Fire monitoring system and method

Abstract

The invention discloses a fire monitoring system and method. The system comprises a terminal layer, an edge layer and a cloud layer, the terminal layer is used for monitoring a target area in real time and transmitting monitoring data acquired by each sensor to the edge layer; the edge layer is used for performing feature fusion of a time dimension and a space dimension on the monitoring data to obtain a space-time fusion feature, and extracting a visual feature from the monitoring data; fire point and smoke recognition is carried out by fusing the space-time fusion features and the visual features, when a fire point or smoke is recognized, a fire alarm is triggered, and monitoring data is transmitted to the cloud layer; and the cloud layer is used for constructing a hierarchical risk assessment system based on the monitoring data, performing fire risk evolution path assessment based on the hierarchical risk assessment system, and generating a fire processing decision according to an assessment result. According to the invention, the fire detection precision is improved, the missing report rate and the false alarm rate are reduced, and the practicability of the system is improved.

Description

Technical Field

[0001] This application relates to the technical field of fire monitoring, and particularly to a fire monitoring system and method. Background Art

[0002] Existing fire monitoring systems capture abnormal information through sensors set in fixed areas to trigger alarms. However, most of them rely on single-sensor technology, with high false alarm rates and missed alarm rates, resulting in waste of emergency resources and possible delay of the best rescue opportunity. Moreover, the existing fire detection systems have single functions, unable to estimate the evolution path of fire risks and unable to give effective fire risk treatment measures, having the problem of low practicability. Summary of the Invention

[0003] This application provides a fire monitoring system and method for solving the technical problems of high false alarm rates and missed alarm rates and low practicability existing in the existing fire detection systems.

[0004] In view of this, in the first aspect of this application, a fire monitoring system is provided, including: a terminal layer, an edge layer, and a cloud layer;

[0005] The terminal layer includes various types of sensors for real-time monitoring of a target area and transmitting the monitoring data collected by each sensor to the edge layer; the monitoring data includes thermal infrared images, point cloud data, video data, and gas concentration;

[0006] The edge layer is used to perform feature fusion in the time dimension and space dimension on the monitoring data to obtain spatio-temporal fusion features, and extract visual features from the monitoring data; fire points and smoke are identified by fusing the spatio-temporal fusion features and the visual features. When a fire point or smoke is identified, a fire alarm is triggered, and the monitoring data is transmitted to the cloud layer;

[0007] The cloud layer is used to construct a hierarchical risk assessment system based on the monitoring data, evaluate the evolution path of fire risks based on the hierarchical risk assessment system, and generate a fire treatment decision according to the evaluation result.

[0008] Optionally, when the edge layer is used to perform feature fusion in the time dimension and space dimension on the monitoring data to obtain spatio-temporal fusion features, it specifically includes:

[0009] Performing feature mapping on the monitoring data collected by each sensor respectively to obtain the mapped feature vectors corresponding to each sensor;

[0010] Calculating the spatial attention weights of each sensor through the mapped feature vectors corresponding to each sensor;

[0011] Calculate the temporal attention weights of each piece of the monitoring data by combining a time decay function and each of the mapped feature vectors;

[0012] Perform weighted summation on each piece of the monitoring data through the temporal attention weights of each piece of the monitoring data to obtain a temporal attention feature; perform weighted fusion on the temporal attention feature through the spatial attention weights of each sensor to obtain a spatio-temporal fusion feature.

[0013] Optionally, the cloud layer is specifically configured to construct a hierarchical risk assessment system, and the hierarchical risk assessment system includes primary indicators, secondary indicators, and tertiary indicators;

[0014] Determine the index weights of each index through the analytic hierarchy process;

[0015] Obtain index values through the monitoring data, and calculate an immediate risk score through the index values and the index weights;

[0016] Taking the index as a node and the fire handling strategy as an action, adopt the Monte Carlo tree search method to optimize the risk assessment path, obtain the optimal risk assessment path, and generate a fire handling strategy through the optimal risk assessment path;

[0017] Among them, during the process of optimizing the risk assessment path, update the cumulative reward for executing each action in each state based on the immediate risk score;

[0018] Calculate the upper confidence bound value of each node based on the cumulative reward for executing each action in each state, the number of times of executing each action in each state, and the number of visits to each state;

[0019] Generate the optimal risk assessment path by selecting the node with the maximum upper confidence bound value.

[0020] Optionally, the primary indicators include temperature gradient, gas concentration, and optical characteristics;

[0021] The secondary indicators include heat conduction rate, substance phase change trend, and oxygen consumption rate;

[0022] The tertiary indicators include explosion equivalent prediction and personnel evacuation difficulty.

[0023] Optionally, the monitoring data further includes the operation data of target devices in the target area, and the operation data includes temperature, pressure, rotation speed, and current;

[0024] The edge layer is further configured to compare the operation data at the current moment with the alarm threshold corresponding to the current moment. If the value of the operation data at the current moment exceeds the alarm threshold corresponding to the current moment, trigger an equipment anomaly alarm and initiate an emergency measure.

[0025] Optionally, the cloud layer is further configured to obtain an anomaly detection model through training. The anomaly detection model includes a generator and a discriminator. The generator is configured to generate a predicted value of the operating state of the target device under normal operating conditions according to the historical operating data of the target device, and the discriminator is configured to discriminate the authenticity of the predicted value of the operating state generated by the generator;

[0026] Obtain the mean and standard deviation of the predicted values of the operating state of the target device under normal operating conditions at W time steps before time t generated by the generator;

[0027] Calculate the alarm threshold at time t through the mean, the standard deviation, and the sensitivity coefficient, and send the alarm threshold at time t to the edge layer.

[0028] Optionally, the cloud layer is further configured to obtain the false alarm rate and the missed alarm rate within a preset time period according to the historical alarm data and the user feedback data;

[0029] Adjust the sensitivity coefficient by using the gradient descent method according to the change trends of the false alarm rate and the missed alarm rate.

