Big data fire safety monitoring management method and system

Abstract

The invention discloses a big data fire safety monitoring management method and system, and the method comprises the steps: sequentially carrying out the cleaning, denoising and normalization processing of collected data, and obtaining standard multi-modal data; extracting an environment trend feature, a spatial distribution feature and a time sequence period feature of the standard multi-modal data by using an improved multi-scale convolutional neural network, and aligning and fusing the features to obtain feature multi-modal data; establishing a risk state assessment model based on a CTHMM continuous time hidden Markov model, defining the safety state of the system as a hidden state, taking the feature multi-modal data as an observation sequence, and outputting a fire risk assessment index; and based on the fire risk assessment index, grading the fire risk by using a hierarchical decision-making mechanism, and triggering an early warning mechanism of a corresponding level according to the risk level. Fire-fighting management is converted from passive disposal to active early warning, and the probability of fire occurrence is remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forestry ecological restoration monitoring, in particular to a big data fire safety monitoring management method and system. BACKGROUND

[0002] At present, fire safety monitoring mainly relies on single sensor technology, such as smoke sensor, infrared flame detector, etc. Using single mode detection is easy to be disturbed by the environment, resulting in high false alarm rate, and it is difficult to distinguish real fire from environmental disturbance. It lacks real-time monitoring of the state of fire-fighting equipment and cannot predictively evaluate fire risk, making it difficult to cope with complex and variable environmental conditions. Although the existing technology usually adopts fixed weight multi-modal fusion, it cannot be dynamically adjusted according to the environment, and lacks predictive maintenance function, making it difficult to improve the intelligent safety monitoring efficiency and accuracy of fire safety. SUMMARY

[0003] The purpose of the present application is to solve the above problems, and a big data fire safety monitoring management method and system are designed.

[0004] To achieve the above purpose, the technical scheme of the present application is that, further, in the above-mentioned big data fire safety monitoring management method, the fire safety monitoring management method comprises the following steps: Collecting environmental perception data, equipment state data, building and personnel dynamic data, and external and historical data, and sequentially cleaning, denoising and normalizing the collected data to obtain standard multi-modal data; Using an improved multi-scale convolutional neural network to extract environmental trend features, spatial distribution features and time sequence period features of the standard multi-modal data, and aligning and fusing the features to obtain feature multi-modal data; Establishing a risk state evaluation model based on a CTHMM continuous time hidden Markov model, defining the safety state of the system as a hidden state, taking the feature multi-modal data as an observation sequence, and outputting a fire risk evaluation index; Based on the fire risk evaluation index, using a hierarchical decision mechanism to classify fire risk, and triggering the corresponding level of early warning mechanism according to the risk level.

[0005] Further, in the above-mentioned big data fire safety monitoring management method, the collecting environmental perception data, equipment state data, building and personnel dynamic data, and external and historical data, and sequentially cleaning, denoising and normalizing the collected data to obtain standard multi-modal data, comprises: The environmental perception data, device state data, building and personnel dynamic data, and external and historical data are collected, the environmental perception data including temperature, humidity, smoke concentration, harmful gas concentration, and light intensity, the device state data including device operating parameters, fault codes, and maintenance records, the building and personnel dynamic data including fire-fighting facility distribution, personnel density, and flow trajectory, the external and historical data including fire cases, hidden danger records, and early warning disposal results; Short-time missing data is completed by using linear interpolation, abnormal data is identified based on the 3σ principle, real mutation signals are retained by wavelet transform denoising, video trajectory data is smoothed by Kalman filtering to remove jumping points caused by occlusion, and high-frequency fluctuation parameters are smoothed by moving average to obtain preprocessed data. The preprocessed data is mapped to the [0, 1] interval based on Min-Max normalization to obtain standard multi-modal data.

[0006] Further, in the above-mentioned big data fire safety monitoring management method, the improved multi-scale convolutional neural network extracts the environmental trend feature, spatial distribution feature and time series period feature of the standard multi-modal data, and aligns and fuses the features to obtain feature multi-modal data, including: The improved multi-scale convolutional neural network sets three convolution kernels, and adds an SE attention mechanism after the convolution layer to enhance the weight of key features; The 1x3 convolution kernel is used to process temperature and humidity time series data to extract rising and falling trends to obtain environmental trend features; The 3x3 convolution kernel is used to convolve the building area grid data to extract high-risk area clustering features to obtain spatial distribution features; Based on the 5x5 convolution kernel combined with the sliding window, the daily and weekly regular fluctuations are extracted to obtain time series period features.

