Fire-fighting facility management system based on multi-dimensional perception and monitoring method thereof

Abstract

The application discloses a fire-fighting facility management system based on multi-dimensional perception and a monitoring method thereof, and relates to the technical field of fire-fighting facility management. The fire-fighting facility management system based on multi-dimensional perception comprises a video perception unit, a multi-source data acquisition unit, a digital twin construction unit, an inference engine and an adaptive feedback unit. The video perception unit is arranged in the associated visual field area of the fire-fighting facility and is used for continuously acquiring video streams of appearance and environment. The multi-source data acquisition unit is used for acquiring the running parameters, sensing data and maintenance operation logs of the fire-fighting facility in real time. The digital twin construction unit is used for creating a virtual image for each fire-fighting facility, integrating the static attributes, installation position topological relationship of the corresponding fire-fighting facility, and constructing the dynamic state mapping relationship with the physical entity. The inference engine is connected with the video perception unit, the multi-source data acquisition unit and the digital twin construction unit respectively. The adaptive feedback unit dynamically adjusts the inspection plan, generates differentiated maintenance work orders and pushes the early warning information to the specified terminal according to the result output by the inference engine. Through multi-dimensional data fusion, the fire-fighting facility state judgment accuracy is improved.

Description

Technical Field

[0001] This invention relates to the technical field of fire protection facility management, and in particular to a fire protection facility management system and its monitoring method based on multi-dimensional perception. Background Technology

[0002] As a critical infrastructure for protecting people's lives and property, the stability and reliability of fire protection facilities directly affect the effectiveness of emergency response in the event of fires and other emergencies. However, traditional fire protection facility management models have many drawbacks: on the one hand, relying on manual inspections is inefficient, highly susceptible to human factors, and makes it difficult to achieve real-time and comprehensive monitoring of fire protection facilities, easily overlooking potential safety hazards; on the other hand, existing monitoring systems mostly use single data source collection and analysis methods, which can only obtain partial status information of fire protection facilities and cannot comprehensively consider the multi-dimensional status of facilities such as appearance and internal performance, resulting in insufficient accuracy in judging the health status of facilities. Summary of the Invention

[0003] Therefore, the purpose of this invention is to provide a management system and its monitoring method for achieving refined management of fire protection facilities through technologies such as multi-dimensional perception, digital twins, intelligent reasoning and adaptive feedback.

[0004] The technical solution adopted by this invention to solve its technical problem is as follows: A fire protection facility management system based on multi-dimensional perception includes: The video sensing unit is deployed in the associated field of view area of ​​the fire protection facility to continuously acquire high-definition video streams of its appearance and corresponding environment; The multi-source data acquisition unit is used to acquire the operating parameters, sensor data, and maintenance operation logs of fire protection facilities in real time. A digital twin construction unit is used to create a virtual image for each fire protection facility. The virtual image integrates the static attributes and installation location topology of the corresponding fire protection facility and constructs a dynamic state mapping relationship with the physical entity. The inference engine is connected to the video perception unit, the multi-source data acquisition unit, and the digital twin construction unit, respectively. The adaptive feedback unit dynamically adjusts the inspection plan, generates differentiated maintenance work orders, and pushes early warning information to designated terminals based on the results output by the inference engine. The inference engine is configured as follows: A. Perform time-series analysis on high-definition video streams to extract the first sequence features of fire protection facility appearance integrity, signage clarity, physical obstruction, and human operation behavior; B. The operating parameters, sensor data, and maintenance operation logs of fire protection facilities are spatiotemporally aligned and standardized to generate a second sequence of features characterizing the intrinsic performance of the facilities; C. The first sequence features and the second sequence features are synchronously injected into the digital twin of the corresponding fire protection facility, and multimodal feature fusion and contradiction detection are performed based on the predefined health status model and state consistency map, wherein the state consistency map defines the physical or logical constraint relationship between different modal state parameters under normal working conditions. D. When a contradiction is detected or any feature sequence deviates from the preset benchmark range, a root cause analysis based on topological relationships and historical state change data is triggered, generating an assessment report that includes anomaly confidence, suspected root causes, and transmission paths.

[0005] In a preferred embodiment of the present invention, when the inference engine executes step C, it is configured to: calculate dynamic confidence scores for various types of data in the first sequence features and the second sequence features respectively, wherein the confidence scores are determined based on the health status of the data source, real-time data quality, and historical accuracy; and when features of different modalities conflict in their determination of the same facility status, a conflict arbitration mechanism based on the dynamic confidence scores is activated.

[0006] In a preferred embodiment of the present invention, the inference engine, when executing step D, is configured to execute a predefined cross-validation process, the process including at least any of the following scenarios: Scenario 1: When the video sensing unit detects that the pressure gauge pointer of the fire extinguisher is in the green zone, and the corresponding pressure sensor data is continuously lower than the safety threshold, it queries the historical calibration data of the pressure sensor, retrieves high frame rate video clips to analyze the dial status, and associates it with recent maintenance logs. Second scenario: When the video sensing unit detects that the fire hydrant box door is closed, and the corresponding door magnetic sensor returns an open signal or the water pressure sensor returns a zero pressure signal, the video history is called to analyze the sudden change in the door status and lock details, and the data of the regional water supply network is associated. Third scenario: When the video sensing unit detects that the emergency lighting fixtures are not lit in a dark environment or during the test period, and the corresponding power sensors report that the battery voltage and current data are normal, it retrieves the video and current curves from historical power outage events for comparison, and remotely queries or triggers self-test command response data.