[0030] Optionally, the cloud layer is further configured to combine the location information of the devices in the target area with the geographical environment based on the three-dimensional model and the geographical information system data, and use a real-time rendering engine to realize the dynamic display of the device operating state;

[0031] In response to a user's gesture operation, obtain the interaction content corresponding to the current gesture operation according to the correspondence between each gesture operation and the interaction content;

[0032] In response to a user's voice control instruction, execute the corresponding operation.

[0033] Optionally, the sensor integrates a first communication module and a second communication module;

[0034] When the sensor performs data transmission, when the transmission distance is greater than the distance threshold, the amount of data to be transmitted is less than the data amount threshold, and the transmission power consumption is lower than the power consumption threshold, use the first communication module for data transmission; when the transmission distance is greater than or equal to the distance threshold, or the amount of data to be transmitted is greater than or equal to the data amount threshold, use the second communication module for data transmission.

[0035] The second aspect of the present application provides a fire monitoring method, including:

[0036] The terminal layer monitors the target area in real time through various types of sensors, and transmits the collected monitoring data to the edge layer; the monitoring data includes thermal infrared images, point cloud data, video data, and gas concentration;

[0037] The edge layer performs feature fusion on the monitoring data in the time dimension and the space dimension to obtain spatio-temporal fusion features, and extracts visual features from the monitoring data; fire points and smoke are identified by fusing the spatio-temporal fusion features and the visual features. When a fire point or smoke is identified, a fire alarm is triggered, and the monitoring data is transmitted to the cloud layer;

[0038] The cloud layer constructs a hierarchical risk assessment system based on the monitoring data, and conducts an assessment of the fire risk evolution path based on the hierarchical risk assessment system, and generates a fire handling decision according to the assessment result.

[0039] It can be seen from the above technical solutions that the present application has the following advantages:

[0040] The fire monitoring system provided by the present application deploys various types of sensors at the terminal layer for data monitoring. The edge layer extracts multimodal features from multi-source monitoring data for fire point and smoke identification, which helps to improve the identification accuracy of fire points and smoke, thereby improving the high false alarm rate and missed alarm rate problems existing in fire monitoring using a single sensor and extracting a single feature; the present application performs fire point and smoke identification through the edge layer and responds in real time, which helps to improve the data processing efficiency and response speed; by further constructing a hierarchical risk assessment system and conducting an assessment of the fire risk evolution path through the cloud layer, and giving corresponding fire handling decisions according to the assessment results of the fire risk evolution path, it provides a reference for subsequent fire handling and improves the practicability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0042] Figure 1 FIG. 1 is a schematic structural diagram of a fire monitoring system provided by an embodiment of the present application;

[0043] Figure 2 FIG. 2 is a schematic flowchart of a fire monitoring method provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0045] The fire monitoring system in this application is a system that uses advanced sensing technology, data fusion technology, data analysis technology, artificial intelligence technology, etc. to monitor, analyze, and give early warnings of potential risks of fires and explosions in real time. The system obtains an accurate understanding and prediction of the environmental status and development trend by collecting, integrating, and analyzing data from various data sources, and helps relevant personnel discover potential fire and explosion hazards in a timely manner. The main application fields include:

[0046] Industrial field, high-risk industries such as petrochemical, coal mine, and power, for real-time monitoring of production equipment, storage facilities, pipelines, etc., to prevent the occurrence of fire and explosion accidents. Timely discover potential risks such as equipment failures and leaks, and avoid fires and explosions caused by high temperatures, sparks, etc.

[0047] Commercial buildings, including crowded places such as shopping malls, office buildings, and hotels, can monitor fire hazards in real time, such as electrical failures and illegal use of fire. When a fire occurs, the system can quickly locate the fire source, provide support for personnel evacuation and fire fighting and rescue, and ensure the safety of personnel's lives and property.

[0048] Warehousing and logistics, for places such as warehouses and logistics centers where a large number of flammable and explosive items are stored, the system can realize real-time monitoring of the inventory goods and discover potential hazards such as spontaneous combustion of goods and electrical fires. It is also linked with the fire-fighting facilities and ventilation systems in the warehouse to take fire-fighting measures in a timely manner when a fire occurs and reduce losses.

[0049] Public areas, such as transportation hubs such as airports, railway stations, and subway stations, as well as public places such as parks and squares. Through sensors and monitoring devices installed in these areas, real-time monitoring of fire and explosion risks is carried out to prevent the occurrence of fire accidents and ensure public safety.

[0050] For ease of understanding, please refer to Figure 1 , a fire monitoring system provided by an embodiment of this application includes: a terminal layer, an edge layer, and a cloud layer;

[0051] The terminal layer includes various types of sensors for real-time monitoring of the target area and transmitting the monitoring data collected by each sensor to the edge layer; the monitoring data includes thermal infrared images, point cloud data, video data, and gas concentration;

[0052] The edge layer is used to perform feature fusion on the monitoring data in the time dimension and the space dimension to obtain spatio-temporal fusion features, and extract visual features from the monitoring data; fire point and smoke are identified by fusing the spatio-temporal fusion features and the visual features. When a fire point or smoke is identified, a fire alarm is triggered, and the monitoring data is transmitted to the cloud layer;

[0053] The cloud layer is used to construct a hierarchical risk assessment system based on the monitoring data, and perform an assessment on the evolution path of the fire risk based on the hierarchical risk assessment system, and generate a fire handling decision according to the assessment result.

