Intelligent building fire risk assessment method and system based on sensor network

Abstract

The invention relates to the technical field of intelligent building operation and maintenance management, in particular to an intelligent building fire risk assessment method and system based on a sensor network. The method comprises the following steps: continuously acquiring environment information in a building by deploying a multi-source sensor network, and introducing time continuous change modeling and space consistency constraint to convert monitoring data into networked environment state expression; performing risk-oriented reconstruction on the environment state, and extracting a low-dimensional risk feature vector through risk sensitivity analysis, component collaborative change analysis and historical reference contrast; constructing a fire risk space transmission network based on the risk characteristics; multi-time-scale trend prediction is carried out by constructing a unit-level risk time sequence, and the prediction confidence is quantified; the prediction risk is converted into an executable intervention decision, and an active prevention and control closed loop is realized through risk grading, response action generation, strategy optimization and effect feedback. According to the invention, the accuracy and robustness of building fire risk assessment and early warning are improved.

Description

A Sensor Network-Based Intelligent Building Fire Risk Assessment Method and System Technical Field

[0001] This invention relates to the field of intelligent building operation and maintenance management technology, specifically to an intelligent building fire risk assessment method and system based on sensor networks. Background Technology

[0002] With the widespread application of IoT technology and smart sensing devices in building management, it has become possible to achieve comprehensive monitoring of environmental conditions by deploying multi-source sensor networks. Existing fire early warning systems typically use sensors such as smoke and temperature for real-time monitoring and issue alarms when the detected values ​​exceed preset thresholds. At the same time, the research field is actively exploring how to use time series analysis, data fusion, and spatial correlation models to conduct earlier and more comprehensive assessments of fire risks.

[0003] Chinese invention patent application CN120071585A discloses an intelligent fire monitoring method and system for high-rise buildings, including the following steps: deploying multimodal sensor nodes at the construction site of the high-rise building to collect environmental data; transmitting the environmental data to a central monitoring center via a wireless sensor network; denoising the environmental data using a Kalman filter and fusing the multimodal sensor data using a dynamically adjusted weighted fusion algorithm to generate preliminary judgment parameters for fire monitoring; and assessing the fire risk based on a dynamic Bayesian decision network to generate the probability of fire occurrence.

[0004] However, achieving a complete technological closed loop from multi-dimensional environmental information perception, dynamic risk feature extraction, spatial propagation modeling to multi-timescale prediction and proactive intervention strategy generation remains a topic worthy of in-depth research in the field of intelligent building fire prevention and control, in order to promote the intelligent evolution of fire safety management from post-event alarm to pre-event risk assessment and proactive prevention and control. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a smart building fire risk assessment method and system based on sensor networks.

[0006] The technical solution of this invention is a smart building fire risk assessment method based on sensor networks, comprising the following specific implementation steps: S1. Continuously collect environmental information through a multi-source sensor network deployed in various spatial units within the building, and introduce time-continuous change modeling and spatial consistency constraints to transform the monitoring data into a networked environmental state expression; S2. Perform risk-oriented reconstruction based on the environmental state expression, and extract low-dimensional risk feature vectors through risk sensitivity analysis, component collaborative change analysis, and historical benchmark comparison; S3. Based on the low-dimensional risk feature vectors, combine building spatial adjacency relationships and airflow information to construct a fire risk spatial transmission network, simulating the dynamic propagation of risk in the building space; S4. Based on the output of the fire risk spatial transmission network, construct a unit-level risk time series, perform trend prediction at multiple time scales, and quantify the prediction confidence level; S5. Transform the trend prediction results into executable intervention decisions, and achieve a proactive prevention and control closed loop through risk classification, response action generation, strategy optimization, and post-implementation effect feedback.

[0007] Preferably, in step S1, the modeling of continuous temporal change specifically involves: for each environmental component collected by each sensor node, calculating the change of that component within adjacent sampling time intervals, which is used to characterize the continuous evolution trend of the environmental state; the spatial consistency constraint specifically involves: weighted summarization of the environmental states of multiple sensor nodes belonging to the same spatial unit, using the average state of all nodes in the spatial unit as the comprehensive environmental state of the unit, and assigning lower weights to the states of individual nodes that deviate too much from the average state, thereby obtaining the corrected comprehensive environmental state vector of the spatial unit.

[0008] Preferably, in step S2, the risk sensitivity analysis specifically involves: for each environmental state component of a spatial unit, firstly calculating the average change amplitude of the component within a preset time window; simultaneously calculating the consistency index of the change direction of the component at each sampling moment within the time window; then converting the consistency index into a risk weighting coefficient through a nonlinear mapping function; finally combining the average change amplitude, the risk weighting coefficient, and an initial influence coefficient reflecting the difference in the physical contribution of the component to calculate the risk sensitivity index of the environmental state component.

[0009] Preferably, in step S2, the component synergistic change analysis specifically involves: calculating the correlation coefficients between pairs of multiple key environmental state components during the change process within the same spatial unit, and comprehensively processing the correlation coefficients of all key component pairs to obtain a synergy index that characterizes the degree of synergy in the changes of multiple environmental state components within the spatial unit.