[0007] In a preferred embodiment of the invention, a risk transmission simulator is further included, which is configured as follows: a. Based on the topological relationships in the digital twin, fire protection facilities are abstracted as nodes, and the physical connections, functional dependencies, or information flow relationships between fire protection facilities are abstracted as directed edges. At the same time, each edge is assigned a dependency strength and an influence propagation delay weight to construct a directed graph model. b. When a single facility is diagnosed as being in an abnormal state, Monte Carlo simulation or deterministic propagation calculation is performed on the directed graph model, taking the node corresponding to the fire protection facility as the source, to deduce the path, probability and time series of the abnormal state spreading to the associated nodes along each edge; c. Based on the simulation results, output the systemic risk level and a list of critical dependent facilities, wherein the list of critical dependent fire protection facilities includes at least the top N fire protection facilities with the highest probability of being affected and the main paths of their impact.

[0008] In a preferred embodiment of the present invention, the condition for triggering the risk transmission simulator must satisfy at least one of the following: (1) One or more of the first sequence features or the second sequence features continuously deviate from their preset benchmark threshold range; (2) In step C, the multimodal feature fusion and contradiction detection determine that there is a fundamental conflict in the description of the same state by different data sources that cannot be resolved by confidence arbitration; (3) The early performance degradation warning level output by the health status model exceeds the preset risk threshold.

[0009] In a preferred embodiment of the present invention, the adaptive feedback unit activates an enhanced monitoring mode for facilities in the list of critical dependent facilities based on the list of critical dependent facilities. The enhanced monitoring mode includes at least one of increasing the data acquisition frequency, increasing the video analysis frequency, or temporarily activating a backup sensor channel.

[0010] In a preferred embodiment of the present invention, the adaptive feedback unit maintains a facility health risk profile database, the risk profile being dynamically updated based on the facility's recent abnormal history, its criticality in risk transmission, and its predicted remaining lifespan; the dynamic adjustment of the inspection plan specifically involves: based on the risk profile, automatically shortening the fixed inspection cycle for high-risk facilities or inserting them into a temporary key inspection queue.

[0011] In a preferred embodiment of the present invention, the health status model is a machine learning-based predictive model, which continuously learns from historical data to predict the remaining service life of key components of fire protection facilities and identify the progressive degradation mode of status parameters, and issues an early warning before the parameters actually exceed the standard.

[0012] The present invention also provides a method for monitoring the status of fire protection facilities, based on the above-described system, comprising the following steps: S1: Parallel acquisition of video stream data and non-video multidimensional status data of the target fire protection facilities; S2: In the digital twin of fire protection facilities, perform spatiotemporal synchronization and alignment of the two types of data; S3: Extract the visually observable state set from the video stream and the intrinsically measurable state set from the non-video data respectively; S4: Apply the preset conflict determination rule set and correlation analysis model to perform fusion analysis and contradiction diagnosis on the two state sets; S5: When a contradiction or anomaly is diagnosed, combine facility topology and historical data to conduct root cause reasoning and risk assessment, and generate structured diagnostic conclusions. S6: Automatically trigger differentiated response strategies based on the level and type of the diagnostic conclusion.

[0013] In a preferred embodiment of the present invention, step S4 further includes: S41. Confidence assessment: Calculate dynamic confidence for each data point in the visually observable state set and the intrinsically measurable state set, wherein the confidence is determined based on the health status of the data source, real-time data quality, and historical diagnostic accuracy. S42. Consistency Map Verification: Compare two state sets with a predefined state consistency map, which defines the physical or logical constraints that should be satisfied between different modal parameters, and identify and record all violated constraints. S43. Conflict Arbitration and Fusion: When two state sets conflict in their description of the same fire protection facility status, the one with higher confidence level shall be adopted first, or a weighted calculation based on confidence level weights shall be performed to generate a fused comprehensive status judgment. Step S5 also includes: S51. Scenario-based cross-validation: Based on the type of contradiction, invoke the preset verification process corresponding to the scenario, wherein the process includes the verification steps in the first scenario, second scenario or third scenario as described in claim 3; S52. Risk Transmission Simulation: Based on the digital twin topology of the facility, construct or simulate the potential impact paths, probabilities, and timing of abnormal states on associated fire protection systems using a directed graph model, and assess the systemic risk level; S53. Structured Report Generation: Generate structured diagnostic conclusions that include outliers, confidence levels, causes of conflict, verification evidence, risk levels, and a list of critical dependent facilities; Step S6 also includes: S61. Strategy Matching: Based on the anomaly type, risk level, and list of critical dependent facilities in the diagnostic conclusion, match at least one response strategy from the pre-set strategy library; S62. Strategy Execution: The response strategies include: pushing differentiated early warning information containing a visual chain of evidence to managers at different levels, generating differentiated maintenance work orders containing specific maintenance verification steps, or dynamically adjusting the inspection frequency and monitoring level of affected facilities based on risk assessment results.

[0014] The beneficial effects of this invention are: 1. Multi-dimensional data fusion improves the accuracy of status judgment: Through video perception unit and multi-source data acquisition unit, the appearance status and internal operating parameters of fire protection facilities are fully acquired. Combined with dynamic confidence assessment and conflict arbitration mechanism, the problem of judgment bias from a single data source is effectively solved, ensuring the accuracy and reliability of facility status judgment.