[0054] This application constructs a three-level intelligent perception system of the terminal, edge, and cloud to achieve efficient data collection and processing. A multi-modal intelligent sensor array is deployed at the terminal layer. According to the risk characteristics of different scenarios, a risk assessment model is used to determine the deployment density and location of the sensors. In a chemical industrial park, through the risk analysis of areas such as the storage tank area and the reaction device area, the spatial grid method is adopted to divide the high-risk areas into several grids, and sensors such as infrared thermal imagers, laser gas detectors, millimeter-wave radars, and multi-spectral cameras are deployed at the center or key positions of each grid to achieve comprehensive collection of data such as temperature, gas concentration, object position, and visual features. An edge intelligent gateway is set at the edge layer, and a lightweight AI model is built in to achieve preliminary data fusion, fire and smoke detection, and real-time response, which helps to reduce the data transmission pressure and improve the processing efficiency. The cloud layer integrates a multi-source heterogeneous data processing engine to achieve in-depth analysis and mining of data. The overall architecture of the system adopts an adaptive topology structure, supporting star / mesh hybrid networking, which improves the flexibility and reliability of the system.

[0055] The sensor integrates a first communication module (such as ZigBee3.0) and a second communication module (such as LoRaWAN). When the sensor performs data transmission, when the transmission distance is greater than the distance threshold, the amount of data to be transmitted is less than the data volume threshold, and the transmission power consumption is lower than the power consumption threshold, the first communication module is used for data transmission; when the transmission distance is greater than or equal to the distance threshold, or the amount of data to be transmitted is greater than or equal to the data volume threshold, the second communication module is used for data transmission. The sensor performs data transmission through the dual communication modules, using the first communication module for low-power and short-distance data transmission, and using the second communication module to achieve long-distance and large-data-volume data transmission.

[0056] The sensor is based on self-organizing network routing protocols such as AODV (Ad-Hoc On-Demand Distance Vector). When a new node joins the network, it sends a routing request packet, and the existing nodes assist the new node to complete network topology construction and routing update through routing reply packets. For OTA (Over-the-Air Technology) upgrade, upgrade instructions and firmware data are sent to the node through a secure channel, and the node performs software and firmware updates after receiving and verifying them to ensure continuous improvement of system performance.

[0057] Specifically, the terminal layer is used to monitor the target area in real time and transmit the monitoring data collected by each sensor in the target area to the edge layer. The monitoring data specifically includes thermal infrared images, point cloud data, video data, gas concentration, etc.

[0058] After receiving the monitoring data, the edge layer extracts spatio-temporal fusion features and visual features from the monitoring data; fire points and smoke are identified by fusing spatio-temporal fusion features and visual features. When a fire point or smoke is identified, a fire alarm is triggered, and the monitoring data is transmitted to the cloud layer.

[0059] When the edge layer is used to perform feature fusion on the monitoring data in the time dimension and the space dimension to obtain spatio-temporal fusion features, it specifically includes:

[0060] Perform feature mapping on the monitoring data collected by each sensor respectively to obtain the mapped feature vectors corresponding to each sensor;

[0061] Calculate the spatial attention weights of each sensor through the mapped feature vectors corresponding to each sensor;

[0062] Combine the time decay function and each mapped feature vector to calculate the time attention weights of each monitoring data;

[0063] Perform weighted summation on each monitoring data through the time attention weights of each monitoring data to obtain time attention features; perform weighted fusion on the time attention features through the spatial attention weights of each sensor to obtain spatio-temporal fusion features.

[0064] The edge layer is deployed with a lightweight AI model based on spatio-temporal attention mechanism for preliminary processing of the monitoring data. Let the monitoring data collected by the sensor be , where is the feature vector of the th sensor at a certain moment. The data types in the calculation of spatial attention weights are mainly the feature vectors corresponding to the sensors, including data information such as temperature, gas concentration, and object position collected by sensors such as infrared thermal imaging, laser gas detection, and millimeter wave radar.

[0065] The edge layer calculates the spatial attention weights to obtain the attention weights of different sensors in the spatial dimension, clarify the importance of each sensor feature during fusion, and thus achieve the fusion of monitoring data in the spatial dimension and enhance the perception ability of spatial features. Among them, the calculation formula for the spatial attention weights is:

[0066]

[0067] In the formula, , , 、 are learnable weight matrices, and Q i 、K j are the mapped feature vectors corresponding to sensors i and j; is the spatial attention weight, indicating the attention weight of sensor i to sensor j in the spatial dimension; is the dimension scaling factor, used to adjust the calculation scale to prevent the calculation result from being too large or too small to ensure the stability of the calculation result.

[0068] The calculation of the temporal attention weights is based on the timestamp t of the above-mentioned monitoring data and the set temporal decay factor to form the temporal decay function , and calculate the temporal attention weights of the above-mentioned monitoring data at different times to capture the change trend and correlation of the data in the time series, and reflect the change of the importance of the data over time. The calculation formula for the temporal attention weights is:

[0069]

[0070] In the formula, represents the temporal attention weight of the monitoring data at time point t i to the monitoring data at time point t j . is the temporal decay factor, , and as the time interval increases, gradually decreases, highlighting the importance of recent data.

[0071] Through the above calculations, the temporal attention weights are used to perform weighted fusion on the monitoring data, and then the spatial attention is used to perform weighted fusion on the fused features to obtain spatio-temporal fusion features. This application fuses multi-source data based on the spatio-temporal attention mechanism, providing accurate basic data for subsequent data analysis and decision-making.

[0072] The edge layer is also used to extract visual features from video data, and the visual features include color histogram features and gray-level co-occurrence matrix texture features; the video features and spatio-temporal fusion features are fused by means of splicing, weighted summation, etc. to obtain multi-modal features. Let the visual feature vector be , and the spatio-temporal fusion feature vector be , and the fused feature vector , where , are learnable weights, and the optimal values can be determined through training, providing auxiliary support for fire point and smoke recognition. The multi-modal features are input into a pre-set fire recognition model for fire point and smoke recognition. When a fire point or smoke is recognized, a fire alarm is triggered, and the monitoring data is transmitted to the cloud layer. The pre-set fire recognition model can be constructed based on the YOLOv11 architecture, and transfer learning can be adopted to fine-tune the model based on a large-scale fire point and smoke annotation data set. When training, optimization algorithms such as Stochastic Gradient Descent (SGD) and Adam can be used to adjust the model parameters (such as convolutional kernel weights, biases, etc.), and the recognition accuracy and robustness can be improved by minimizing the loss function (such as a comprehensive loss function including classification loss and regression loss). An evaluation system with accuracy, recall rate, etc. as the core indicators is established, and the model performance is evaluated based on the test set to ensure that the fire point recognition accuracy and smoke detection accuracy reach the threshold.