[0010] Preferably, in step S2, the historical benchmark comparison specifically involves: establishing a stable state benchmark for each spatial unit based on long-term historical data, which includes the long-term average value of each environmental state component and its normal fluctuation range; comparing the current environmental state vector of the spatial unit with the stable state benchmark, calculating the ratio of the deviation of the current value of each key component from the historical average value to its historical normal fluctuation range, and combining these ratios to obtain a deviation index characterizing the deviation of the current state from the historical normal level.

[0011] Preferably, in step S3, constructing the fire risk spatial transmission network specifically includes: S31, obtaining the adjacency relationship of spatial units based on the building's floor plan layout to form an initial adjacency matrix, and integrating the airflow velocity and direction information collected by sensors to calculate the airflow-guided propagation weight between each pair of adjacent spatial units; S32, using the low-dimensional risk feature vector extracted in step S2, calculating the potential risk impact of any spatial unit on its adjacent spatial units, and then combining it with the airflow-guided propagation weight to obtain the final risk coupling weight between spatial units; S33, constructing a dynamic weighted network with spatial units as nodes and risk coupling weights as edge weights, and defining an iterative update formula for the risk state of network nodes to simulate the propagation process of risk along the network edge weights.

[0012] Preferably, step S3 further includes step S34: during the iterative update process of the risk propagation network, a neighbor average correction mechanism is introduced, that is, the average risk state of all neighboring units of a spatial unit is used to correct the risk state of the unit itself; and a time smoothing coefficient is introduced to weight and smooth the corrected risk state with the risk state at the previous moment, so as to obtain a more robust spatial unit risk state for prediction.

[0013] Preferably, step S4, which involves trend prediction at multiple time scales, specifically includes: S41, constructing a historical risk time series for each spatial unit based on its corrected and smoothed risk status; S42, using a hybrid prediction model that integrates a time series model and a spatial graph convolutional network to predict the short-term risk trend of each spatial unit, wherein the spatial graph convolutional network is used to incorporate the risk status influence of its neighboring units; S43, performing multi-step rolling recursion on the short-term risk trend prediction results to obtain the medium- and long-term risk evolution trend; simultaneously, spatially aggregating the risk prediction results of multiple spatial units belonging to the same floor or functional area to generate a regional risk index and an overall building risk evolution curve.

[0014] Preferably, in step S5, the response action generation specifically involves: selecting a set of candidate intervention actions from a predefined response action library based on the risk level of the spatial unit that triggers the response; the candidate intervention actions include adjusting the status of the ventilation and smoke extraction system, limiting the operating power of specific electrical equipment, activating local sprinklers or alarm preparation, guiding personnel through broadcasts or signs, and temporarily closing or isolating some passages; and evaluating the expected risk reduction efficiency and execution cost of each candidate action; the strategy optimization specifically involves: when multiple spatial units need to respond simultaneously, calculating the response priority based on the risk prediction value, prediction confidence, and spatial importance weight of each unit, and solving for the optimal combination of intervention actions and execution sequence with the optimization objective of maximizing the overall risk reduction efficiency of the system and minimizing the execution cost.

[0015] The technical solution of this invention is as follows: A smart building fire risk assessment system based on sensor networks, used to execute the aforementioned smart building fire risk assessment method based on sensor networks, comprising: a multi-dimensional building environment perception and networked representation module, used to collect environmental information through a deployed multi-source sensor network, and perform time-continuous change modeling and spatial consistency constraint processing to generate a networked environmental state expression; a spatial risk correlation modeling and neighborhood relationship calculation module, used to receive the environmental state expression, perform risk sensitivity analysis, component collaborative change analysis and historical benchmark comparison, and extract low-dimensional risk feature vectors; and a dynamic risk network correction and risk transmission analysis module. The system is used to construct and iteratively update a fire risk spatial transmission network based on the low-dimensional risk feature vector, combined with building spatial structure and airflow information, to simulate the dynamic propagation and evolution of risk; the fire risk trend prediction and multi-timescale evolution analysis module is used to construct a unit-level risk time series based on the output of the fire risk spatial transmission network, perform short-term and medium-to-long-term trend prediction, and conduct quantitative assessment of prediction confidence; the risk response strategy generation and adaptive intervention execution module is used to classify risks based on trend prediction results, generate and optimize response action strategies, execute intervention actions, and collect feedback data after execution to correct system model parameters, forming closed-loop control.