[0015] 2. Precise anomaly diagnosis and root cause localization: Based on the topological relationships and historical data of digital twins, combined with scenario-based cross-validation, it can quickly and accurately diagnose data contradictions and abnormal states, locate the root cause, avoid blind investigation, and improve the efficiency of anomaly handling.

[0016] 3. Systemic risk prediction and reduction of safety hazards: The risk transmission simulator can predict the transmission impact of a single facility anomaly on the entire fire protection system, identify the risk level and key dependent facilities, and facilitate managers to take preventive measures in advance to reduce systemic risks and avoid safety accidents. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the architecture of the management system of this invention; Figure 2 This is a schematic diagram of the workflow of the inference engine in this invention; Figure 3 This is a schematic diagram of the risk transmission simulation in this invention; Figure 4 This is a flowchart of the monitoring method of the present invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will now be clearly and completely described in conjunction with the accompanying drawings.

[0019] Reference Figures 1 to 3 A fire protection facility management system based on multi-dimensional perception includes: The video sensing unit, deployed within the associated field of view of the fire protection facilities, continuously acquires high-definition video streams of their appearance and corresponding environment. Specifically, the video sensing unit employs an ultra-high-definition network camera whose lens angle can adaptively adjust according to the type of fire protection facility. For small facilities such as fire extinguishers and emergency lighting fixtures, a corresponding fixed-focus lens is used to ensure clear capture of details such as the position of pressure gauge pointers and the illumination status of lights. For large facilities or regional facilities such as fire hydrants and fire pipeline valves, a zoom lens with a corresponding focal length is used to cover panoramic information such as the overall appearance of the facility, the opening and closing status of the cabinet doors, and the surrounding environmental obstructions. Simultaneously, the camera supports infrared night vision, automatically switching to infrared mode at night or in low-light environments to ensure the clarity of the video stream.

[0020] A multi-source data acquisition unit is used to acquire real-time operating parameters, sensor data, and maintenance operation logs of fire protection facilities. Specifically, the multi-source data acquisition unit includes a sensor module, a data transmission module, and a log recording module. The sensor module is configured with differentiated sensing components according to the type of fire protection facility: for fire extinguishers, pressure sensors and temperature sensors are configured; for fire hydrants, water pressure sensors, door magnetic sensors, and flow sensors are configured; for emergency lighting fixtures, voltage sensors, current sensors, and battery capacity sensors are configured. The data transmission module supports multiple communication methods such as 5G, LoRa, and Ethernet, and automatically switches according to the network coverage of the deployment environment. The log recording module is linked with the maintenance management platform to automatically record maintenance personnel information, maintenance time, maintenance content (such as pressure replenishment, component replacement, cleaning and maintenance, etc.), post-maintenance test data, etc. The log data is stored in an encrypted manner, making it tamper-proof and ensuring the traceability of maintenance operations.

[0021] A digital twin building unit is used to create a virtual image for each fire protection facility. This virtual image integrates the static attributes and installation location topology of the corresponding fire protection facility, and constructs a dynamic state mapping relationship with the physical entity. Specifically, the digital twin building unit uses a combination of BIM and GIS technologies to construct the virtual image. Static attribute integration includes information such as the fire protection facility's model and specifications, manufacturer, manufacturing date, material parameters, design service life, installation time, latitude and longitude coordinates of the installation location, and the building floor and area. Through structured data storage and visual modeling, the static information of the facility is presented intuitively. When constructing the installation location topology, a topology graph algorithm is used to associate and model fire protection facilities within the same building and area according to physical connections (such as the connection between fire hydrants and fire pipe networks), functional dependencies (such as the linkage dependency between emergency lighting fixtures and fire alarm systems), and information flow transmission (such as the transmission path between sensor data and the monitoring platform), forming a visual topology graph. The dynamic state mapping relationship is achieved through a real-time data synchronization mechanism. The digital twin receives data uploaded by the video sensing unit and the multi-source data acquisition unit every 100ms, mapping the dynamic information such as the physical entity's appearance and operating parameters to the virtual image in real time. Users can view the real-time status of the facility through a 3D visualization interface, supporting operations such as zooming, rotating, and sectioning, facilitating a clear understanding of facility details. This integrates the static attributes and dynamic status of fire protection facilities, achieving a digital replication of the physical entity; through topological relationship modeling and visualization, the relationships between facilities are readily apparent.

[0022] The inference engine is connected to the video perception unit, the multi-source data acquisition unit, and the digital twin construction unit, respectively.

[0023] The inference engine is configured as follows: A. Perform time-series analysis on the high-definition video stream to extract first-sequence features of fire-fighting facility appearance integrity, signage clarity, physical obstruction, and human operation behavior. Specifically, the inference engine uses a deep learning model combining CNN and LSTM to perform time-series analysis on the high-definition video stream. Regarding appearance integrity extraction, the model compares the current video frame with a standard image of the facility's initial state to identify whether the facility is damaged (e.g., deformed fire hydrant box door, dented fire extinguisher bottle) or missing components (e.g., missing fire hydrant interface, damaged emergency lighting fixture cover). For signage clarity, the model analyzes the pixel contrast and text recognition of the signage area to determine whether the signage is blurry, faded, or obstructed. When the text recognition is below 80%, the signage clarity is deemed substandard. In physical occlusion identification, the model analyzes the temporal changes of 10 consecutive video frames to determine whether a facility is permanently occluded (e.g., by goods or billboards) or temporarily occupies (e.g., by temporary personnel presence). Physical occlusion is defined as the occluded area occupying more than 30% of the total facility area and lasting for more than 5 minutes. For human operation behavior identification, the model uses a human posture recognition algorithm to identify whether there are unauthorized operations (e.g., unauthorized dismantling of fire protection facilities or malicious damage) or normal maintenance operations. The response time for identifying unauthorized operations is ≤2 seconds, and the video segment is automatically marked as evidence. Through temporal analysis using a deep learning model, the model can accurately and efficiently extract the first-sequence features related to the appearance and environment of fire protection facilities. Compared to traditional image recognition algorithms, it has higher recognition accuracy and anti-interference capabilities, effectively distinguishing between normal and abnormal states, and providing reliable visual feature support for facility health diagnosis.