[0007] Further, in the above-mentioned big data fire safety monitoring management method, the improved multi-scale convolutional neural network extracts the environmental trend feature, spatial distribution feature and time series period feature of the standard multi-modal data, and aligns and fuses the features to obtain feature multi-modal data, including: The step length of the environmental trend feature, spatial distribution feature and time series period feature with different sampling frequencies is unified by linear interpolation to obtain unified feature data; The unified feature data is concatenated in the channel dimension, the Cross-Attention cross-attention mechanism is used to learn the correlation between features, and the feature multi-modal data is output.

[0008] Further, in the above-mentioned fire safety monitoring and management method of big data, the risk state evaluation model based on the CTHMM continuous time hidden Markov model defines the safety state of the system as a hidden state, takes the feature multi-modal data as an observation sequence, and outputs a fire risk evaluation index, including: The risk state evaluation model based on the CTHMM continuous time hidden Markov model divides the safety state into four hidden states, including safe, low risk, medium risk, and high risk. The feature multi-modal data is mapped to a probability distribution by the GMM Gaussian mixture model, the transition probability matrix between hidden states is defined, and the historical fire data is initialized.

[0009] Further, in the above-mentioned fire safety monitoring and management method of big data, the risk state evaluation model based on the CTHMM continuous time hidden Markov model defines the safety state of the system as a hidden state, takes the feature multi-modal data as an observation sequence, and outputs a fire risk evaluation index, including: The transition probability and observation probability of the CTHMM are iteratively optimized by the EM expectation maximization algorithm until the log-likelihood function of the verification set converges. The most likely hidden state sequence is solved by the Viterbi algorithm, the posterior probability of each state is calculated, the hidden state is mapped to a risk index of 0-100, and the fire risk evaluation index is output.

[0010] Further, in the above-mentioned fire safety monitoring and management method of big data, based on the fire risk evaluation index, a hierarchical decision mechanism is used to classify fire risks, and a corresponding level of early warning mechanism is triggered according to the risk level, including: Based on the fire risk evaluation index, a hierarchical decision mechanism is used to classify fire risks, including first-level risk, second-level risk, third-level risk, and fourth-level risk.

[0011] Further, in a fire safety monitoring and management system of big data, the fire safety monitoring and management system includes the following modules: A multi-modal data acquisition module is used to collect environmental perception data, device state data, building and personnel dynamic data, and external and historical data, and to sequentially clean, denoise, and normalize the collected data to obtain standard multi-modal data. A feature alignment and fusion module is used to extract environmental trend features, spatial distribution features, and time series period features of the standard multi-modal data using an improved multi-scale convolutional neural network, and to align and fuse the features to obtain feature multi-modal data. a risk index evaluation module configured to establish a risk state evaluation model based on a CTHMM continuous time hidden Markov model, define a safety state of the system as a hidden state, take the feature multi-modal data as an observation sequence, and output a fire safety risk evaluation index; a fire safety grading module configured to grade the fire risk based on the fire safety risk evaluation index using a hierarchical decision mechanism, and trigger a corresponding level of early warning mechanism according to the risk level.

[0012] Further, in the big data fire safety monitoring management system, the risk index evaluation module comprises the following sub-modules: an iteration sub-module configured to iteratively optimize transition probabilities and observation probabilities of the CTHMM using an EM expectation maximization algorithm until a log-likelihood function of a verification set converges; an output sub-module configured to solve a most probable hidden state sequence by a Viterbi algorithm, calculate posterior probabilities of each state, map the hidden state to a risk index of 0-100, and output the fire safety risk evaluation index.

[0013] Further, in the big data fire safety monitoring management system, the fire safety grading module comprises the following sub-modules: a grading sub-module configured to grade the fire risk based on the fire safety risk evaluation index using a hierarchical decision mechanism, including a first-level risk, a second-level risk, a third-level risk, and a fourth-level risk.

[0014] The beneficial effects are that, through multi-source collection and standardized preprocessing, the problems of traditional data fragmentation and low quality are solved, ensuring accurate and reliable analysis foundation; in the feature extraction link, a multi-scale CNN combined with an attention mechanism is improved to accurately capture multi-dimensional features of time, space, and trend, and after fusion, data correlation is strengthened, breaking through the limitations of single feature analysis and improving the comprehensiveness of risk identification; in the risk evaluation stage, the CTHMM model converts the safety state into a quantitative index, accurately divides the risk, effectively reduces false positives and false negatives, avoids resource waste or omission of hidden dangers; the hierarchical decision and early warning mechanism matches the response measures according to the risk level, greatly shortens the response time, and at the same time, the closed-loop management feeds back the model iteration, continuously optimizes the monitoring accuracy. The overall realization of fire management from passive disposal to active early warning significantly reduces the probability of fire occurrence. BRIEF DESCRIPTION OF DRAWINGS

[0015] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not considered as limiting the application.