[0073] The monitoring data also includes the operation data of the target device in the target area, and the operation data includes temperature, pressure, rotation speed, current, etc.; the edge layer is also used to compare the operation data at the current moment with the corresponding alarm threshold at the current moment. If the value of the operation data at the current moment exceeds the corresponding alarm threshold at the current moment, an equipment anomaly alarm is triggered, and emergency measures are started. Suppose the alarm threshold set is , and this set contains alarm thresholds for different types of data, such as the temperature alarm threshold T temp , the gas concentration alarm threshold T gas , etc. When the edge layer detects that a certain type of operation data i satisfies x i >T i (T i is the alarm threshold corresponding to i ), a local alarm is immediately triggered, and within a preset time (such as 500 ms), the preliminary processing results (such as feature vectors, anomaly type identifiers, etc.) and the monitoring data are uploaded to the cloud layer through a secure network channel, and local emergency measures are started, such as controlling the ventilation system, closing relevant valves, etc.

[0074] In the embodiments of the present application, an embedded device based on a high-performance and low-power processor such as ARM Cortex-A72 can be selected as an edge computing node. Optimize memory management, adopt a paged memory allocation and caching mechanism to reduce memory access latency; in the software architecture, adopt a multi-threaded programming model, divide threads according to the data processing flow, such as data acquisition threads, preprocessing threads, model inference threads, etc., to improve data processing efficiency and response speed.

[0075] After receiving the monitoring data, the cloud layer can perform data cleaning on the monitoring data, such as normalization processing, outlier processing, and missing value processing, etc. It can also perform data conversion, unify different format data (such as converting binary data collected by sensors into floating-point format) into a format suitable for analysis. By establishing a data mapping relationship, integrate data from different sensors and different formats into a unified data warehouse, laying a foundation for subsequent in-depth analysis and mining.

[0076] When the edge layer detects a fire point or smoke, it can upload the detection results and monitoring data to the cloud layer together, and the cloud layer can further perform fire and explosion risk prediction. The cloud layer can build a large fire and explosion warning model based on the Transformer architecture. In the training stage, the model can be trained through a large amount of historical monitoring data, and the model parameters can be optimized by minimizing the loss function (such as cross-entropy loss function) through optimization algorithms such as Stochastic Gradient Descent (SGD), Adagrad, and Adadelta. In the inference stage, input real-time detection data, and after the forward propagation of the model, output the fire and explosion risk prediction results and analysis and inference conclusions.

[0077] The large fire and explosion warning model includes a spatio-temporal position encoding module, a multi-head attention mechanism module, a feature fusion module, and a prediction module. The process of the cloud layer performing fire and explosion analysis and prediction through the large fire and explosion warning model includes:

[0078] S11. Perform spatial position encoding and temporal position encoding on the monitoring data collected by each sensor at each time step to obtain encoded features;

[0079] Suppose there are sensors, and data is collected at time steps. Each sensor will collect -dimensional feature data at each time step. Then, at the rd time step, the monitoring data (such as temperature, gas concentration, object position, etc.) collected by all sensors can be represented as a matrix , and the monitoring data of the entire time series can be represented as a three-dimensional tensor .

[0080] To enable the fire and explosion warning large model to capture the spatio-temporal information of monitoring data, spatio-temporal position encoding is added to the input data through the spatio-temporal position encoding module. Let the spatial position encoding function be and the temporal position encoding function be . For the input data (the feature vector of the th sensor at the th time step), the encoded feature is obtained after spatial position encoding and temporal position encoding:

[0081]

[0082] In the formula, is the spatial position encoding vector of the th sensor, and is the temporal position encoding vector of the th time step.

[0083] Sine and cosine functions can be used to complete position encoding, that is:

[0084]

[0085]

[0086] In the formula, pos represents the position (for spatial position encoding, pos = i; for temporal position encoding, pos = t), and i represents the dimension index.

[0087] S12. Perform a linear transformation on the encoded feature based on the multi-head attention mechanism to obtain the multi-head attention feature;

[0088] Through the multi-head attention mechanism module, the encoded feature is linearly transformed to obtain the query (Query), key (Key), and value (Value) to complete the attention calculation.

[0089] Let the number of heads be H. For the hth head (h = 1, 2,..., H), the calculation of the query, key, and value is as follows:

[0090]

[0091]

[0092]

[0093] In the formula, is the feature matrix composed of the encoded features; , , are all learnable weight matrices.

[0094] The attention output of the first head is:

[0095]

[0096] The outputs of all heads are concatenated and linearly transformed to achieve the final output of multi - head attention:

[0097]

[0098] where is a learnable weight matrix.

[0099] The fire and explosion warning large model can also extract temporal attention and spatial attention. Spatial attention is used to determine the importance of different sensors during feature fusion in the spatial dimension. For the th time step, the spatial attention is calculated as follows:

[0100]

[0101] where and are the query and key vectors of the th and th sensors at the th time step respectively, and is the dimension of the key vector (usually ).

[0102] Temporal attention is used to capture the dependencies of data in the temporal dimension. For the i - th sensor, the temporal attention is calculated as follows:

[0103]

[0104] where and are the query and key vectors of the th sensor at the and th time steps respectively.

[0105] The feature fusion module performs feature fusion on the encoded features using the attention weights extracted above to obtain the fused features. The fused features are input into a prediction layer (such as a fully - connected layer) for prediction to obtain the final fire and explosion risk prediction result :

[0106]

[0107] where ​is the weight vector of the fully connected layer, is the bias term, is the activation function (such as the Sigmoid function), which is used to map the output to interval, representing the probability of the occurrence of fire and explosion risks; is the fused feature. When it is predicted that there is a fire and explosion risk in the target area, an explosion warning is triggered, and emergency measures are taken to reduce property losses and ensure personal safety.