[0016] Compared with existing technologies, the above-mentioned technical solution of the present invention has the following beneficial technical effects: The present invention designs an intelligent building fire risk assessment method and system based on sensor networks. First, the system transforms discrete environmental monitoring data into continuous and stable networked state expressions through collaborative perception and spatiotemporal fusion modeling of multi-source sensor networks, significantly improving the reliability of environmental perception and the ability to capture early abnormal signals, overcoming the limitations of traditional point-based alarms that are susceptible to interference and have high false alarm and missed alarm rates. Second, through risk-oriented feature reconstruction and spatial consistency analysis, a multi-dimensional and low-dimensional representation of the fire incubation state is achieved, making risk assessment not only dependent on absolute thresholds but also focusing on the changing trends and synergistic effects of the state, thereby greatly improving the accuracy of risk identification. The system ensures accuracy and robustness; furthermore, by constructing a dynamic risk propagation network that integrates physical structure and aerodynamics, it achieves the quantitative transmission and evolution simulation of fire risks in building spaces, enabling the system to predict the impact of local risks on the overall situation, providing a scientific basis for early warning and regional coordination; in addition, the system establishes a multi-timescale prediction system from short-term early warning to medium- and long-term trend analysis, and combines confidence assessment to form a reliable risk situation assessment, supporting managers to conduct tiered responses and forward-looking planning; finally, the system forms a complete closed loop from risk perception, assessment, prediction to the generation of optimized intervention strategies and implementation feedback, realizing the transformation of fire prevention and control from post-event response to pre-event proactive intervention, enhancing the foresight, adaptability, and overall effectiveness of building safety management. Attached Figure Description

[0017] Figure 1 is a flowchart of a smart building fire risk assessment method based on sensor networks proposed in this invention; Figure 2 is a system architecture diagram of a smart building fire risk assessment system based on sensor networks proposed in this invention. Detailed Implementation

[0018] Example 1, as shown in Figure 1, proposes an intelligent building fire risk assessment method based on sensor networks. The specific implementation steps are as follows: S1. By constructing a multi-source sensor network within the building, continuous collection of environmental information closely related to fire formation, such as temperature, smoke, combustible gases, and airflow, is performed. Temporal continuous variation modeling and spatial consistency constraints are introduced to transform discrete, instantaneous monitoring data into a networked state expression that reflects the environmental evolution process. Specifically: S11. By collaboratively collecting multi-dimensional environmental parameters within building spatial units, various environmental quantities are uniformly regarded as components characterizing the fire incubation state, making the environmental perception results closer to the actual fire formation mechanism. Specifically: the building is divided into several spatial units according to usage function, personnel density, and spatial connectivity, and several sensor nodes are deployed in each spatial unit; each sensor node synchronously collects multiple environmental parameters highly related to fire formation, enabling the collection results to reflect the environmental change state from multiple physical dimensions; for any sensor node i, the original environmental observation vector formed at time t is defined as: ;in, This represents the original environmental observation vector collected by sensor node i at time t, which is used to comprehensively reflect the multidimensional environmental state at the location of the node; This represents the ambient temperature value collected by sensor node i at time t; This represents the smoke concentration value collected by sensor node i at time t; This represents the volume fraction of combustible gas collected by sensor node i at time t; This represents the airflow velocity collected by sensor node i at time t; S12, by modeling the changes in environmental parameters within adjacent time windows, environmental monitoring is transformed from instantaneous numerical judgment to continuous evolution process analysis, thereby highlighting the changing trends and cumulative effects of environmental states and providing a time dimension basis for early risk identification. Specifically, for any environmental component x at node i, its continuous temporal change is defined as: ;in, This indicates that the environmental component x is within an adjacent sampling time interval at sensor node i. The amount of change within; This represents the value of the environmental component x at sensor node i; S13. By introducing a multi-node state consistency constraint within a spatial unit, local abnormal fluctuations are flexibly corrected. This improves overall state stability while preserving weak signals of potential risks, making the environmental state representation more reliable and less likely to mask early anomalies. Specifically, considering that a single sensor node may be affected by local disturbances or temporary anomalies, a spatial consistency constraint mechanism is introduced to comprehensively correct the environmental state of multiple nodes within the same spatial unit; let spatial unit k contain a set of nodes. Then the comprehensive environmental state of the spatial unit at time t is defined as follows: Based on this, by comparing the deviation between the state of a single node and the overall spatial state, nodes with larger deviations are assigned lower weights, so that potential local anomalies can still be preserved as weak signals; among them, This represents the comprehensive environmental state vector of spatial unit k at time t; This represents the number of sensor nodes within space unit k, used for normalization when calculating the overall environmental state of the space unit; S14. Based on the aforementioned processing, the stable state characteristics and key change trends of each spatial unit are uniformly mapped into a standardized state vector to construct an environmental state network that can be transmitted and evolved in the building space, providing structured input for subsequent fire risk assessment and predictive analysis. Specifically, after completing time-continuous modeling and spatial consistency correction, the environmental state of each spatial unit is uniformly mapped into a standardized state vector to construct the building environmental state network. The state vector of spatial unit k at time t is expressed as: ;in, This represents the final environmental state vector of spatial unit k at time t, which comprehensively reflects the current environmental state and key change trends of the spatial unit. This represents the average ambient temperature state of space cell k at time t, used to describe the overall thermal environment level of the space cell; This represents the average smoke concentration state of spatial cell k at time t, used to reflect the smoke accumulation situation within the spatial cell; It represents the average combustible gas volume fraction state of space cell k at time t, and is used to describe the potential fuel conditions of the space cell; It represents the average airflow velocity state of space cell k at time t, and is used to characterize the gas flow characteristics within the space cell; This represents the temperature change trend of spatial cell k within adjacent sampling time intervals, used to reflect whether heat shows a continuous accumulation trend; This represents the trend of smoke concentration change in spatial cell k over adjacent sampling time intervals, used to reflect whether the smoke shows a continuous increasing trend.

