alarm rule entry set. Based on the alarm rule set, the risk level adjudication record set is matched with the rules one by one. The threshold over-limit judgment result, the trend abnormality judgment result, and the multi-source collaborative alarm judgment result are triggered and aggregated to generate a trigger summary. Based on the trigger summary and the risk level adjudication result, the alarm level, alarm type and alarm time window of the graded alarm event are determined to form a graded alarm event. Specifically, for threshold over-limit triggering conditions, the gas concentration monitoring results, temperature monitoring results, and liquid level monitoring results associated with the risk level adjudication results are read and over-limit judgments are made to obtain threshold over-limit judgment results. For trend anomaly triggering conditions, the risk trend analysis results associated with the risk level adjudication results are read and anomaly judgments are made to obtain trend anomaly judgment results. For multi-source collaborative alarm triggering conditions, the fusion risk assessment results, evidence credibility configuration, and personnel intrusion identification results associated with the risk level adjudication results are read and collaborative judgments are made to obtain multi-source collaborative alarm judgment results. The alarm types include gas exceedance alarm, abnormal trend alarm, personnel intrusion alarm, and multi-source collaborative alarm, and the alarm time window is the time window range corresponding to the risk level adjudication result; The workflow engine of the background management subsystem receives hierarchical alarm events, generates alarm work orders for hierarchical alarm events based on monitoring point identifiers and alarm types, and writes them into the pending processing queue to form an alarm work order set. The alarm work order includes work order number, alarm level, alarm type, monitoring point identifier, alarm time window, risk level adjudication result, integrated risk assessment result, and evidence credibility configuration; Based on the alarm work order set, a handling instruction is generated, including control instructions for remote start and stop of ventilation equipment, handling task instructions from the mobile application subsystem, and flow instructions for the three-level response of village-town-district. The handling instructions are then bound to the corresponding alarm work orders to form a work order-instruction association result.

[0024] In this embodiment, the generation of the closed-loop monitoring result set specifically includes: based on the work order-instruction association result, the handling instructions corresponding to the alarm work orders are uniformly scheduled and issued in the background management subsystem, remote start and stop control of ventilation equipment, voice alarm triggering, and three-level response flow from village to town to district are executed, and handling status and response time are established at each response level. When any level fails to complete the response confirmation within the preset time window, the hierarchical timeout escalation mechanism is triggered, and the alarm work order is automatically escalated and flowed in hierarchical order. During the handling process, the mobile application subsystem collects the gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results of the on-site confirmation and verification by the handling personnel, and forms the handling feedback. The background management subsystem performs consistency verification on the handling feedback and the multi-source historical dataset, ventilation equipment execution receipt and voice alarm receipt within the alarm time window, generates the handling verification result, and writes it back to the InfluxDB time series database and the electronic file management of hazardous sources to form a closed-loop monitoring result set.

[0025] A smart monitoring system for confined spaces in rural areas based on multi-source data fusion includes: The intelligent sensing module is used to collect gas concentration data, temperature data, liquid level height, and electronic fence monitoring and human recognition results in limited spaces in rural areas, forming a standardized multi-source monitoring dataset; The data transmission module is used to encrypt and encapsulate standardized multi-source monitoring datasets based on the TCP / IP and MQTT protocol stack, and upload them with breakpoint resumption to form a secure data transmission stream; The data platform module is used to receive securely transmitted data streams and write them into the InfluxDB time series database to establish continuously covered historical time series data and form a searchable multi-source historical dataset. The risk analysis module is used to construct a dual-branch long short-term memory network based on multi-source historical datasets, perform risk trend analysis, form evidence credibility configuration through cross-branch consistency gating units, fuse multi-source data using DS evidence theory, and generate fused risk assessment results and risk level adjudication results. The alarm and work order management module is used to generate graded alarm events based on the three-level trigger conditions of the alarm rule base, and form work order-instruction association results through the workflow engine of the background management subsystem. The emergency response and closed-loop monitoring module is used to execute remote start / stop of ventilation equipment, voice alarms, village-town-district three-level response and graded timeout escalation mechanism, collect the response feedback from the mobile application subsystem and perform intelligent verification of the response results, write back to the time-series database and the electronic file management of hazard sources to form a closed-loop monitoring result set.