[0108] It should be noted that when training the fire and explosion warning large model, the loss value can be calculated based on the binary cross-entropy loss function, and the model parameters are updated through this loss value. The binary cross-entropy loss function is:

[0109]

[0110] In the formula, is the number of training samples, is the th sample's true label ( represents no risk, represents there is a risk), is the predicted probability of the model.

[0111] Taking into account the spatio-temporal information of the monitoring data, the multi-head attention mechanism is used to complete feature fusion to achieve the prediction of fire and explosion risks. By continuously training the model, its prediction accuracy and reliability are improved.

[0112] When a fire point or smoke is detected at the edge layer, the cloud layer can construct a hierarchical risk assessment system based on the monitoring data, and conduct an assessment of the fire risk evolution path based on the hierarchical risk assessment system, and generate a fire treatment decision according to the assessment results. Specifically, the cloud layer is used to construct a hierarchical risk assessment system, determine the index weights of each index through the analytic hierarchy process; obtain the index values through the monitoring data, calculate the instant risk score through the index values and index weights; use the index as the node and the fire treatment strategy as the action, and use the Monte Carlo tree search method to optimize the risk assessment path to obtain the optimal risk assessment path, and generate a fire treatment strategy through the optimal risk assessment path; among them, during the process of optimizing the risk assessment path, the cumulative reward for executing each action in each state is updated based on the instant risk score; the upper confidence bound value of each node is calculated based on the cumulative reward for executing each action in each state, the number of times of executing each action in each state, and the number of visits to each state; the optimal risk assessment path is generated by selecting the node with the largest upper confidence bound value.

[0113] The hierarchical risk assessment system in the embodiment of the present application includes a first-level indicator I1, a second-level indicator I2 and a third-level indicator I3; the first-level indicator includes temperature gradient, gas concentration, and optical characteristics; the second-level indicator includes heat transfer rate, material phase change trend, and oxygen consumption rate; the third-level indicator includes explosion equivalent prediction and personnel evacuation difficulty. The indicators I1, I2, and I3 can be determined by the hierarchical analysis method. i The indicator weight , n is the number of indicators;

[0114] There are a large number of possible risk evolution paths and decision combinations in fire risk assessment scenarios. Monte Carlo tree search is used to optimize the risk assessment path to improve decision efficiency and accuracy. In Monte Carlo tree search, the node selection strategy uses the upper confidence bound algorithm, and the formula is:

[0115]

[0116] In the formula, Status Next action The cumulative reward of Status Next action The number of times Status The number of visits, is the exploration coefficient (e.g. ). Among them, nodes are the basic elements in Monte Carlo tree search, and each node represents a specific state in the risk assessment process. In this model, a node can correspond to a specific combination state composed of the first-level indicators (temperature gradient, gas concentration, optical characteristics), second-level indicators (heat conduction rate, material phase change trend, oxygen consumption rate) and third-level indicators (explosion equivalent prediction, difficulty of personnel evacuation) at a certain moment.

[0117] The above node selection strategy balances the value estimation and exploration potential of the current node, and quickly screens out a better risk assessment path. The risk value propagation mechanism is:

[0118]

[0119] In the formula, is the discount factor (e.g. ), Score the immediate risk, To perform actions After reaching the new state, It is an optional action in the new state. Through this mechanism, the node value estimation is updated according to the simulation results to optimize the risk assessment results.

[0120] Furthermore, the cloud layer can dynamically adjust the exploration coefficient c by analyzing historical data and simulation scenarios. At the initial stage of risk assessment, increase the value of c to encourage the algorithm to explore more unknown nodes. As the assessment process progresses, gradually decrease the value of c according to the node access situation and the stability of value estimation, so that the algorithm is more dependent on existing experience to select nodes with high value, and improve the ability of the algorithm to find a better path in complex fire risk assessment scenarios.

[0121] In the risk value propagation mechanism, the immediate risk score can be calculated through the metric value and metric weight. This application preferably calculates the immediate risk score by comparing the metric data of each current state with the metric threshold and combining the metric weight. . Such as (when ), in the formula is the i-th metric value, is the metric threshold corresponding to the i-th metric. Through this mechanism, the immediate risk score and the maximum value estimation of the subsequent state are incorporated into the update formula, and the value estimation of the node can reflect the risk information on the entire search path, gradually optimizing the risk assessment result.

[0122] If there is a large amount of risk assessment data for past similar fire scenarios, the actual risk results after performing action a in the similar state s can be statistically analyzed, and their average value can be calculated as the initial value estimation. Alternatively, experts in the field of fire risk assessment can subjectively evaluate the expected risk benefits of performing action a in different states s based on professional knowledge and experience, and give the initial value estimation. The experts can comprehensively consider the common development patterns of fires, the mutual relationships between various metrics, and past experience in dealing with similar fires to determine the initial value of Q(s,a).

[0123] Furthermore, the cloud layer is also used to obtain an anomaly detection model through training. The anomaly detection model includes a generator and a discriminator. The generator is used to generate the predicted operating state value of the target device under normal working conditions according to the historical operating data of the target device, and the discriminator is used to discriminate the authenticity of the predicted operating state value generated by the generator; obtain the mean and standard deviation of the predicted operating state values of the target device under normal working conditions at W time steps before the t-th moment generated by the generator; calculate the alarm threshold at the t-th moment through the mean, standard deviation and sensitivity coefficient, and send the alarm threshold at the t-th moment to the edge layer.