[0024] B. The operating parameters, sensor data, and maintenance logs of fire protection facilities are spatiotemporally aligned and standardized to generate a second sequence of features characterizing the facility's intrinsic performance. Specifically, during spatiotemporal alignment, the inference engine uses GPS timestamps as a benchmark to synchronize video frame data collected by the video perception unit with operating parameters and sensor data acquired by the multi-source data acquisition unit. For spatial alignment, based on the installation location topology in the digital twin, various types of data for the same fire protection facility are associated with corresponding virtual image nodes to ensure accurate data-facility correspondence. During standardization, the Z-score standardization method is used to convert operating parameters and sensor data (such as pressure, temperature, and voltage) with different dimensions and value ranges into standardized data with a mean of 0 and a standard deviation of 1. For maintenance operation logs, Natural Language Processing (NLP) technology is used to convert unstructured log text (such as "Replenish fire extinguisher pressure to 1.2MPa on May 20, 2024") into structured data (maintenance time: 2024-05-20, maintenance type: pressure replenishment, post-maintenance parameter: pressure 1.2MPa). The second sequence features include real-time values, rates of change, fluctuation amplitudes, deviations from historical data for the same period, maintenance cycles, and post-maintenance performance recovery levels of facility operating parameters, comprehensively characterizing the facility's intrinsic operating performance. This spatiotemporal alignment ensures the correlation and consistency of data from different sources, avoiding analytical errors caused by data asynchrony or inconsistency. Furthermore, standardization eliminates the impact of differences in data dimensions, making various types of data comparable. Furthermore, the multi-dimensional construction of the second sequence features can comprehensively and objectively reflect the intrinsic performance status of fire protection facilities, laying the foundation for subsequent multi-modal feature fusion analysis.

[0025] C. The first sequence features and the second sequence features are simultaneously injected into the digital twin of the corresponding fire protection facility, and multimodal feature fusion and contradiction detection are performed based on a predefined health status model and state consistency map. Specifically, the predefined state consistency map is constructed based on the physical principles, design specifications, and actual operating experience of the fire protection facility. For example, for fire extinguishers, the state consistency map stipulates that "the pressure gauge pointer is in the green area (0.8-1.2MPa) as perceived by the video" and "the pressure sensor data is within the range of 0.8-1.2MPa" are logically consistent; "the fire extinguisher bottle is undamaged as perceived by the video" and "the pressure sensor data does not drop suddenly" are physically constrained. Multimodal feature fusion employs a combination of attention mechanism and weighted summation. The inference engine calculates dynamic confidence scores for each type of data in the first and second sequence features. The confidence scores are determined based on the health status of the data source (e.g., whether the sensor is within the calibration validity period and whether the camera is working properly), real-time data quality (e.g., data transmission error rate and video clarity), and historical accuracy (e.g., historical sensor measurement error and historical video recognition accuracy). The confidence score ranges from 0 to 1, with higher values ​​indicating stronger data reliability. When different modalities conflict in their assessment of the same facility's status, a conflict arbitration mechanism based on dynamic confidence weighting is activated. For example, if the video sensing unit identifies the fire extinguisher pressure gauge pointer as being in the green zone (confidence 0.9), while the pressure sensor data shows 0.6 MPa (confidence 0.8), the weighted result is calculated as (0.9 × 1 + 0.8 × 0) / (0.9 + 0.8) ≈ 0.53 (where 1 represents a normal state and 0 represents an abnormal state). Combining this with the constraints of the state consistency map, a data contradiction is identified, and further verification is initiated. This state consistency map provides a clear basis for multimodal data fusion and contradiction detection, ensuring the logic and rationality of the analysis. Simultaneously, the introduction of dynamic confidence and the conflict arbitration mechanism effectively resolves contradictions between different data sources, improves the accuracy of status assessment, and avoids misjudgments caused by the failure or error of a single data source.