[0026] Example 1: To verify the feasibility of this invention in practice, it was applied to a confined space monitoring scenario in a rural area of ​​North China, involving concentrated underground sewage inspection wells, biogas digesters, and agricultural sewage drainage ditches. This area presents a dispersed and complex environment with a long-term risk of toxic and harmful gas accumulation, such as methane and hydrogen sulfide. Furthermore, manual inspections rely on experience-based judgment, leading to delayed responses, frequent false alarms, and incomplete closed-loop management, placing significant pressure on grassroots safety supervision.

[0027] In practical applications, intelligent sensing modules are deployed at entrances and key locations within confined spaces to continuously monitor methane and hydrogen sulfide concentrations, ambient temperature, liquid levels, and surrounding activity. An electronic fence is constructed using AI cameras to identify unauthorized entry. After local standardization, the sensed data is encrypted and uploaded via a data transmission module with breakpoint resumption, continuously writing to a data platform to form a multi-source historical dataset. The backend risk analysis module automatically performs risk trend analysis based on historical time-series data, characterizing gas changes and temperature evolution through a dual-branch long short-term memory network, and constraining multi-source trends using cross-branch consistency gating units to avoid misjudgments caused by single sensor anomalies. Based on this, the system integrates liquid level changes and unauthorized entry information to output a stable risk level assessment result.

[0028] When the risk level reaches the alarm criteria, the system automatically generates a tiered alarm event based on the alarm rule base, and the workflow engine establishes a linkage mechanism between work orders and disposal instructions. Disposal instructions are simultaneously sent to the ventilation equipment control terminal and the mobile application subsystem, triggering on-site ventilation, voice prompts, and multi-level response processes at the village, town, and district levels. During the disposal process, the system continuously collects disposal feedback information and verifies the consistency of the feedback results with multi-source historical data before and after the alarm, ensuring that the work order is closed only after the risk is resolved. All process data is automatically archived in the electronic hazard source file management system.

[0029] From an operational perspective, this implementation method significantly improves the real-time performance and reliability of rural confined space monitoring. The system can identify the evolutionary trends of gas accumulation and environmental anomalies in advance, transforming alarm triggering from passive exceeding limits to trend-aware driving, thus reducing the probability of sudden risks. Multi-source evidence fusion and consistency gating mechanisms effectively reduce false alarms caused by single-point sensor fluctuations, increasing the trust of grassroots personnel in alarm results. Simultaneously, through closed-loop management of work orders and handling instructions, on-site response efficiency and cross-level collaboration capabilities are significantly improved. The monitoring process shifts from post-event traceability to full-process tracking and continuous optimization, fully demonstrating the practical value and promotional significance of this invention in rural confined space safety monitoring.

[0030] Table 1. Performance comparison between the intelligent monitoring method for rural confined spaces based on multi-source data fusion and traditional methods.

[0031] As can be seen from Table 1, this invention achieves a systematic improvement in overall regulatory capabilities. First, regarding the coverage of regulatory points, traditional methods rely on manual inspections and decentralized equipment deployment, which can only cover a limited number of locations. In contrast, this invention, through intelligent sensing modules and a unified data access mechanism, enables continuous online monitoring of all regulatory points, achieving full coverage and providing a complete data foundation for subsequent risk analysis.

[0032] In terms of risk detection and early warning capabilities, this invention demonstrates significant advantages. The average detection time for gas anomalies is shortened from a relatively long period to identification within a single time window, and the early identification rate of risk trends is significantly improved. This improvement mainly stems from the joint modeling capability of the dual-branch long short-term memory network for the temporal evolution of gas and temperature, enabling the system to no longer rely on a single threshold trigger but instead capture the accumulation and changing trends of risks, thereby achieving early warning.