[0124] Specifically, the cloud layer can build an anomaly detection model. By learning the characteristics of the device under normal operating conditions through the anomaly detection model, it can identify abnormal situations during the device operation. The traditional fixed threshold detection method may cause false alarms or missed alarms due to changes in the device operating state, environmental factors, etc. The dynamic alarm threshold setting can continuously adjust the alarm threshold according to the real-time operation data of the device, better adapt to various complex situations, reduce the occurrence of false alarms and missed alarms, improve the reliability of the detection results, and avoid resource waste caused by false alarms and safety accidents caused by missed alarms. The dynamic alarm threshold is calculated based on the normal operating condition data predicted by the generator and adjusted as the device operating state changes, making the anomaly detection more accurate, timely discovering potential fire and explosion risks, and issuing an alarm when the fire and explosion risks just emerge, so as to gain more time for taking emergency measures.

[0125] The anomaly detection model in this application includes a generator and a discriminator. The generator is constructed using a long short-term memory network, and the state equation of the long short-term memory network is:

[0126]

[0127] In the formula, 、 、 are the input gate, forget gate, and output gate respectively, is the candidate memory unit, is the memory unit, is the hidden state, is the weight matrix, is the bias vector, is the sigmoid activation function, is the hyperbolic tangent activation function. By training the long short-term memory network, it learns the characteristics of the device under normal operating conditions and realizes the anomaly detection during the device operation.

[0128] The loss function of the discriminator is:

[0129]

[0130] In the formula, represents the mathematical expectation, is the real data distribution, is the real data sample, is the judgment result of the discriminator on the real data; is the input noise distribution of the generator, is the noise sample, is the false data sample generated by the generator according to the noise, is the judgment result of the discriminator on the false data. By minimizing the discriminator loss function , enabling the discriminator to accurately distinguish between real data and the fake data generated by the generator, while prompting the generator to generate samples closer to the real data distribution and improving the accuracy of anomaly detection.

[0131] The dynamic alarm threshold is calculated based on the normal operating condition data predicted by the generator. Suppose the normal operating condition data sequence predicted by the generator within a period of time (the time step is , and the sliding window length is ) is . First, calculate the mean and the standard deviation of this sequence:

[0132]

[0133]

[0134] Then, combine with the sensitivity coefficient (which can be set according to the actual scenario requirements, and the value range is generally between 0.5 and 2) to calculate the alarm threshold :

[0135]

[0136] As the operating state of the device changes, the normal operating condition data predicted by the generator will also change accordingly, thereby dynamically adjusting the alarm threshold, improving the sensitivity and accuracy of anomaly detection, and reducing false alarms and missed alarms.

[0137] Further, the cloud layer is also used to obtain the false alarm rate and missed alarm rate within a preset time period according to the historical alarm data and user feedback data; according to the changing trends of the false alarm rate and missed alarm rate, the gradient descent method is used to adjust the sensitivity coefficient.

[0138] To meet the requirements of different application scenarios for the sensitivity of anomaly detection, a flexible sensitivity adjustment mechanism is established. Through the user interface or system configuration file, the operator is allowed to adjust the sensitivity coefficient . When in a high-risk environment or strictly monitoring abnormal situations, adjust the sensitivity coefficient so as to trigger alarms more easily; in a relatively stable environment that is more sensitive to false alarms, adjust the sensitivity coefficient to make the alarm threshold more conservative and reduce the probability of false alarms. The system can automatically learn and optimize the sensitivity coefficient according to the historical alarm data and the feedback of actual abnormal situations. For example, statistically calculate the false alarm rate and missed alarm rate within a period of time; according to the changing trends of the false alarm rate and missed alarm rate, use optimization algorithms such as gradient descent to adjust the sensitivity coefficient so that the system can maintain good anomaly detection performance in different scenarios.

[0139] The cloud layer can also be used to build an auxiliary knowledge base for fire and explosion warning. It widely collects materials such as professional literature, regulations and standards, and accident reports in the field of fire and explosion. Using natural language processing technology and knowledge graph construction tools, it extracts entities (such as fire types, explosive substances, fire-fighting equipment, etc.) and relationships (such as causal relationships, association relationships, composition relationships, etc.) from them. A graph database such as Neo4j is used to store the knowledge graph to ensure efficient query and update of data. Through a combination of manual review and automatic verification, the accuracy and integrity of entity relationships are guaranteed, and a domain knowledge graph containing more than 500,000 entity relationships is constructed to provide rich knowledge support for the system. The cloud layer can also design an incremental learning framework based on online learning. When new data or cases arrive, the cloud layer first preprocesses and extracts features from the new data, and then matches it with the existing knowledge graph and model. For newly emerged entities and relationships, they are identified through entity extraction and relationship extraction algorithms and integrated into the knowledge graph. At the same time, the new data is used to fine-tune the model and update the model parameters. For example, during the training of a large model, when new data arrives, the mini-batch gradient descent algorithm is used to update based on the original model parameters, enabling the system to continuously learn new knowledge and improve its intelligence level.

[0140] The cloud layer can also collect and organize a large number of typical domestic and foreign fire and explosion accident cases, and conduct a detailed analysis of each case, including information such as the time, location, cause, process, consequences of the accident, and emergency measures taken. These cases are classified and stored according to dimensions such as accident types, industry fields, and accident scales to establish a structured accident case library. Through case-based reasoning technology, when the system detects a similar risk situation, it can quickly retrieve relevant historical cases to provide a reference basis for the current risk assessment and decision-making.

[0141] The cloud layer can also establish a module for real-time monitoring of the system operation status and data changes. When the cloud layer obtains new data and meets certain update conditions (such as the new data volume reaches a certain threshold, the model performance index drops by more than a certain range, etc.), the model update process is automatically triggered. First, the quality of the new data is evaluated and preprocessed, and then it is combined with the historical data to retrain the large model. During the training process, technologies such as model compression and transfer learning are used to reduce the training time and resource consumption. After training, through a series of performance tests and verifications, it is ensured that the accuracy and stability of the new model are better than those of the old model, and then the new model is deployed into the system to ensure the accuracy and timeliness of the system.

[0142] The cloud layer can also be used to combine the location information of devices in the target area with the geographical environment based on 3D models and geographic information system data, and use a real-time rendering engine to achieve dynamic display of the device operation status; in response to the user's gesture operation, obtain the interaction content corresponding to the current gesture operation according to the corresponding relationship between each gesture operation and the interaction content; in response to the user's voice control command, execute the corresponding operation.