[0026] D. When a contradiction or any feature sequence deviates from the preset baseline range, a root cause analysis based on topological relationships and historical state change data is triggered, generating an assessment report including anomaly confidence, suspected root causes, and transmission paths. Specifically, the preset baseline range is determined based on the design standards, industry specifications, and historical operating data of fire protection facilities. For example, the baseline range for fire extinguisher pressure is 0.8-1.2 MPa, and the baseline range for emergency lighting battery voltage is 12-14V. When a feature sequence exceeds this range and lasts for more than 30 seconds, it is determined to deviate from the baseline range. During the root cause analysis, the inference engine first checks the status of facilities associated with the abnormal facility based on the topological relationships of the digital twin (e.g., when the fire hydrant water pressure is abnormal, check the status of the associated water supply network valves); at the same time, it retrieves the historical state change data of the facility for the past year, analyzing parameter change trends (e.g., whether the pressure is gradually decreasing), historical fault records, maintenance records, and other information. For example, when video detects that an emergency lighting fixture is not emitting light in a dark environment (first-sequence feature anomaly), while the power sensor reports normal battery voltage and current data (second-sequence feature normal), the inference engine retrieves the fixture's maintenance records for the past six months. If there are records of untimely lampshade cleaning, combined with comparative analysis of video and current curves from historical power outage events, the suspected root cause is determined to be dust accumulation on the lampshade preventing light from passing through. If the maintenance records show that the bulb was recently replaced, and the current curve differs from when it is emitting light normally, the suspected root cause is poor bulb contact. The anomaly confidence level in the assessment report is calculated comprehensively based on the degree of feature deviation, data confidence level, and matching degree of similar historical cases. Suspected root causes are ranked by probability, and the transmission path clearly identifies the related facilities that the abnormal state may affect and the propagation sequence. This root cause analysis based on topological relationships and historical data can accurately locate the root cause of anomalies or contradictions, avoiding blind investigation. The assessment report, containing multi-dimensional information, provides managers with comprehensive and clear anomaly analysis results, facilitating the rapid development of targeted handling solutions.

[0027] This solution also includes a risk transmission simulator, which is configured as follows: a. Based on the topological relationships in the digital twin, fire protection facilities are abstracted as nodes, and the physical connections, functional dependencies, or information flow relationships between fire protection facilities are abstracted as directed edges. Each edge is assigned a dependency strength and an impact propagation delay weight to construct a directed graph model. Specifically, during node abstraction, each fire protection facility node is assigned a unique identifier, associated with its type, importance, and operational status. Importance is categorized into three levels based on the facility's role in the fire protection system: core nodes (e.g., fire pumps, main fire hydrants), important nodes (e.g., area fire hydrants, emergency lighting main controllers), and ordinary nodes (e.g., individual emergency lighting fixtures, small fire extinguishers). The dependency strength of directed edges is determined based on the closeness of the connections between facilities, using a 1-10 scoring mechanism: in physical connections, the dependency strength between fire pumps and the main fire protection network is 10, and the dependency strength between area fire hydrants and the main fire protection network is 8; in functional dependencies, the dependency strength between emergency lighting fixtures and emergency power supplies is 9, and the dependency strength between fire alarm detectors and fire controllers is 9; in information flow relationships, the dependency strength between sensors and the inference engine is 7. The weights for transmission delays are determined based on data transmission speed and the time it takes for physical actions to propagate. For example, the weight for electrical signal transmission delay is 0.1 seconds, for water pressure transmission delay it is 1 second, and for mechanical structure linkage delay it is 0.5 seconds. The directed graph model is presented visually, with nodes distinguished by different colors to indicate their importance, and edge thickness corresponding to dependency strength, facilitating intuitive viewing. Through the abstraction and weighting of nodes and directed edges, the constructed directed graph model accurately reflects the relationships and influence characteristics between fire protection facilities. Simultaneously, the visual presentation improves the model's usability, making it easier for managers to understand the dependency logic between facilities.

[0028] b. When a single facility is diagnosed as abnormal, Monte Carlo simulation or deterministic propagation calculation is performed on the directed graph model, taking the node corresponding to that fire protection facility as the source, to deduce the path, probability, and time series of the abnormal state spreading to associated nodes along each edge. Specifically, the simulation calculation method is selected according to the type of abnormality: for abnormalities with high uncertainty (such as those caused by sensor data drift), Monte Carlo simulation is used, with 1000 simulations. Each simulation randomly generates an impact path based on dependency strength and propagation delay weights, and the probability of different paths occurring is statistically analyzed; for abnormalities with high determinism (such as water pressure drop caused by fire pump failure), deterministic propagation calculation is used to determine the impact path based on physical laws and logical relationships. For example, when a fire pump (a core node) malfunctions and cannot supply water, deterministic propagation calculations show that: the main fire pipeline water pressure drops after 1 second, the regional fire hydrant water pressure falls below the safety threshold after 2 seconds, and the fire extinguisher refilling system, which relies on the regional fire hydrant for water supply, cannot replenish pressure normally after 5 seconds, with a 100% probability of impact in all cases. If a regional fire hydrant (an important node) experiences abnormal water pressure due to valve damage, Monte Carlo simulations show that: the emergency sprinkler system in that region is 95% affected, the adjacent regional fire hydrants are 30% affected, and the fire pump is 5% affected, with time series of different impact paths output (e.g., the emergency sprinkler system experiences insufficient water pressure after 3 seconds, and the adjacent regional fire hydrants experience water pressure fluctuations after 10 seconds). This allows for the selection of appropriate simulation calculation methods for different types of anomalies, improving the accuracy of risk propagation simulations; the detailed path, probability, and time series outputs provide managers with a scientific basis for predicting the scope and development trend of risk impacts.