[0033] Regarding alarm quality, this invention effectively suppresses false triggering caused by occasional sensor fluctuations and environmental interference through cross-branch consistency gating units and evidence credibility configuration mechanisms, significantly reducing the false alarm rate and substantially improving the accuracy of tiered alarms. This demonstrates that multi-source evidence fusion and consistency constraints possess higher stability and credibility in complex rural environments.

[0034] In terms of emergency response and handling execution, the efficiency of issuing handling instructions, the success rate of remote start / stop of ventilation equipment, and the effectiveness of the timeout escalation mechanism are all significantly better than traditional methods. This is thanks to the unified scheduling capability of the workflow engine in the back-end management subsystem for alarm work orders and handling instructions, which transforms the response process from manual dependence to automatic system flow, reducing human delays.

[0035] Finally, regarding closed-loop monitoring and data accumulation, this invention achieves near-complete completion of risk response closed loops and high hazard source file integrity. Through consistency verification of response feedback, equipment acknowledgments, and historical data, it ensures that every alarm generates a complete record and traceable results, providing a reliable basis for subsequent regulatory decisions and risk reviews.

[0036] In summary, this invention, through the collaborative design of multi-source data fusion, risk trend analysis, hierarchical alarm and closed-loop handling, has achieved substantial improvements in key indicators such as risk detection speed, alarm accuracy, response efficiency and regulatory integrity, fully verifying its technical advantages and application value in the intelligent supervision scenario of rural confined spaces.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of rural confined spaces based on multi-source data fusion, characterized in that, Includes the following steps: Collect data on methane and hydrogen sulfide concentrations, temperature, liquid level, and AI camera electronic fence monitoring and human recognition results to form a standardized multi-source monitoring dataset; The standardized multi-source monitoring dataset is encrypted with AES-256 using the TCP / IP+MQTT protocol stack and uploaded with breakpoint resume capability to form a secure data transmission stream. Securely transmit data streams to the InfluxDB time series database and build historical time series data to form a searchable multi-source historical dataset. Risk trend analysis is performed on multi-source historical datasets using a dual-branch long short-term memory network. Evidence credibility configuration is formed based on cross-branch consistency gating units. Multi-source data are fused using DS evidence theory to obtain fused risk assessment results. The risk trend analysis results are used as prior constraints to generate risk level adjudication results. Based on the three-level triggering conditions in the alarm rule base, the risk level adjudication results generate graded alarm events, and the workflow engine of the back-end management subsystem forms work order-instruction association results; The system executes remote start / stop of ventilation equipment, voice alarms, and a three-tiered response mechanism (village-town-district) and a tiered timeout escalation mechanism for alarm work orders and handling instructions. It collects handling feedback from the mobile application subsystem and intelligently verifies the handling results through the back-end management subsystem before writing them back to the InfluxDB time-series database and the electronic file management of hazard sources, forming a closed-loop regulatory result set.

2. The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to claim 1, characterized in that, The generation of the standardized multi-source monitoring dataset specifically includes: Intelligent sensing units are deployed within the rural confined space monitoring area, and each sensing unit is synchronized in a unified time and calibrated in its installation location to form a sensing baseline state. The intelligent sensing unit includes a methane sensing unit, a hydrogen sulfide sensing unit, a temperature sensing unit, a liquid level sensing unit, and an AI camera sensing unit. The raw monitoring data of methane concentration and hydrogen sulfide concentration are obtained through the methane sensing unit and the hydrogen sulfide sensing unit, and range consistency correction and zero point offset compensation are performed to generate gas concentration monitoring results. The temperature of the internal environment of the confined space is periodically collected by the temperature sensing unit, and abnormal jitter suppression and time window smoothing are performed to form temperature monitoring results. The liquid level is detected in real time within a confined space by a liquid level sensing unit. The detection results are then processed to a uniform scale based on the installation reference height of the liquid level sensing unit to generate liquid level monitoring results. The AI ​​camera sensing unit collects video of the surrounding and internal areas of a limited space, sets up an electronic fence monitoring area, performs human recognition processing on targets entering the electronic fence monitoring area, and generates personnel intrusion recognition results. Perform unified timestamp alignment and data structure encapsulation on gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results to form a multi-source monitoring dataset containing multiple monitoring dimensions. Perform data integrity verification and validity marking on the multi-source monitoring dataset, remove missing or abnormal monitoring data items, and generate a standardized multi-source monitoring dataset.