[0143] A real-time rendering engine can be developed based on game development engines such as Unity, leveraging GPU acceleration technology. Conduct 3D modeling and visual presentation of information such as device operation status and fire and explosion risk situation, and improve the realism of visualization through means such as lighting effects and material textures. For example, map temperature data to the 3D model through color gradients to visually display the temperature distribution; use a particle system to simulate the dynamic effects of smoke and fire to provide intuitive and accurate information for operation and maintenance personnel.

[0144] Collect a large number of voice samples including various dialects such as Mandarin, Shandong dialect, Cantonese, Sichuan dialect, and Northeast dialect, and train using a deep learning-based speech recognition model (such as DeepSpeech, Wav2Vec2.0, etc.). During the training process, analyze and extract the voice features of different dialects, and improve the accuracy of dialect recognition by adjusting model parameters and adding dialect-specific language models. Integrate the trained speech recognition model into the system, enabling operation and maintenance personnel to interact with the system through voice commands, facilitating the use of operation and maintenance personnel in different regions. Design a detailed voice control instruction set according to the actual needs of operation and maintenance work. The voice control instructions include several categories such as device operation, information query, and task navigation. In the device operation category, it includes instructions such as start, stop, and parameter adjustment; the information query category includes instructions such as querying device status and fault information; the task navigation category is used to guide users to complete specific operation and maintenance task processes.

[0145] The system can also provide a human-computer interaction interface, integrating gesture recognition technology, allowing users to interact with the 3D model through simple gesture operations. Users can view different parts of the device, obtain detailed information, or execute specific operations through gestures such as finger clicking, zooming, and rotating. For example, in a device repair scenario, users can select the device fault location through gestures, and the system will immediately display relevant repair guides. The system provides convenient perspective switching and navigation functions. Users can freely switch different observation perspectives, such as first-person perspective, third-person perspective, and global perspective, etc.; provide navigation guidance to achieve quick positioning of targets in complex device scenarios. For example, during device inspection, users can quickly find the location of the device to be inspected through the navigation function.

[0146] Develop a virtual operation guidance function to provide virtual guidance for operation and maintenance personnel in aspects such as equipment operation and repair. Through 3D modeling and animation production, present the operation process and repair steps of the equipment in the form of a virtual scene. After wearing the MR device, operation and maintenance personnel can see virtual operation steps and prompt information around the actual equipment. For example, in the equipment repair scenario, the virtual model will indicate the parts to be disassembled, the tools to be used, and the operation sequence, etc., improving the operation and maintenance efficiency and accuracy.

[0147] The fire monitoring system provided by this application deploys various types of sensors at the terminal layer for data monitoring. The edge layer extracts multi-modal features from multi-source monitoring data for fire point and smoke recognition, which helps to improve the recognition accuracy of fire points and smoke, thereby improving the problems of high false alarm rate and missed alarm rate existing in fire monitoring using a single sensor and extracting a single feature. This application conducts fire point and smoke recognition through the edge layer and responds in real time, which helps to improve the data processing efficiency and response speed. By further constructing a hierarchical risk assessment system and conducting an assessment of the fire risk evolution path at the cloud layer, and giving corresponding fire handling decisions according to the assessment results of the fire risk evolution path, it provides a reference for subsequent fire handling and improves the practicability of the system.

[0148] After detecting a fire point or smoke, the fire monitoring system in this application can further predict the fire explosion risk to determine whether there is a fire explosion risk in the target area for further processing. Moreover, the system can dynamically adjust the alarm threshold, which can continuously adjust the alarm threshold according to the real-time operation data of the equipment, better adapt to various complex situations, reduce the occurrence of false alarms and missed alarms, improve the reliability of the detection results, and avoid resource waste caused by false alarms and safety accidents caused by missed alarms.

[0149] The fire monitoring system in this application provides a human-computer interaction interface. Through natural language processing technology, it provides the operation status, warning information, operation suggestions, etc. of the system to the operator in an easy-to-understand way. Even if the operator lacks professional technical knowledge, they can perform effective operations and management under the guidance of the system, reducing the requirements for personnel quality and skills, and giving full play to the system functions. Through the learning and training of a large model on various data, it has stronger generality and adaptability, can be applied to different types of scenarios, and improves the popularization of the system.

[0150] Please refer to Figure 2 , this application embodiment also provides a fire monitoring method, including:

[0151] Step 210, the terminal layer monitors the target area in real time through various types of sensors and transmits the collected monitoring data to the edge layer; the monitoring data includes thermal infrared images, point cloud data, video data, and gas concentration;

[0152] Step 220: The edge layer performs feature fusion on the monitoring data in the time dimension and the space dimension to obtain spatio-temporal fusion features, and extracts visual features from the monitoring data; fire points and smoke are identified by fusing the spatio-temporal fusion features and the visual features. When a fire point or smoke is identified, a fire alarm is triggered, and the monitoring data is transmitted to the cloud layer;

[0153] Step 230: The cloud layer constructs a hierarchical risk assessment system based on the monitoring data, and conducts an assessment of the fire risk evolution path based on the hierarchical risk assessment system, and generates a fire handling decision according to the assessment result.

[0154] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the methods described above can refer to the corresponding processes in the foregoing system embodiments, and will not be elaborated herein.

[0155] As used in the specification of this application and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that comprises a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0156] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B may be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (piece) of the following" or similar expressions refer to any combination of these items, including any combination of single items (pieces) or plural items (pieces). For example, at least one (piece) of a, b or c may mean: a, b, c, "a and b", "a and c", "b and c", or "a, b and c", where a, b, c may be single or multiple.

[0157] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0158] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0159] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0160] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. And the aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (English full name: Read-Only Memory, English abbreviation: ROM), random access memories (English full name: Random Access Memory, English abbreviation: RAM), magnetic disks or optical discs and other various media that can store program codes.