[0029] c. Based on the simulation results, output the systemic risk level and a list of critical dependent facilities. The list of critical dependent fire protection facilities must include at least the top N fire protection facilities with the highest probability of being affected and their primary paths of impact. Specifically, the systemic risk level is comprehensively assessed based on the number, importance, and scope of affected facilities, and is divided into four levels: Level 1 (extremely high risk), Level 2 (high risk), Level 3 (medium risk), and Level 4 (low risk). Level 1 risk occurs when a core node is abnormal and affects more than 50% of important nodes; Level 2 risk occurs when an important node is abnormal and affects more than 30% of ordinary nodes; Level 3 risk occurs when an ordinary node is abnormal and the impact is limited to a single area; and Level 4 risk occurs when the abnormality only affects itself and no related facilities are affected. The N value in the critical dependent facilities list can be flexibly set according to actual management needs (default N=10). The list details the name, type, location, probability of being affected, and main cause path of each facility (e.g., "Fire hydrant A is affected, cause path is fire pump failure → main fire pipeline water pressure drop → insufficient water pressure in fire hydrant A"), and marks the importance of each facility, making it easier for managers to prioritize high-priority facilities. This clearly defined systemic risk level provides managers with a standard for judging the urgency of risks. The critical dependent facilities list allows managers to quickly identify key areas of concern and take targeted prevention and control measures to reduce losses caused by the spread of risks.

[0030] In this scheme, the conditions for triggering the risk transmission simulator must meet at least one of the following: (1) One or more of the first sequence features or the second sequence features continuously deviate from their preset benchmark threshold range; (2) In step C, the multimodal feature fusion and contradiction detection determine that there is a fundamental conflict in the description of the same state by different data sources that cannot be resolved by confidence arbitration; (3) The early performance degradation warning level output by the health status model exceeds the preset risk threshold.

[0031] For condition (1), the criterion for "continuous deviation" is that the duration of the feature sequence deviating from the baseline threshold range exceeds 1 minute, and the deviation exceeds 20% of the baseline value, thus avoiding unnecessary simulations triggered by instantaneous fluctuations. For condition (2), a fundamental conflict refers to an irreconcilable contradiction in the judgment results of different data sources after dynamic confidence weighted arbitration, and the contradiction lasts for more than 30 seconds. For example, the video sensing unit continuously identifies that the fire hydrant box door is closed for 30 consecutive seconds (confidence 0.95), while the door magnetic sensor continuously feeds back an open signal (confidence 0.9). Even after weighted arbitration, the true state cannot be determined, and it is judged as a fundamental conflict. For condition (3), the early performance degradation warning level of the health status model is divided into three levels: mild, moderate, and severe. The preset risk threshold is moderate. When the model outputs a severe warning, the risk transmission simulator is triggered. This ensures that the risk transmission simulator is only activated when there is a real potential systemic risk, thus avoiding resource waste and timely capturing of key risk points.

[0032] This solution also includes an adaptive feedback unit that dynamically adjusts the inspection plan, generates differentiated maintenance work orders, and pushes early warning information to designated terminals based on the results output by the inference engine.

[0033] The adaptive feedback unit has a built-in processor and storage module. The storage module maintains a facility health risk profile library and a pre-set strategy library. The facility health risk profile library is dynamically updated based on the facility's recent anomaly history (number of anomalies, anomaly type, and anomaly duration in the past 3 months), its criticality in risk transmission (whether it is a core node, an important node, and the probability of being affected), and its predicted remaining lifespan. Each facility's risk profile is divided into three levels: high risk, medium risk, and low risk, according to its risk score. When dynamically adjusting the inspection plan, for high-risk facilities, the original fixed inspection cycle (e.g., once a month) is shortened to once every half month, or it is directly inserted into the temporary key inspection queue to ensure that the inspection is completed within 72 hours; for medium-risk facilities, the original inspection cycle remains unchanged, but inspection items are increased (e.g., sensor data calibration and appearance detail inspection); for low-risk facilities, the original inspection plan is maintained.

[0034] When generating differentiated maintenance work orders, targeted work orders are generated based on the anomaly type, suspected root cause, and facility type in the assessment report. For example, for anomalies of insufficient fire extinguisher pressure, the work order clearly indicates the pressure value to be replenished, the operating steps (such as closing the valve, connecting the inflation equipment, slowly inflating to the standard pressure, and checking the seal), the required tools (inflation pump, pressure gauge), and precautions (avoiding over-pressure inflation). For anomalies of deformed fire hydrant box doors, the work order details the repair process (such as disassembling the box door, correcting the deformed parts, replacing the damaged hinges, and installation and debugging) and the repair time limit (within 48 hours).

[0035] When pushing early warning information to designated terminals, the target audience and method are determined based on the risk level and anomaly type: Level 1 risks are pushed to enterprise leaders, fire safety managers, and on-site maintenance personnel via SMS, app notifications, and telephone voice reminders; Level 2 risks are pushed to fire safety managers and on-site maintenance personnel via SMS and app notifications; Level 3 and 4 risks are pushed to on-site maintenance personnel via app notifications. The early warning information includes the name and location of the abnormal facility, anomaly type, risk level, suspected root cause, and a visual evidence chain (such as abnormal video clips or data curve screenshots) to facilitate quick understanding of the situation by recipients.

[0036] In addition, the adaptive feedback unit activates an enhanced monitoring mode for facilities on the list of critical dependent facilities. The enhanced monitoring mode includes at least one of the following: increasing the data acquisition frequency (e.g., increasing the sensor data acquisition frequency from once every 10 seconds to once every 2 seconds), increasing the video analysis frequency (e.g., increasing the video frame analysis frequency from once per frame to three times per five frames), and temporarily activating backup sensing channels (e.g., activating backup pressure sensors) to continuously monitor changes in facility status.