[0019] S2. Based on the spatial unit environmental state vector output in step S1, risk-oriented reconstruction and feature extraction of the building environment state are performed to extract low-dimensional, evolvable, and transferable risk features, providing input for subsequent fire risk transmission and trend prediction. Through risk sensitivity analysis, component collaborative change analysis, historical stable benchmark comparison, and risk feature vector reconstruction, the high-dimensional, multi-source, and continuous environmental state is transformed into a risk feature vector reflecting the fire evolution trend, improving the accuracy and robustness of risk identification. Specifically: S21. Component-level risk sensitivity calculation is performed on the environmental state vector of each spatial unit. The contribution of each component to the potential fire risk is dynamically evaluated by comprehensively considering the component change amplitude and consistency, forming a single-component risk index, providing basic input for subsequent collaborative analysis. Specifically: In step S1, the state vector of each spatial unit k... The current state and key trends have already been included; to identify which state components are most sensitive to fire risk, a risk sensitivity calculation formula is designed: ; ; ;in, This represents the risk sensitivity index of state component j in spatial unit k; This represents the average variation amplitude of state component j in spatial unit k; The consistency index represents the direction of change of state component j within the time window; Indicates the direction of change at time point t (rising, falling, or remaining stable); N represents the number of sampling points within the time window; Represents a symbolic function; This represents a nonlinear mapping function that transforms the consistency index of change into a risk-weighted coefficient. The initial influence coefficient is used to reflect the differences in the physical contribution of different state components to the formation of fire risk; S22, analyze the coordinated change relationship of multiple state components within a spatial unit, identify synchronous enhancement or continuous linkage trends by calculating the correlation between changes in components, extract the combined signals that truly indicate fire risk, and improve the ability and accuracy of identifying early fire risks. Specifically, early signs of fire are usually manifested as joint changes in temperature, smoke, and combustible gases. Therefore, a coordinated change index is established for the multi-component vector of each spatial unit k: ;in, The index represents the degree of coordination of changes in state components within spatial unit k; n represents the number of state components. Represents the correlation coefficient; S23. Establish a historical stable state benchmark for each spatial unit, including the long-term mean and standard deviation. Compare the deviation of the current state from the historical benchmark to identify abnormal changes that deviate from the normal range, and achieve spatial adaptive judgment. Specifically, establish a long-term historical stable state benchmark for each spatial unit: The deviation between the current state and the historical benchmark is defined as: ;in, This represents the historical stable state benchmark of spatial unit k, used as a reference for comparing the current state. , , and These represent the long-term average values ​​of the components, i.e., the normal operating status of the spatial unit; and The standard deviation of the key components is used to define the normal range of fluctuations. An indicator representing the deviation of the current state from the historical benchmark; Represents a small constant to prevent division by zero; J represents the set of summed state components, selecting the set of key risk components; S24, integrating risk sensitivity, component synergy, and historical deviation, constructs a low-dimensional risk feature vector, including the comprehensive risk activation degree, synergistic change intensity, and deviation degree, providing structured, risk-oriented input for subsequent risk transmission modeling. Specifically, after completing component sensitivity analysis, synergistic change analysis, and historical benchmark comparison, the spatial unit state is compressed into a risk-oriented feature vector: ; ;in, This represents the final risk feature vector of spatial unit k; The degree of activation of the overall risk of a space unit is represented by a weighted average of the risk sensitivity of individual components.