3. The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to claim 1, characterized in that, The generation of the secure transmission data stream specifically includes: encapsulating a standardized multi-source monitoring dataset based on the monitoring point identifier and data type; constructing a communication session using TCP / IP and MQTT protocols; adaptively selecting the transmission mode between fourth-generation mobile communication and narrowband IoT based on the current communication signal quality; enabling LoRa self-organizing network for data forwarding in areas without network coverage; performing AES-256 encryption on the data to be transmitted and attaching integrity verification information; recording and controlling the data transmission status through a breakpoint resume mechanism; updating the status of the transmitted data based on the confirmation receipt in the communication session; forming continuously confirmed upload results; and aggregating them into a secure transmission data stream in chronological order.

4. The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to claim 1, characterized in that, The generation of the multi-source historical dataset specifically includes: performing decapsulation and integrity verification on the securely transmitted data stream, restoring the gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results contained therein, performing structured parsing and sequential organization of various monitoring results based on monitoring point identifiers and timestamps, and writing the parsed monitoring records into the InfluxDB time-series database. During the writing process, a time-series storage structure is established according to the monitoring point dimension and monitoring type to form continuously covered historical time-series data. Based on the historical time-series data, a multi-source historical dataset supporting time alignment and conditional retrieval is constructed.

5. The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to claim 1, characterized in that, The generation of the risk level determination result specifically includes: Extract gas concentration monitoring result sequences, temperature monitoring result sequences, liquid level monitoring result sequences, and personnel intrusion identification result sequences that are aligned to timestamps from multi-source historical datasets according to monitoring point identifiers. Slice and resample each sequence according to preset time windows to form a windowed time series sample set. The windowed time series sample set is divided into monitoring dimensions to construct the input of a dual-branch long short-term memory network. The gas concentration monitoring result sequence is used as the gas branch input and the temperature monitoring result sequence is used as the temperature branch input. Time series encoding and memory state update are performed to generate gas branch trend representation and temperature branch trend representation. In the cross-branch consistency gating unit, consistency calculation and gating mapping are performed on the trend representations of the gas branch and the trend representations of the temperature branch, the gating results are output and mapped to the evidence credibility configuration; Risk trend analysis results are generated based on gas branch trend characterization and temperature branch trend characterization. Based on the credibility of the evidence, evidence is constructed from gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, personnel intrusion identification results, and risk trend analysis results. Evidence allocation functions are generated for each, and the results are input into the DS evidence theory fusion process to obtain the fusion risk assessment results. The DS evidence theory fusion process calculates the support combination result of the two evidences on the risk level value set for any two evidence allocation functions, calculates the conflict degree formed by the incompatible parts of the two evidences, normalizes the conflict degree, normalizes and synthesizes each pair of support combination results to obtain the fusion allocation function of the two evidences, and iterates and fuses the fusion allocation function with the other evidence allocation functions one by one according to the same rules to obtain the final fusion allocation function covering all evidence allocation functions. The fusion risk assessment result is determined by the final fusion allocation function. The risk trend analysis results are used as a priori constraints to perform consistency adjudication and level correction on the integrated risk assessment results, thereby generating risk level adjudication results. The level correction limits the upward adjustment of the risk level of the fusion risk assessment result when the risk trend analysis result indicates a continuously rising state, limits the maintenance of the risk level of the fusion risk assessment result when the risk trend analysis result indicates a stable state, and limits the delayed confirmation of the risk level of the fusion risk assessment result when the risk trend analysis result indicates an abnormal fluctuation state.