[0016] Figure 1A first embodiment schematic view of a fire safety monitoring management method for big data in the embodiment of the present application is shown in the figure. Figure 2 A second embodiment schematic view of a fire safety monitoring management method for big data in the embodiment of the present application is shown in the figure. Figure 3 A first embodiment schematic view of a fire safety monitoring management system for big data in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0018] Those skilled in the art can understand that, unless specifically stated, the singular form "one", "an", and "the" used herein can also include the plural form. It should be further understood that the use of the word "comprise" in the specification of the present application means that the features, integers, steps, operations, elements, and / or components described in the specification exist, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present application is described in detail below in combination with the drawings, Figure 1 A fire safety monitoring management method for big data is shown in the figure, which comprises the following steps: Step 101, collecting environmental perception data, equipment state data, building and personnel dynamic data, and external and historical data, sequentially cleaning, denoising and normalizing the collected data to obtain standard multi-modal data; Specifically, in the present embodiment, environmental perception data, equipment state data, building and personnel dynamic data, and external and historical data are collected, the environmental perception data includes temperature, humidity, smoke concentration, harmful gas concentration and light intensity, the equipment state data includes equipment operating parameters, fault codes and maintenance records, the building and personnel dynamic data includes fire-fighting facility distribution, personnel density and flow trajectory, and the external and historical data includes fire cases, hidden danger records and early warning disposal results; short-time missing data is completed by using linear interpolation method, abnormal data is identified based on 3σ principle, real mutation signals are retained by wavelet transform denoising, video trajectory data is smoothed by Kalman filter to remove jump points caused by occlusion, high-frequency fluctuation parameters are subjected to moving average to obtain preprocessed data; the preprocessed data is mapped to the [0, 1] interval based on Min-Max normalization to obtain standard multi-modal data.

[0020] The comprehensiveness and pertinence of the data are achieved through classified collection: the environmental perception data are obtained by a sensor network deployed in various areas of the building, and the core includes temperature, precision ±0.5℃, humidity, precision ±3%RH, smoke concentration, unit mg / m³, harmful gas concentration, such as CO, unit ppm, and light intensity, etc. key indicators, and the collection frequency is dynamically adjusted according to the regional risk level. High-risk areas, such as power distribution rooms and flammable material warehouses, collect data every 1-5 seconds to capture real-time changes. Ordinary areas, such as office areas and corridors, collect data every 10-30 seconds to balance monitoring accuracy and resource consumption. Device status data are collected in real time through the Internet of Things interface of fire-fighting equipment, such as sprinkler systems, smoke alarms, fire extinguishers, and fire pumps, including operating parameters such as sprinkler water pressure and pump current, as well as device fault codes, maintenance records, etc. Once the device status changes, such as abnormal pressure or fault triggering, it is immediately uploaded to ensure that the device status can be tracked in real time. Building and personnel dynamic data are divided into static and dynamic data. Building static data are imported through BIM models, including floor structure, room layout, evacuation passage location, fire extinguisher location, and fire hydrant distribution, etc. information. The building is updated synchronously within 24 hours after the reconstruction to ensure the accuracy of the space data. Personnel dynamic data are identified by video monitoring and face recognition technology to identify personnel identity, heat map to present density distribution, and access control system to collect, including personnel density, flow trajectory, and residence time, etc. The collection frequency is linked with the environmental perception data in the same area to realize the synchronization of space-time data. External and historical data are used as supplementary support. External data are connected with the meteorological department to obtain rainfall, wind speed, and other environmental factors that affect fire risk and city fire pipe network pressure monitoring system, updated once a day to reflect external environmental changes. Historical data are imported into fire cases, hidden danger records, and early warning disposal results of the building and similar buildings in the past 5 years, updated quarterly to provide historical reference for risk assessment. In terms of collection device deployment and interface specification, sensors use LoRa / NB-IoT low-power protocol to reduce energy consumption and ensure data transmission stability. At the same time, at least one comprehensive environmental sensor is deployed per 50㎡ to ensure that there are no monitoring dead angles in the building, and all sensors are connected to the edge gateway to realize data aggregation. The device interface strictly follows the "Technical Standard for Fire Fighting Facilities Internet of Things System", supports Modbus / OPCUA protocol, and ensures that the data of different brands and types of fire-fighting equipment can be compatible and transmitted, realizing real-time data upload to the cloud platform. The data storage adopts edge cache + distributed storage dual mechanism. The collected data are cached to the edge node to avoid data loss caused by network interruption, and then synchronized to the distributed database such as HBase, and the raw data backup is kept for at least 6 months to provide basis for subsequent data tracing and model iteration.