[0020] S3. Based on the risk feature vector output in step S2, a spatial transmission network of fire risk within the building is constructed through physical adjacency relationships, aerodynamic paths, and risk feature weighting. This enables dynamic propagation and evolvable modeling of local risks to the global risk profile. Neighbor correction and smoothing optimization are used to improve prediction stability and accuracy, providing a scientific basis for subsequent trend prediction. Specifically: S31. A spatial unit adjacency matrix is ​​constructed using building floor plan layout and airflow information. Airflow direction and wind speed are introduced to adjust the adjacency weights, forming a potential risk propagation path coupled with physical and aerodynamic principles. Specifically: using the building blueprint, the planar coordinates, door and window positions, and connectivity relationships of each spatial unit are obtained to generate an initial adjacency matrix; the airflow information from step S1 is then used to construct the spatial unit adjacency matrix. Incorporating adjacency weights, calculate the airflow impact for each pair of spatial cells k and l: Modeling of building air ducts, air conditioning systems, and natural ventilation paths is used as a supplement to weighted adjacency edges to ensure that risk propagation paths include both spatially adjacent elements and remote units that may be affected by aerodynamics. This represents the angle between the airflow direction of spatial cell k and the direction of the line connecting the neighboring cell l; This represents the geometric center distance between spatial cell k and its neighboring cell l; S32 represents the airflow guidance weight from spatial unit k to neighbor l; combining the risk feature vector from step S2, the potential impact of each unit on its neighbors is weighted and calculated, then multiplied by the airflow guidance matrix to obtain the risk coupling weight matrix between units, quantifying the risk transmission intensity. Specifically, this is done by using the risk feature vector from step S2. Define the potential influence of unit k on its neighbor l: aerodynamic weights As an adjustment factor, the final coupling weights are obtained: Combine all neighbor relationships to generate a complete weighted risk transmission matrix W(t), and record the source, destination, and weight of each edge for dynamic analysis; among which, , and The weighting coefficient representing the transmission of local risk characteristics to neighboring risks; This represents the potential risk impact of spatial unit k on its neighbor l (without considering airflow direction); Let represent the final risk coupling weight of spatial unit k with respect to its neighbor l (in conjunction with the airflow direction); W(t) represents the risk propagation weighting matrix of the entire building, composed of all Composition; S33, using spatial units as network nodes and risk weights between neighbors as edges, a dynamic and iterative risk propagation network is constructed. The node risk state is updated through an iterative formula to achieve continuous propagation of local risks to the entire building. Specifically: each spatial unit is treated as a network node, and each adjacency relationship... weighted value As edge weights; set the node attributes as the risk feature vector of step S2. This ensures that each node not only represents its location but also contains dynamic risk information; the network risk iterative update formula is defined as follows: Network iteration can be performed using discrete time steps. Simulates the propagation trend of fire risk within buildings, supporting multi-timescale analysis; among which... This indicates the predicted state of the risk network at the next time step; S34, introduce a neighbor average correction and smoothing mechanism to correct the risk state of each node, suppress local abnormal fluctuations, and improve the robustness of network prediction. This can also be optimized by considering personnel density and equipment operating status. Specifically, a neighbor average correction mechanism is introduced: Smoothly update the node's own risk status: ;in, Represents the set of neighbors of spatial unit k; This represents the average correction factor of node k's neighbors; This represents the time smoothing coefficient, with a value in [0,1]. This indicates the node risk status after correction and smoothing.

[0021] S4. Based on the dynamic risk network output of step S3, a multi-timescale fire risk trend prediction system is formed by constructing unit-level risk time series, short-term trend prediction, medium- and long-term risk evolution analysis, and confidence quantification. This enables the understanding of the risk situation at the unit, regional, and building levels. Specifically: S41. For each spatial unit, a historical risk series with a continuous time window is constructed based on the corrected risk status. Through normalization, denoising, and historical event enhancement, the time series is made robust and reliable, providing accurate input data for trend prediction. Specifically, for each spatial unit k, the corrected risk status output in step S34 is recorded. In continuous time window Internal historical sequence: The time series data is normalized and denoised using methods such as moving averages and exponentially weighted filtering to reduce the interference of instantaneous environmental fluctuations on the prediction results. Historical event labels (such as small smoke alarms and fire drill records) can be selectively introduced for sequence augmentation to improve the model's sensitivity to real risk signals. This represents the corrected risk state of spatial unit k at time t; Let k represent the set of risk time series for spatial unit k; S42, using a hybrid prediction model, perform short-term trend prediction on the risk time series of each unit and the risk status of its neighbors, outputting the future risk increment and direction of change for the unit, achieving highly sensitive early risk warning, specifically: for each unit's risk series Using a hybrid prediction model: Predicted output This characterizes the potential increase and trend direction of risk for a given unit in the short term; among them, This represents the predicted risk state of spatial unit k at a future time step; This represents a mixed prediction function for short-term risks, which can be combined with a lightweight LSTM model and a risk network graph convolution (GCN) to capture both time dependence and the influence of neighboring risks. Represents the set of neighboring units of spatial unit k; This represents the set of corrected risk states of spatial unit k's neighborhood at time t; S43, the short-term prediction results are rolled out to the medium and long term, combined with spatial aggregation to generate a regional risk index, and then weighted to form a building-wide risk evolution curve, realizing multi-level, multi-time-scale risk situation analysis and visualization, providing a basis for medium- and long-term prevention and control, specifically: using the short-term prediction results of step S42 as rolling input, a multi-step prediction sequence is constructed: Layered trend analysis is employed: Spatial aggregation layer: The risk of units on the same floor or functional area is weighted and averaged to form a regional risk index. Building-wide risk level: Weighted average of risk indices for all areas to generate a building-wide risk evolution curve. Weights can be set based on regional population density, combustible material distribution, and historical fire risk data; it generates risk trend visualizations: heat maps, risk curves, and risk level change animations, supporting managers in proactively intervening in and optimizing strategies for medium- and long-term fire risks; among which, This represents the prediction step index, which is the number of time steps that are extrapolated backwards from the current time. Indicates that spatial unit k is in the future. Predicted risk status at each time step; This represents the set of risk prediction states of spatial unit k-neighborhood in the previous prediction step; This represents a risk trend recursive function used to map the current risk state and its spatial correlation effects to the risk state at the next time step, thereby characterizing the accumulation, amplification, or attenuation trend of risk; S44, quantifying the confidence level of short-term and medium-to-long-term forecast results, combining forecast uncertainty with risk trends to generate a multi-timescale comprehensive risk index, providing robust short-term early warning, medium-term intervention, and long-term planning decision support for building management, specifically: introducing confidence level measures for short-term and medium-to-long-term forecast results: A comprehensive risk index with multiple time scales is generated by combining confidence level and predicted trend. A comprehensive multi-timescale analysis of the entire building is conducted, integrating short-term early warning, medium-term intervention, and long-term planning into unified risk situation decision-making information to support real-time early warning, emergency response, and long-term fire prevention strategies; among which, This represents the confidence level of risk prediction for spatial unit k at a future time. This represents the variance of the predicted risk outcome; This indicates the preset maximum allowable reference value for prediction variance; Represents a small positive number that is protected against division by zero; This represents the multi-timescale comprehensive risk index of spatial unit k.