[0037] Reference Figure 4 The present invention also provides a method for monitoring the status of fire protection facilities, based on the above-described system, comprising the following steps: S1: Parallel acquisition of video stream data and non-video multidimensional status data of the target fire protection facilities; S2: In the digital twin of fire protection facilities, perform spatiotemporal synchronization and alignment of the two types of data; S3: Extract the visually observable state set from the video stream and the intrinsically measurable state set from the non-video data respectively; S4: Apply the preset conflict determination rule set and correlation analysis model to perform fusion analysis and contradiction diagnosis on the two state sets; S5: When a contradiction or anomaly is diagnosed, combine facility topology and historical data to conduct root cause reasoning and risk assessment, and generate structured diagnostic conclusions. S6: Automatically trigger differentiated response strategies based on the level and type of the diagnostic conclusion.

[0038] Step S4 further includes: S41. Confidence assessment: Calculate dynamic confidence for each data point in the visually observable state set and the intrinsically measurable state set, wherein the confidence is determined based on the health status of the data source, real-time data quality, and historical diagnostic accuracy. S42. Consistency Map Verification: Compare two state sets with a predefined state consistency map, which defines the physical or logical constraints that should be satisfied between different modal parameters, and identify and record all violated constraints. S43. Conflict Arbitration and Fusion: When two state sets conflict in their descriptions of the same fire protection facility status, the one with higher confidence level is adopted first, or a weighted calculation is performed based on the confidence level weight to generate a fused comprehensive status judgment.

[0039] Step S5 further includes: S51. Scenario-based cross-validation: Based on the type of contradiction, invoke the preset verification process corresponding to the scenario, wherein the process includes the verification steps in the first scenario, second scenario or third scenario as described in claim 3; S52. Risk Transmission Simulation: Based on the digital twin topology of the facility, construct or simulate the potential impact paths, probabilities, and timing of abnormal states on associated fire protection systems using a directed graph model, and assess the systemic risk level; S53. Structured Report Generation: Generate structured diagnostic conclusions that include outliers, confidence levels, causes of conflict, verification evidence, risk levels, and a list of critical dependent facilities.

[0040] Step S6 further includes: S61. Strategy Matching: Based on the anomaly type, risk level, and list of critical dependent facilities in the diagnostic conclusion, match at least one response strategy from the pre-set strategy library; S62. Strategy Execution: The response strategies include: pushing differentiated early warning information containing a visual chain of evidence to managers at different levels, generating differentiated maintenance work orders containing specific maintenance verification steps, or dynamically adjusting the inspection frequency and monitoring level of affected facilities based on risk assessment results.

[0041] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A fire protection facility management system based on multi-dimensional perception, characterized in that, Including: The video sensing unit is deployed in the associated field of view area of ​​the fire protection facility to continuously acquire high-definition video streams of its appearance and corresponding environment; The multi-source data acquisition unit is used to acquire the operating parameters, sensor data, and maintenance operation logs of fire protection facilities in real time. A digital twin construction unit is used to create a virtual image for each fire protection facility. The virtual image integrates the static attributes and installation location topology of the corresponding fire protection facility and constructs a dynamic state mapping relationship with the physical entity. The inference engine is connected to the video perception unit, the multi-source data acquisition unit, and the digital twin construction unit, respectively. The adaptive feedback unit dynamically adjusts the inspection plan, generates differentiated maintenance work orders, and pushes early warning information to designated terminals based on the results output by the inference engine. The inference engine is configured as follows: A. Perform time-series analysis on high-definition video streams to extract the first sequence features of fire protection facility appearance integrity, signage clarity, physical obstruction, and human operation behavior; B. The operating parameters, sensor data, and maintenance operation logs of fire protection facilities are spatiotemporally aligned and standardized to generate a second sequence of features characterizing the intrinsic performance of the facilities; C. The first sequence features and the second sequence features are synchronously injected into the digital twin of the corresponding fire protection facility, and multimodal feature fusion and contradiction detection are performed based on the predefined health status model and state consistency map, wherein the state consistency map defines the physical or logical constraint relationship between different modal state parameters under normal working conditions. D. When a contradiction is detected or any feature sequence deviates from the preset benchmark range, a root cause analysis based on topological relationships and historical state change data is triggered, generating an assessment report that includes anomaly confidence, suspected root causes, and transmission paths.

2. The fire protection facility management system based on multi-dimensional perception according to claim 1, characterized in that, When the inference engine executes step C, it is configured to: calculate dynamic confidence scores for various types of data in the first sequence features and the second sequence features, wherein the confidence scores are determined based on the health status of the data source, real-time data quality, and historical accuracy; and when features of different modalities conflict in their determination of the same facility status, a conflict arbitration mechanism based on the dynamic confidence scores is activated.

3. A fire protection facility management system based on multi-dimensional perception according to claim 1, characterized in that, When executing step D, the inference engine is configured to perform a predefined cross-validation process, which includes at least one of the following scenarios: Scenario 1: When the video sensing unit detects that the pressure gauge pointer of the fire extinguisher is in the green zone, and the corresponding pressure sensor data is continuously lower than the safety threshold, it queries the historical calibration data of the pressure sensor, retrieves high frame rate video clips to analyze the dial status, and associates it with recent maintenance logs. Second scenario: When the video sensing unit detects that the fire hydrant box door is closed, and the corresponding door magnetic sensor returns an open signal or the water pressure sensor returns a zero pressure signal, the video history is called to analyze the sudden change in the door status and lock details, and the data of the regional water supply network is associated. Third scenario: When the video sensing unit detects that the emergency lighting fixtures are not lit in a dark environment or during the test period, and the corresponding power sensors report that the battery voltage and current data are normal, it retrieves the video and current curves from historical power outage events for comparison, and remotely queries or triggers self-test command response data.