[0021] The collected raw data is systematically optimized to eliminate data defects and unify data formats, and finally a standard multi-modal data set meeting the model input requirements is generated. In the data cleaning stage, common data problems are classified and processed: missing value processing distinguishes between short-term and long-term missing values. When the data missing time is ≤5 minutes, linear interpolation is used to complete the data based on the effective data before and after the missing data to ensure data continuity. When the missing time is >5 minutes, the data in this section is marked as invalid, and at the same time, device failure warning is triggered to remind staff to troubleshoot sensor or transmission link problems. Outlier rejection is based on the 3σ principle, such as extreme high temperature and sudden humidity caused by sensor false alarms, etc., combined with device status for further verification. If the device is offline, the abnormal data is determined to be caused by transmission or device failure and is directly excluded. If the device is running normally, the data is retained and marked for verification by the staff to confirm whether it is a real anomaly. Format standardization unifies the three dimensions of data: time stamp is unified to UTC+8 time zone to avoid cross-time zone data confusion; data units are unified, such as temperature is converted to ℃, concentration is unified to standard units to eliminate dimension difference hazards; data structure adopts JSON format, including collection time-device ID-region ID-data value core field to ensure data traceability and easy parsing. In the denoising processing stage, different algorithms are used according to the noise characteristics of different data types: there is noise in environmental perception data caused by high-frequency jitter of the sensor, such as small temperature fluctuations, which is denoised by wavelet transform to filter out high-frequency noise while preserving real risk signals such as sudden increase in smoke concentration and sudden increase in temperature; personnel dynamic data may have trajectory jump points due to video monitoring obstruction, which is smoothed by Kalman filter algorithm to correct abnormal points caused by obstruction and restore the real flow path of personnel; high-frequency fluctuation parameters in device status data are processed by moving average method to extract the steady-state characteristics of device operation and avoid instantaneous fluctuation interference analysis. In the normalization processing stage, different methods are used for different types of data: numerical data is normalized by Min-Max normalization to map the data to the [0,1] interval, such as temperature conversion according to the rule of 0℃→0, 80℃→1 to eliminate the influence of different index dimension differences on model training; categorical data is encoded by One-HotEncoding to convert discrete categories into a vector form that the model can recognize. Finally, the pre-processing stage outputs a standard multi-modal data set containing three types of features: time series features, spatial features, and static attributes, providing high-quality, standardized input data for subsequent multi-modal feature extraction and risk assessment models.

[0022] Step 102, using the improved multi-scale convolutional neural network to extract environmental trend features, spatial distribution features and time series period features of the standard multi-modal data, and aligning and fusing the features to obtain feature multi-modal data; Specifically, in the improved multi-scale convolutional neural network in this embodiment, three convolution kernels are set, and SE attention mechanism is added after the convolution layer to enhance the weight of key features; 1x3 convolution kernel is used to process temperature and humidity time series data to extract rising and falling trends to obtain environmental trend features; 3x3 convolution kernel is used to convolve building area grid data to extract high-risk area clustering features to obtain spatial distribution features; 5x5 convolution kernel is used in combination with a sliding window to extract daily and weekly regular fluctuations to obtain time series period features. Linear interpolation is used to unify the step of environmental trend features, spatial distribution features and time series period features with different sampling frequencies to obtain unified feature data; the unified feature data is spliced in the channel dimension, and Cross-Attention cross-attention mechanism is used to learn the correlation between features to output feature multi-modal data.

[0023] The improved multi-scale CNN model construction link designs the core architecture and clearly defines the execution process around feature accurate extraction. In the model structure design in 2.1.1, three types of differentiated convolution kernels are used to realize multi-dimensional feature capture: 1x3 convolution kernel focuses on temperature, humidity and other time series data, which can effectively extract the short-term rising or falling trend of environmental parameters, such as the risk signal of continuous temperature rise; 3x3 convolution kernel is used for building area grid data, such as smoke concentration distribution matrix of each room, which can accurately identify the spatial clustering features of high-risk areas, such as local smoke concentration increase on a certain floor; 5x5 convolution kernel combined with a 1-hour sliding window processes data, which can capture daily or weekly long-period regular fluctuations, such as the periodic characteristics of sudden increase in personnel density during the morning peak period on weekdays. To improve feature effectiveness, the model adds key improvements: SE attention mechanism is embedded after the convolution layer to strengthen the weight of key risk features such as smoke concentration and high temperature, while suppressing irrelevant noise features such as weak light changes, further optimizing feature extraction accuracy. In the feature extraction process in 2.1.2, the input data needs to be first reorganized according to the time-space dimension of the standard multi-modal data to form the environmental, equipment and personnel data matrix of each building area at a certain time; then the model is forward propagated to output three types of feature maps corresponding to trend, space and period features, and the dimensions of the three types of feature maps are unified to [batch_size, 64, 64, 128], laying a foundation for subsequent feature processing.