[0022] S5. Based on the multi-timescale predicted risk results formed in step S4, the predicted fire risk is transformed into actionable intervention decisions. Through risk classification, response action generation, strategy optimization and execution, and effect feedback correction, a closed-loop evolution from risk assessment to proactive prevention and control is achieved, enabling the system to have the ability to intervene in advance, adjust dynamically, and learn continuously. Specifically: S51. Based on the comprehensive risk index and prediction confidence of the spatial unit at future moments, the risk situation is classified and analyzed, and a mapping relationship between risk level and intervention intensity is established. Under the premise of ensuring prediction reliability, the corresponding level of response is triggered, providing clear and controllable triggering conditions for subsequent intervention decisions. Specifically: For each spatial unit k, the comprehensive risk index of its future key moments is read. Based on building management objectives and safety regulations, risk continuity values ​​are mapped to response levels: Simultaneously, a confidence constraint is introduced: when the prediction confidence level falls below a preset lower limit, only a low-intrusion response is allowed to be triggered, avoiding misjudgments that could lead to resource waste or staff panic; among other things, This indicates the risk response level of spatial unit k in the future. The risk level mapping function is designed based on safety regulations, building management strategies, or empirical parameters. S52, Combining the building's spatial structure and risk distribution characteristics, dynamically generates multiple optional intervention actions for the spatial unit triggering the response, and evaluates the impact of each action on risk reduction and operational costs, forming a candidate set of response action combinations that balances safety and operational feasibility. Specifically, for the spatial unit k triggering the response, a set of candidate intervention actions is generated: The actions include, but are not limited to: adjusting ventilation or smoke extraction status, limiting the operating power of combustible equipment, activating local fire-fighting preparation units, and pre-emptively evacuating people or closing passageways; each action corresponds to an effectiveness function. Based on the spatial coupling relationship of buildings, the chain reaction of actions on neighboring areas should be considered to avoid local intervention causing increased risks in other areas; among them, Represents the set of candidate response actions for spatial unit k; This represents the i-th specific response action in the candidate action set; m represents the response action set in the candidate action set. Indicates a response action Expected effectiveness in reducing risk; Indicates a response action The execution costs or disruption to the normal operation of the building; Indicates a response action The resulting risk reduction; S53. In multi-spatial unit concurrent risk scenarios, determine response priorities based on predicted risk levels and spatial importance, and select the optimal set of intervention actions and execution order through optimization strategies to maximize the overall risk control effect under limited resource conditions. Specifically, construct a unit-level response optimization objective function: When multiple spatial units trigger responses simultaneously, a priority sorting method is introduced: ;in, This indicates the subset of response actions ultimately selected for execution; This represents the objective function for optimizing the response strategy. Indicates the trade-off coefficient; This indicates the response execution priority of spatial unit k; The importance weight of spatial unit k is indicated; S54. After the response action is executed, the environmental state after intervention is acquired in real time and compared with the prediction results. The intervention effect and prediction deviation are fed back to correct the parameters of the risk assessment and prediction model, thereby continuously improving the system's adaptability to the evolution of fire risk and response effect. Specifically, after the response action is executed, the environmental state is re-acquired to form the risk state after intervention: Compare the predicted risks with the actual changes in risks: Based on the deviation results, adaptive adjustments are made to: risk propagation weights, prediction model parameters, and response action effectiveness evaluation functions; the updated model parameters are then fed back to steps S3 and S4, forming a complete closed loop; among which, This indicates the amount of prediction error or intervention deviation; Indicates that spatial unit k is in the future time. Predicting risk status.

[0023] Example 2, as shown in Figure 2, presents a sensor network-based intelligent building fire risk assessment system proposed in this invention. This system is used to execute the sensor network-based intelligent building fire risk assessment method proposed in Example 1. It includes: a multi-dimensional building environment perception and networked representation module, a spatial risk association modeling and neighborhood relationship calculation module, a dynamic risk network correction and risk transmission analysis module, a fire risk trend prediction and multi-timescale evolution analysis module, and a risk response strategy generation and adaptive intervention execution module.