6. The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to claim 1, characterized in that, The generation of the work order-instruction association result specifically includes: The risk level determination results are collected based on the monitoring point identification to form a risk level determination record set for a single monitoring point; In the alarm rule base, load the risk level adjudication record set with three-level triggering conditions corresponding to the rural confined space type. The three-level triggering conditions include threshold over-limit triggering conditions, trend abnormality triggering conditions, and multi-source collaborative alarm triggering conditions. Encapsulate the three-level triggering conditions into a set of executable rule entries to form an alarm rule entry set. Based on the alarm rule set, the risk level adjudication record set is matched with the rules one by one. The threshold over-limit judgment result, the trend abnormal judgment result, and the multi-source collaborative alarm judgment result are triggered and aggregated to generate a trigger summary. Based on the trigger summary and the risk level adjudication result, the alarm level, alarm type and alarm time window of the graded alarm event are determined to form a graded alarm event. The workflow engine of the background management subsystem receives hierarchical alarm events, generates alarm work orders for hierarchical alarm events based on monitoring point identifiers and alarm types, and writes them into the pending processing queue to form an alarm work order set. Based on the alarm work order set, a handling instruction is generated, including control instructions for remote start and stop of ventilation equipment, handling task instructions from the mobile application subsystem, and flow instructions for the three-level response of village-town-district. The handling instructions are then bound to the corresponding alarm work orders to form a work order-instruction association result.

7. The intelligent monitoring method for rural confined spaces based on multi-source data fusion according to claim 1, characterized in that, The generation of the closed-loop monitoring result set specifically includes: based on the work order-instruction association result, the handling instructions corresponding to the alarm work orders are uniformly scheduled and issued in the background management subsystem, remote start and stop control of ventilation equipment, voice alarm triggering, and three-level response flow from village to town to district are executed, and handling status and response time are established at each response level. When any level fails to complete the response confirmation within the preset time window, the hierarchical timeout escalation mechanism is triggered, and the alarm work order is automatically escalated and flowed in hierarchical order. During the handling process, the mobile application subsystem collects the gas concentration monitoring results, temperature monitoring results, liquid level monitoring results, and personnel intrusion identification results of the on-site confirmation and verification by the handling personnel, and forms the handling feedback. The background management subsystem performs consistency verification on the handling feedback and the multi-source historical dataset, ventilation equipment execution receipts and voice alarm receipts within the alarm time window, generates the handling verification result, and writes it back to the InfluxDB time series database and the electronic file management of hazardous sources to form the closed-loop monitoring result set.

8. A rural confined space intelligent monitoring system based on multi-source data fusion, comprising the rural confined space intelligent monitoring method based on multi-source data fusion as described in any one of claims 1 to 7, characterized in that, include: The intelligent sensing module is used to collect gas concentration data, temperature data, liquid level height, and electronic fence monitoring and human recognition results in limited spaces in rural areas, forming a standardized multi-source monitoring dataset; The data transmission module is used to encrypt and encapsulate standardized multi-source monitoring datasets based on the TCP / IP and MQTT protocol stack, and upload them with breakpoint resumption to form a secure data transmission stream; The data platform module is used to receive securely transmitted data streams and write them into the InfluxDB time series database to establish continuously covered historical time series data and form a searchable multi-source historical dataset. The risk analysis module is used to construct a dual-branch long short-term memory network based on multi-source historical datasets, perform risk trend analysis, form evidence credibility configuration through cross-branch consistency gating units, fuse multi-source data using DS evidence theory, and generate fused risk assessment results and risk level adjudication results. The alarm and work order management module is used to generate graded alarm events based on the three-level trigger conditions of the alarm rule base, and form work order-instruction association results through the workflow engine of the background management subsystem. The emergency response and closed-loop monitoring module is used to execute remote start / stop of ventilation equipment, voice alarms, village-town-district three-level response and graded timeout escalation mechanism, collect the response feedback from the mobile application subsystem and perform intelligent verification of the response results, write back to the time-series database and the electronic file management of hazard sources to form a closed-loop monitoring result set.

Images