[0024] The multi-source features are cooperatively integrated by a two-step process to form comprehensive risk features. In the feature alignment stage, the following are optimized respectively to address the heterogeneity of multi-modal data: in terms of time alignment, linear interpolation is used to unify all features to 10 seconds per step to ensure data synchronization in the time dimension, in the face of feature data with different sampling frequencies; in terms of spatial alignment, the features extracted from each region are mapped to a unified spatial grid with a precision of 1 m x 1 m based on the coordinates of the building BIM model, to ensure accurate correspondence of features in different regions in the spatial position and avoid feature misplacement due to differences in spatial reference. In the feature fusion stage, feature integration is completed in two stages: primary fusion is achieved by channel dimension splicing, which combines the three types of feature maps, trend, space and cycle, at the channel level to achieve preliminary aggregation of features; advanced fusion introduces a cross-attention mechanism, Cross-Attention, which learns the correlation between different features to uncover potential risk associations between features, and finally outputs a fusion feature vector with a dimension of [batch_size, 1024]. This step finally generates feature multi-modal data containing spatio-temporal correlation information, providing comprehensive and cooperative feature input for the subsequent risk assessment model.

[0025] Step 103, establishing a risk state assessment model based on a CTHMM continuous time hidden Markov model, defining the safety state of the system as a hidden state, taking the feature multi-modal data as an observation sequence, and outputting a fire risk assessment index; Specifically, in this embodiment, the risk state assessment model based on the CTHMM continuous time hidden Markov model divides the safety state into four hidden states, including safe, low risk, medium risk and high risk; the feature multi-modal data is mapped to a probability distribution by a GMM Gaussian mixture model, a transition probability matrix between hidden states is defined, and initialization is combined with historical fire data. The transition probability and observation probability of the CTHMM are iteratively optimized using the EM expectation maximization algorithm until the log-likelihood function of the validation set converges; the most likely hidden state sequence is solved by the Viterbi algorithm, the posterior probability of each state is calculated, the hidden state is mapped to a risk index of 0-100, and the fire risk assessment index is output.

[0026] Model parameter definition is the basis for the risk state evaluation model based on CTHMM to work, focusing on defining hidden states, observation sequences, and transition probabilities. In the hidden state, the model divides the system security state into four levels of clear hidden states, ensuring accurate and controllable risk classification: S0 is the safe state, representing all monitoring indicators such as temperature, smoke concentration, and equipment operating parameters within the normal range, with no risk hidden dangers; S1 is the low-risk state, with only a single indicator showing slight abnormalities, such as a local area temperature slightly higher than the conventional threshold, but without other abnormal signals; S2 is the medium-risk state, characterized by multiple indicators showing abnormalities simultaneously, such as a sustained increase in temperature accompanied by an upward trend in smoke concentration, with enhanced risk correlation; S3 is the high-risk state, which has approached the conditions for fire occurrence, with typical characteristics of sustained high temperature, smoke concentration exceeding the standard, and some fire-fighting equipment may respond abnormally. In the observation sequence and transition probability setting, the observation sequence directly uses the feature multi-modal data vector output in step 2, which is mapped to a probability distribution that meets the model's requirements through the Gaussian Mixture Model (GMM), achieving the conversion of feature data to observation signals; the state transition probability matrix is initialized based on historical fire data, following the actual laws of risk evolution, such as the probability of S1, low-risk, transferring to S2, medium-risk, being significantly higher than the probability of S0, safe, transferring to S3, high-risk, ensuring that state switching conforms to the risk progression logic in real-world scenarios.

[0027] Through scientific data division and algorithm optimization, the accuracy and stability of the model evaluation are ensured. In the data set construction stage, the model uses a historical data set containing building normal operation data and past fire case data, considering both risk-free and risky scenarios, ensuring the comprehensiveness of the training samples. The data set is divided into a training set and a validation set in a 7:3 ratio, with 70% and 30% respectively, and each group of data is labeled with the corresponding hidden state label, such as labeling the monitoring data 10 minutes before the fire as S3, high-risk, according to the fire development law, to provide a clear supervision learning goal for the model. In the parameter optimization stage, the model uses the Expectation-Maximization (EM) algorithm for iterative optimization, adjusting two key parameters of CTHMM: the state transition probability, which optimizes the probability values of switching between different hidden states; and the observation probability, which adjusts the probability distribution of feature multi-modal data and observation sequences for each hidden state. During the iteration process, the log-likelihood function loss of the validation set is used as the convergence judgment standard, and when the loss value no longer significantly decreases and tends to be stable, the iteration is stopped, at which point the model parameters have reached the optimal state, accurately capturing the association between data and risk states.