[0161] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A fire monitoring system, characterized in that, Including: The terminal layer, the edge layer, and the cloud layer; The terminal layer includes various types of sensors for real-time monitoring of the target area and transmitting the monitoring data collected by each sensor to the edge layer; the monitoring data includes thermal infrared images, point cloud data, video data, and gas concentration; The edge layer is used to perform feature fusion on the monitoring data in the time dimension and the space dimension to obtain spatio-temporal fusion features, and extract visual features from the monitoring data; trigger a fire alarm when a fire point or smoke is identified by fusing the spatio-temporal fusion features and the visual features, and transmit the monitoring data to the cloud layer; The cloud layer is used to construct a hierarchical risk assessment system based on the monitoring data, evaluate the evolution path of the fire risk based on the hierarchical risk assessment system, and generate a fire handling decision according to the evaluation result.

2. The fire monitoring system according to claim 1, wherein When the edge layer is used to perform feature fusion on the monitoring data in the time dimension and the space dimension to obtain spatio-temporal fusion features, it specifically includes: Performing feature mapping on the monitoring data collected by each sensor respectively to obtain the mapped feature vectors corresponding to each sensor; Calculating the spatial attention weights of each sensor through the mapped feature vectors corresponding to each sensor; Combining the time decay function and each of the mapped feature vectors to calculate the time attention weights of each of the monitoring data; Performing weighted summation on each of the monitoring data through the time attention weights of each of the monitoring data to obtain the time attention features; performing weighted fusion on the time attention features through the spatial attention weights of each sensor to obtain spatio-temporal fusion features.

3. The fire monitoring system according to claim 1, characterized in that, The cloud layer is specifically used to construct a hierarchical risk assessment system, and the hierarchical risk assessment system includes first-level indicators, second-level indicators, and third-level indicators; Determining the index weights of each index through the analytic hierarchy process; Obtaining the index values through the monitoring data, and calculating the immediate risk score through the index values and the index weights; Taking the index as a node and the fire handling strategy as an action, using the Monte Carlo tree search method to optimize the risk assessment path, obtaining the optimal risk assessment path, and generating a fire handling strategy through the optimal risk assessment path; Among them, during the process of optimizing the risk assessment path, updating the cumulative rewards for executing each action in each state based on the immediate risk score; Calculating the upper confidence bound value of each node based on the cumulative rewards for executing each action in each state, the number of times of executing each action in each state, and the number of visits to each state; Generating the optimal risk assessment path by selecting the node with the largest upper confidence bound value.

4. The fire monitoring system according to claim 3, characterized in that, The first-level indicators include temperature gradient, gas concentration, and optical characteristics; The second-level indicators include heat conduction rate, substance phase change trend, and oxygen consumption rate; The third-level indicators include explosion equivalent prediction and personnel evacuation difficulty.

5. The fire monitoring system according to claim 1, characterized in that, The monitoring data also includes the operation data of the target equipment in the target area, and the operation data includes temperature, pressure, rotation speed, and current; The edge layer is further configured to compare the operation data at the current moment with the alarm threshold corresponding to the current moment. If the value of the operation data at the current moment exceeds the alarm threshold corresponding to the current moment, it triggers an equipment anomaly alarm and activates emergency measures.

6. The fire monitoring system according to claim 5, characterized in that, The cloud layer is further configured to obtain an anomaly detection model through training. The anomaly detection model includes a generator and a discriminator. The generator is configured to generate a predicted value of the operation state of the target equipment under normal working conditions based on the historical operation data of the target equipment, and the discriminator is configured to discriminate the authenticity of the predicted value of the operation state generated by the generator. Obtain the mean and standard deviation of the predicted values of the operation state of the target equipment under normal working conditions at W time steps before time t generated by the generator. Calculate the alarm threshold at time t through the mean, the standard deviation, and the sensitivity coefficient, and send the alarm threshold at time t to the edge layer.

7. The fire monitoring system according to claim 6, characterized in that, The cloud layer is further configured to obtain the false alarm rate and the miss alarm rate within a preset time period based on the historical alarm data and the user feedback data. Adjust the sensitivity coefficient by using the gradient descent method according to the changing trends of the false alarm rate and the miss alarm rate.

8. The fire monitoring system according to claim 1, wherein The cloud layer is further configured to combine the position information of the equipment in the target area with the geographical environment based on the three-dimensional model and the geographical information system data, and use a real-time rendering engine to realize the dynamic display of the operation state of the equipment. In response to the user's gesture operation, obtain the interaction content corresponding to the current gesture operation according to the corresponding relationship between each gesture operation and the interaction content. In response to the user's voice control instruction, execute the corresponding operation.

9. The fire monitoring system according to claim 1, wherein The sensor integrates a first communication module and a second communication module. When the sensor performs data transmission, when the transmission distance is greater than the distance threshold, the data volume of the transmitted data is less than the data volume threshold, and the transmission power consumption is lower than the power consumption threshold, the first communication module is used for data transmission; when the transmission distance is greater than or equal to the distance threshold, or the data volume of the transmitted data is greater than or equal to the data volume threshold, the second communication module is used for data transmission.

10. A fire monitoring method, characterized in that, It includes: The terminal layer monitors the target area in real time through various types of sensors, and transmits the collected monitoring data to the edge layer; the monitoring data includes thermal infrared images, point cloud data, video data, and gas concentration. The edge layer performs feature fusion in the time dimension and the space dimension on the monitoring data to obtain spatio-temporal fusion features, and extracts visual features from the monitoring data; it identifies fire points and smoke by fusing the spatio-temporal fusion features and the visual features. When a fire point or smoke is identified, it triggers a fire alarm and transmits the monitoring data to the cloud layer. The cloud layer constructs a hierarchical risk assessment system based on the monitoring data, and performs an assessment on the evolution path of the fire risk based on the hierarchical risk assessment system, and generates a fire handling decision according to the assessment result.

Images