[0024] The building multi-dimensional environmental perception and networked representation module is responsible for the continuous, real-time acquisition and digital representation of the multi-dimensional environmental status within the building. The module is equipped with various types of sensors, including those for temperature, smoke, combustible gases, humidity, and personnel density, transmitting the collected data to the system core via wireless or wired networks. The module performs calibration, noise reduction, and spatiotemporal indexing on the sensor data, generating multi-dimensional environmental state vectors and constructing a preliminary spatial network representation, enabling the environmental status of each spatial unit to form computable structured data within the system. The spatial risk correlation modeling and neighborhood relationship calculation module, based on the multi-dimensional environmental status, assesses the potential fire hazards in each spatial unit within the building. The module performs local risk assessment and constructs a spatial correlation network. By analyzing the spatiotemporal correlation between sensor data, airflow paths, combustible material distribution, and building structure information, it calculates the risk coupling relationships between units and generates a dynamic risk network. This network identifies the neighborhood influence of high-risk areas, quantifies spatial-level risk propagation relationships, and provides a spatial dependency basis for short-term forecasting and trend analysis. The dynamic risk network correction and risk transmission analysis module corrects and adjusts the risk status of each unit to eliminate the impact of isolated measurement point errors on the overall risk assessment and simulates risk transmission. Through dynamic network calculation, it integrates local risk changes and neighborhood influences. For each spatial unit, a corrected risk status network is formed; this module realizes dynamic evolution analysis of risks, providing controllable and reliable input data for trend prediction; the fire risk trend prediction and multi-timescale evolution analysis module performs short-term and medium-to-long-term trend prediction of fire risk for each spatial unit and the building as a whole; the module constructs unit-level risk time series, uses time series analysis and spatial coupling model fusion to predict short-term risk changes, and simultaneously performs hierarchical risk trend analysis for the region and the building as a whole; the module also introduces prediction confidence assessment and multi-timescale integration to generate risk situation information that is suitable for both immediate early warning and supports medium-to-long-term prevention and control decisions, providing the system with... It provides comprehensive and visualized risk evolution analysis; the risk response strategy generation and adaptive intervention execution module analyzes the risk level of each spatial unit within the building that triggers intervention, generating a differentiated and combinable set of response actions; the module optimizes response strategies, comprehensively considering risk reduction effectiveness, operating costs, building space coupling effects, and response priorities to determine the optimal intervention plan; after the action is executed, the module collects the post-intervention environment and risk status, feeding the actual effect back to the prediction and correction module to achieve adaptive correction and continuous system optimization; this module ensures that the system can intervene rationally, dynamically, and controllably before risks become apparent, forming a complete closed-loop management capability.

[0025] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for intelligent building fire risk assessment based on sensor networks, characterized in that, The specific implementation steps include the following: S1. Continuously collect environmental information through a multi-source sensor network deployed in various spatial units within the building, and introduce time-continuous change modeling and spatial consistency constraints to transform the monitoring data into a networked environmental state expression; S2. Perform risk-oriented reconstruction based on the environmental state expression, and extract low-dimensional risk feature vectors through risk sensitivity analysis, component synergistic change analysis, and historical benchmark comparison. S3. Based on low-dimensional risk feature vectors, combined with building space adjacency relationships and air flow information, a fire risk spatial transmission network is constructed to simulate the dynamic propagation of risk in building space. S4. Based on the output of the fire risk spatial transmission network, construct a unit-level risk time series, perform trend prediction at multiple time scales, and quantify the prediction confidence. S5. Transform trend prediction results into actionable intervention decisions, and achieve a proactive prevention and control closed loop through risk classification, response action generation, strategy optimization, and post-implementation effect feedback.

2. The intelligent building fire risk assessment method based on sensor networks according to claim 1, characterized in that, In step S1, the continuous time variation modeling specifically involves: for each environmental component collected by each sensor node, calculating the change of that component within adjacent sampling time intervals. This change is used to characterize the continuous evolution trend of the environmental state. The spatial consistency constraint specifically involves: weighting and synthesizing the environmental states of multiple sensor nodes belonging to the same spatial unit, using the average state of all nodes within the spatial unit as the comprehensive environmental state of the unit, and assigning lower weights to the states of individual nodes that deviate too much from the average state, thereby obtaining the corrected comprehensive environmental state vector of the spatial unit.

3. The intelligent building fire risk assessment method based on sensor networks according to claim 2, characterized in that, In step S2, the risk sensitivity analysis specifically involves: for each environmental state component of a spatial unit, firstly, calculating the average change amplitude of the component within a preset time window; simultaneously, calculating the consistency index of the change direction of the component at each sampling moment within the time window; then, converting the consistency index into a risk weighting coefficient through a nonlinear mapping function; finally, combining the average change amplitude, the risk weighting coefficient, and an initial influence coefficient reflecting the difference in the physical contribution of the component to calculate the risk sensitivity index of the environmental state component.