[0028] The abstract state judgment is converted into intuitive and timely quantitative risk indicators. In the real-time inference stage, the model continuously receives the feature multi-modal data output in step 2, and solves the most possible hidden state sequence corresponding to the current data through the Viterbi algorithm. The algorithm can filter out the sequence with the highest probability from numerous potential state sequences, and calculate the posterior probability of each hidden state, further clarify the confidence of the current safety state, and avoid the deviation of single state judgment. In the risk assessment index calculation, the model converts the abstract hidden state into a quantitative risk index of 0-100, establishes a clear mapping relationship: S0, safe corresponds to 0-20, S1, low risk corresponds to 21-40, S2, medium risk corresponds to 41-70, S3, high risk corresponds to 71-100, so that the management personnel can intuitively perceive the risk degree. In order to ensure real-time response to risk changes, the risk assessment index is updated every 10 seconds, which can dynamically track the subtle changes of the risk in the building and timely capture the risk escalation signal, and provide accurate and timely quantitative basis for subsequent risk level-based hierarchical decision and early warning mechanism.

[0029] Step 104, based on the fire risk assessment index, using hierarchical decision mechanism to classify fire risk, and triggering corresponding level of early warning mechanism according to risk level.

[0030] Specifically, in the present embodiment, based on the fire risk assessment index, the hierarchical decision mechanism is used to classify the fire risk, including first-level risk, second-level risk, third-level risk and fourth-level risk.

[0031] With the risk assessment index as the core, the four-level risk classification standard is first defined, then the decision-making logic and execution plan are designed, and finally the continuous optimization is realized through closed-loop management, forming a whole-process risk response system. Among them, the risk classification standard is based on the risk index of 0-100: the first level of risk corresponds to the index of 0-20, representing no significant risk, and the system only needs to maintain normal monitoring; the second level of risk corresponds to 21-40, which is a slight abnormality, mostly caused by device false alarm and other occasional situations, and needs to be locally investigated; the third level of risk corresponds to 41-70, which is a potential hidden danger, and there are signs of risk that need to be intervened on site, such as local index abnormalities superimposed with device state fluctuations; the fourth level of risk corresponds to 71-100, which is a high fire risk, close to the conditions of fire occurrence, and needs emergency response. Based on this classification standard, a targeted hierarchical decision-making mechanism is established: for the first level of risk, the system automatically records the monitoring data, does not trigger additional warnings, and only generates a regular monitoring report every hour, balancing management efficiency and resource consumption; for the second level of risk, the decision-making logic first compares historical data to determine whether it is a temporary fluctuation of the sensor or an occasional anomaly, and the execution layer pushes a prompt to the property management personnel APP, requiring on-site verification and feedback within 1 hour to timely eliminate false alarms or handle minor problems; for the third level of risk, the decision-making is combined with spatial features to accurately locate high-risk areas, such as the 3rd floor east office, and is associated with the running status of electrical equipment in that area, and in the execution, regional sound and light alarms are triggered, and work orders are pushed to fire patrol personnel to require on-site disposal within 30 minutes, while non-essential power in that area is turned off to reduce risk; for the fourth level of risk, the decision-making logic focuses on the activation of emergency plans, calculates the optimal evacuation path based on real-time personnel distribution and evacuation passage status, and the execution is divided into internal and external linkage: internal activation of full-building sound and light alarms, with emergency broadcast repeating evacuation instructions, automatic elevator drop to the first floor, and automatic unlocking of access control to ensure smooth evacuation; external linkage automatically dials 119, synchronously uploads the building's precise location, current risk index, real-time personnel density, and other key information, and timely notifies the property manager and the jurisdictional fire command center to ensure multi-party coordinated response. At the same time, in order to realize the continuous optimization of the mechanism, a warning closed-loop management process is established: the warning triggering time, specific disposal process and final result are recorded throughout the process, such as third-level risk → 20 minutes to eliminate faults, which will be fed back to the historical database to update the transition probability of the CTHMM model every month, continuously improving the accuracy of risk assessment and decision-making, forming a complete closed loop of monitoring, warning, disposal and optimization.