4. A fire protection facility management system based on multi-dimensional perception according to claim 3, characterized in that, It also includes a risk transmission simulator, which is configured as follows: a. Based on the topological relationships in the digital twin, fire protection facilities are abstracted as nodes, and the physical connections, functional dependencies, or information flow relationships between fire protection facilities are abstracted as directed edges. At the same time, each edge is assigned a dependency strength and an influence propagation delay weight to construct a directed graph model. b. When a single facility is diagnosed as being in an abnormal state, Monte Carlo simulation or deterministic propagation calculation is performed on the directed graph model, taking the node corresponding to the fire protection facility as the source, to deduce the path, probability and time series of the abnormal state spreading to the associated nodes along each edge; c. Based on the simulation results, output the systemic risk level and a list of critical dependent facilities, wherein the list of critical dependent fire protection facilities includes at least the top N fire protection facilities with the highest probability of being affected and the main paths of their impact.

5. A fire protection facility management system based on multi-dimensional perception according to claim 4, characterized in that, The risk transmission simulator must be triggered if at least one of the following conditions is met: (1) One or more of the first sequence features or the second sequence features continuously deviate from their preset benchmark threshold range; (2) In step C, the multimodal feature fusion and contradiction detection determine that there is a fundamental conflict in the description of the same state by different data sources that cannot be resolved by confidence arbitration; (3) The early performance degradation warning level output by the health status model exceeds the preset risk threshold.

6. A fire protection facility management system based on multi-dimensional perception according to claim 4, characterized in that, The adaptive feedback unit activates an enhanced monitoring mode for the facilities in the list of critical dependent facilities based on the list. The enhanced monitoring mode includes at least one of increasing the data acquisition frequency, increasing the video analysis frequency, or temporarily activating a backup sensor channel.

7. A fire protection facility management system based on multi-dimensional perception according to claim 6, characterized in that, The adaptive feedback unit maintains a facility health risk profile database, which is dynamically updated based on the facility's recent abnormal history, its criticality in risk transmission, and its predicted remaining lifespan. The dynamic adjustment of the inspection plan specifically involves: based on the risk profile, automatically shortening the fixed inspection cycle for high-risk facilities or inserting them into a temporary key inspection queue.

8. A fire protection facility management system based on multi-dimensional perception according to claim 1, characterized in that, The health status model is a machine learning-based predictive model that continuously learns from historical data to predict the remaining service life of key components of fire protection facilities and identify the progressive degradation mode of status parameters, and issues early warnings before the parameters actually exceed the standards.

9. A method for monitoring the status of fire protection facilities, based on the system described in any one of claims 5-8, characterized in that, Includes the following steps: S1: Parallel acquisition of video stream data and non-video multidimensional status data of the target fire protection facilities; S2: In the digital twin of fire protection facilities, perform spatiotemporal synchronization and alignment of the two types of data; S3: Extract the visually observable state set from the video stream and the intrinsically measurable state set from the non-video data respectively; S4: Apply the preset conflict determination rule set and correlation analysis model to perform fusion analysis and contradiction diagnosis on the two state sets; S5: When a contradiction or anomaly is diagnosed, combine facility topology and historical data to conduct root cause reasoning and risk assessment, and generate structured diagnostic conclusions. S6: Automatically trigger differentiated response strategies based on the level and type of the diagnostic conclusion.

10. A method for monitoring the status of fire protection facilities according to claim 9, characterized in that: Step S4 also includes: S41. Confidence assessment: Calculate dynamic confidence for each data point in the visually observable state set and the intrinsically measurable state set, wherein the confidence is determined based on the health status of the data source, real-time data quality, and historical diagnostic accuracy. S42. Consistency Map Verification: Compare two state sets with a predefined state consistency map, which defines the physical or logical constraints that should be satisfied between different modal parameters, and identify and record all violated constraints. S43. Conflict Arbitration and Fusion: When two state sets conflict in their description of the same fire protection facility status, the one with higher confidence level shall be adopted first, or a weighted calculation based on confidence level weights shall be performed to generate a fused comprehensive status judgment. Step S5 also includes: S51. Scenario-based cross-validation: Based on the type of contradiction, invoke the preset verification process corresponding to the scenario, wherein the process includes the verification steps in the first scenario, second scenario or third scenario as described in claim 3; S52. Risk Transmission Simulation: Based on the digital twin topology of the facility, construct or simulate the potential impact paths, probabilities, and timing of abnormal states on associated fire protection systems using a directed graph model, and assess the systemic risk level; S53. Structured Report Generation: Generate structured diagnostic conclusions that include outliers, confidence levels, causes of conflict, verification evidence, risk levels, and a list of critical dependent facilities; Step S6 also includes: S61. Strategy Matching: Based on the anomaly type, risk level, and list of critical dependent facilities in the diagnostic conclusion, match at least one response strategy from the pre-set strategy library; S62. Strategy Execution: The response strategies include: pushing differentiated early warning information containing a visual chain of evidence to managers at different levels, generating differentiated maintenance work orders containing specific maintenance verification steps, or dynamically adjusting the inspection frequency and monitoring level of affected facilities based on risk assessment results.

Images