4. The intelligent building fire risk assessment method based on sensor networks according to claim 3, characterized in that, In step S2, the component synergistic change analysis specifically involves: calculating the correlation coefficients between pairs of multiple key environmental state components during the change process within the same spatial unit, and comprehensively processing the correlation coefficients of all key component pairs to obtain a synergy index that characterizes the degree of synergy in the changes of multiple environmental state components within the spatial unit.

5. The intelligent building fire risk assessment method based on sensor networks according to claim 4, characterized in that, In step S2, the historical benchmark comparison specifically involves: establishing a stable state benchmark for each spatial unit based on long-term historical data, which includes the long-term average value of each environmental state component and its normal fluctuation range; comparing the current environmental state vector of the spatial unit with the stable state benchmark, calculating the ratio of the deviation of the current value of each key component from the historical average value to its historical normal fluctuation range, and combining these ratios to obtain a deviation index characterizing the deviation of the current state from the historical normal level.

6. The intelligent building fire risk assessment method based on sensor networks according to claim 5, characterized in that, In step S3, constructing the fire risk spatial transmission network specifically includes: S31, obtaining the adjacency relationships of spatial units based on the building's floor plan layout to form an initial adjacency matrix, and integrating the airflow velocity and direction information collected by sensors to calculate the airflow-guided propagation weight between each pair of adjacent spatial units; S32, using the low-dimensional risk feature vector extracted in step S2, calculating the potential risk impact of any spatial unit on its adjacent spatial units, and then combining it with the airflow-guided propagation weight to obtain the final risk coupling weight between spatial units; S33, constructing a dynamic weighted network with spatial units as nodes and risk coupling weights as edge weights, and defining an iterative update formula for the risk state of network nodes to simulate the propagation process of risk along the network edge weights.

7. The intelligent building fire risk assessment method based on sensor networks according to claim 6, characterized in that, Step S3 also includes step S34: In the iterative update process of the risk propagation network, a neighbor average correction mechanism is introduced, that is, the average risk state of all neighboring units of a spatial unit is used to correct the risk state of the unit itself; and a time smoothing coefficient is introduced to weight and smooth the corrected risk state with the risk state at the previous moment, so as to obtain a more robust spatial unit risk state for prediction.

8. The intelligent building fire risk assessment method based on sensor networks according to claim 7, characterized in that, Step S4, specifically the multi-timescale trend prediction, includes: S41, constructing a historical risk time series for each spatial unit based on its corrected and smoothed risk status; S42, using a hybrid prediction model that integrates a time series model and a spatial graph convolutional network to predict the short-term risk trend of each spatial unit, wherein the spatial graph convolutional network is used to incorporate the risk status influence of its neighboring units; S43, performing multi-step rolling recursion on the short-term risk trend prediction results to obtain the medium- and long-term risk evolution trend; simultaneously, spatially aggregating the risk prediction results of multiple spatial units belonging to the same floor or functional area to generate a regional risk index and an overall building risk evolution curve.

9. The intelligent building fire risk assessment method based on sensor networks according to claim 8, characterized in that, In step S5, the response action generation specifically involves: selecting a set of candidate intervention actions from a predefined response action library based on the risk level of the spatial unit that triggers the response. The candidate intervention actions include one or more of the following: adjusting the status of the ventilation and smoke extraction system, limiting the operating power of specific electrical equipment, activating local sprinklers or alarm preparation, guiding personnel through broadcasts or signs, and temporarily closing or isolating some passages. The expected risk reduction efficiency and execution cost of each candidate action are evaluated. The strategy optimization specifically involves: when multiple spatial units need to respond simultaneously, calculating the response priority based on the risk prediction value, prediction confidence, and spatial importance weight of each unit, and solving for the optimal combination of intervention actions and execution sequence with the optimization objective of maximizing the overall risk reduction efficiency of the system and minimizing the execution cost.

10. A sensor network-based intelligent building fire risk assessment system, used to execute the sensor network-based intelligent building fire risk assessment method according to any one of claims 1 to 9, characterized in that, include: The building multi-dimensional environment perception and networked representation module is used to collect environmental information through a deployed multi-source sensor network, perform time-continuous change modeling and spatial consistency constraint processing, and generate a networked environmental state representation. The spatial risk correlation modeling and neighborhood relationship calculation module is used to receive the environmental state expression, perform risk sensitivity analysis, component synergistic change analysis and historical benchmark comparison, and extract low-dimensional risk feature vectors; the dynamic risk network correction and risk transmission analysis module is used to construct and iteratively update the fire risk spatial transmission network based on the low-dimensional risk feature vectors, combined with building spatial structure and air flow information, to simulate the dynamic propagation and evolution of risks; the fire risk trend prediction and multi-timescale evolution analysis module is used to construct unit-level risk time series based on the output of the fire risk spatial transmission network, perform short-term and medium-to-long-term trend prediction, and conduct quantitative assessment of prediction confidence. The risk response strategy generation and adaptive intervention execution module is used to classify risks based on trend prediction results, generate and optimize response action strategies, execute intervention actions, and collect feedback data after execution to correct system model parameters, thus forming closed-loop control.

Images