[0032] Its beneficial effects lie in solving the problems of traditional data fragmentation and low quality through multi-source collection and standardized preprocessing, ensuring accurate and reliable analysis foundation; in the feature extraction link, improving multi-scale CNN combined with attention mechanism, accurately capturing multi-dimensional features of time, space and trend, strengthening data correlation after fusion, breaking through the limitations of single feature analysis and improving the comprehensiveness of risk identification; in the risk assessment stage, the CTHMM model converts the safety state into a quantitative index, realizes accurate division of risk, effectively reduces false positives and false negatives, and avoids resource waste or omission of hidden dangers; the hierarchical decision and early warning mechanism matches the response measures according to the risk level, greatly shortens the response time, and at the same time, the closed-loop management counteracts the model iteration and continuously optimizes the monitoring accuracy. The whole realizes the transformation of fire management from passive disposal to active early warning, and significantly reduces the probability of fire occurrence.

[0033] Please refer to Figure 2 In a big data fire safety monitoring management method, an improved multi-scale convolutional neural network is used to extract environmental trend features, spatial distribution features and time sequence period features of standard multi-modal data, and the features are aligned and fused to obtain feature multi-modal data, including the following steps: Step 201, the improved multi-scale convolutional neural network is set with three convolution kernels, and an SE attention mechanism is added after the convolution layer to enhance the weight of key features; Step 202, the temperature and humidity time sequence data are processed by a 1x3 convolution kernel to extract the rising and falling trend to obtain the environmental trend features; Step 203, a 3x3 convolution kernel is used to convolve the building area grid data to extract high-risk area clustering features to obtain the spatial distribution features; Step 204, based on a 5x5 convolution kernel combined with a sliding window, the daily and weekly regular fluctuations are extracted to obtain the time sequence period features.

[0034] The above describes an embodiment of the big data fire safety monitoring management method of the application, please refer to Figure 3 In a big data fire safety monitoring management system, the fire safety monitoring management system includes the following modules: A multi-modal data acquisition module is used to collect environmental perception data, equipment state data, building and personnel dynamic data and external and historical data, and sequentially clean, denoise and normalize the collected data to obtain standard multi-modal data; A feature alignment and fusion module is used to extract environmental trend features, spatial distribution features and time sequence period features of standard multi-modal data by using an improved multi-scale convolutional neural network, and align and fuse the features to obtain feature multi-modal data; a risk index evaluation module, configured to establish a risk state evaluation model based on a CTHMM continuous time hidden Markov model, define the safety state of the system as a hidden state, take the feature multi-modal data as an observation sequence, and output a fire-fighting risk evaluation index; a fire safety grading module, configured to grade the fire risk based on the fire-fighting risk evaluation index by using a hierarchical decision mechanism, and trigger a corresponding level of early warning mechanism according to the risk level.

[0035] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and equivalents thereof.

Claims

1. A big data-based fire safety monitoring and management method, characterized in that, The fire safety monitoring and management method includes the following steps: Collect environmental perception data, equipment status data, building and personnel dynamic data, and external and historical data. The collected data are then cleaned, denoised, and normalized to obtain standard multimodal data. An improved multi-scale convolutional neural network is used to extract environmental trend features, spatial distribution features, and temporal periodic features from the standard multimodal data, and the features are aligned and fused to obtain feature multimodal data. A risk status assessment model based on the CTHMM continuous-time hidden Markov model is established, the safety status of the system is defined as the hidden state, the feature multimodal data is used as the observation sequence, and the fire risk assessment index is output. Based on the aforementioned fire risk assessment index, a hierarchical decision-making mechanism is used to classify fire risks, and a corresponding level of early warning mechanism is triggered according to the risk level.

2. The fire safety monitoring and management method based on big data as described in claim 1, characterized in that, The collected environmental perception data, equipment status data, building and personnel dynamic data, and external and historical data are sequentially cleaned, denoised, and normalized to obtain standard multimodal data, including: Collect environmental perception data, equipment status data, building and personnel dynamic data, and external and historical data. Environmental perception data includes temperature, humidity, smoke concentration, harmful gas concentration, and light intensity. Equipment status data includes equipment operating parameters, fault codes, and maintenance records. Building and personnel dynamic data includes the distribution of fire protection facilities, personnel density, and movement trajectories. External and historical data includes fire cases, hazard records, and early warning and response results. Linear interpolation is used to complete short-term missing data, abnormal data is identified based on the 3σ principle, wavelet transform is used to denoise and retain real abrupt change signals, Kalman filtering is used to smooth video trajectory data, occlusion-induced jump points are removed, and high-frequency fluctuation parameters are averaged to obtain preprocessed data. Preprocessed data is mapped to the [0,1] interval based on Min-Max normalization to obtain standard multimodal data.

3. The fire safety monitoring and management method based on big data as described in claim 1, characterized in that, The improved multi-scale convolutional neural network is used to extract environmental trend features, spatial distribution features, and temporal periodic features from the standard multimodal data, and the features are aligned and fused to obtain feature multimodal data, including: The improved multi-scale convolutional neural network sets up three types of convolutional kernels and adds an SE attention mechanism after the convolutional layer to enhance the weights of key features; Temperature and humidity time series data are processed by 1×3 convolution kernels to extract rising and falling trends and obtain environmental trend features; High-risk area clustering features are extracted by convolving the building area grid data with 3×3 convolution kernels to obtain spatial distribution features; Based on the 5×5 convolution kernel combined with the sliding window, the daily and weekly regular fluctuations are extracted to obtain the time-series periodic features.

4. The fire safety monitoring and management method based on big data as described in claim 1, characterized in that, The improved multi-scale convolutional neural network is used to extract environmental trend features, spatial distribution features, and temporal periodic features from the standard multimodal data, and the features are aligned and fused to obtain feature multimodal data, including: By unifying the step size of environmental trend characteristics, spatial distribution characteristics and temporal periodic characteristics at different sampling frequencies through linear interpolation, unified feature data is obtained. The unified feature data is concatenated along the channel dimension, and the correlation between features is learned through the Cross-Attention mechanism to output multimodal feature data.

5. The fire safety monitoring and management method based on big data as described in claim 1, characterized in that, The risk state assessment model based on the CTHMM continuous-time hidden Markov model is established, defining the system's safety state as a hidden state, using the characteristic multimodal data as an observation sequence, and outputting a fire risk assessment index, including: The risk state assessment model based on the CTHMM continuous-time hidden Markov model divides the safe state into four hidden states: safe, low risk, medium risk and high risk. The feature multimodal data is mapped to a probability distribution using a Gaussian mixture model (GMM), a transition probability matrix between hidden states is defined, and historical fire data is used for initialization.

6. The fire safety monitoring and management method based on big data as described in claim 1, characterized in that, The risk state assessment model based on the CTHMM continuous-time hidden Markov model is established, defining the system's safety state as a hidden state, using the characteristic multimodal data as an observation sequence, and outputting a fire risk assessment index, including: The transition probability and observation probability of CTHMM are iteratively optimized using the EM expectation-maximization algorithm until the log-likelihood function of the validation set converges. The Viterbi algorithm is used to solve for the most likely sequence of hidden states, calculate the posterior probability of each state, map the hidden states to a risk index of 0-100, and output the fire risk assessment index.

7. The fire safety monitoring and management method based on big data as described in claim 1, characterized in that, The method of classifying fire risks based on the fire risk assessment index using a hierarchical decision-making mechanism and triggering corresponding early warning mechanisms according to the risk level includes: Based on the aforementioned fire risk assessment index, a hierarchical decision-making mechanism is used to classify fire risks into four levels: Level 1 risk, Level 2 risk, Level 3 risk, and Level 4 risk.

8. A big data-based fire safety monitoring and management system, characterized in that, The fire safety monitoring and management system includes the following modules: The multimodal data acquisition module is used to collect environmental perception data, equipment status data, building and personnel dynamic data, and external and historical data. The collected data is cleaned, denoised, and normalized in sequence to obtain standard multimodal data. The feature alignment and fusion module is used to extract the environmental trend features, spatial distribution features and temporal periodic features of the standard multimodal data using an improved multi-scale convolutional neural network, and to align and fuse the features to obtain feature multimodal data. The risk index assessment module is used to establish a risk state assessment model based on the CTHMM continuous-time hidden Markov model, define the safety state of the system as the hidden state, use the feature multimodal data as the observation sequence, and output the fire risk assessment index. The fire safety classification module is used to classify fire risks based on the fire risk assessment index using a hierarchical decision-making mechanism, and to trigger the corresponding level of early warning mechanism according to the risk level.

9. A big data-based fire safety monitoring and management system as described in claim 8, characterized in that, The risk index assessment module includes the following sub-modules: The iterative submodule is used to iteratively optimize the transition probability and observation probability of CTHMM using the EM expectation-maximization algorithm until the log-likelihood function of the validation set converges. The output submodule is used to solve for the most likely hidden state sequence using the Viterbi algorithm, calculate the posterior probability of each state, map the hidden states to a risk index of 0-100, and output the fire risk assessment index.

10. A big data-based fire safety monitoring and management system as described in claim 8, characterized in that, The fire safety classification module includes the following sub-modules: The classification submodule is used to classify fire risks based on the fire risk assessment index using a hierarchical decision-making mechanism, including Level 1 risk, Level 2 risk, Level 3 risk, and Level 4 risk.

Images