Intelligent safety elimination integrated big data platform system

Abstract

The invention, which relates to the technical field of fire protection and security management, discloses an intelligent fire protection and security integrated big data platform system comprising a data acquisition module, a data storage and preprocessing module, an intelligent analysis module and a linkage control module. The data acquisition module integrates fire-fighting and security sensors and equipment, and transmits data to the convergence unit in a wired and wireless communication mode, so that one-stop acquisition of fire-fighting and security basic data is realized; according to the invention, the intelligent analysis module can carry out fire prediction probability evaluation in combination with security video image features, carries out comprehensive analysis on security videos and time series data through CNN and LSTM technologies, analyzes the internal relation of fire protection and security data through an association rule mining algorithm, and generates a risk early warning report; according to the intelligent analysis method, the accuracy and timeliness of safety early warning are improved, potential safety threats can be found and processed in time, and safety accidents such as fire disasters and the like are effectively prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of fire protection and security management, and in particular to a smart fire protection and security integrated big data platform system. Background Art

[0002] In traditional fire safety and security monitoring systems, there is usually a data island phenomenon, that is, the fire protection and security systems operate independently and lack effective data integration and information sharing mechanisms. This separation results in the security system being unable to provide auxiliary information in a timely manner in emergency situations, such as when a fire occurs, and the fire protection system being unable to use security data for more accurate risk assessment and emergency response. In addition, traditional fire protection systems mostly rely on simple sensor alarms and lack intelligent analysis and prediction capabilities, and are unable to effectively predict and warn of potential fire risks. The security system also has the problem of insufficient intelligence, making it difficult to identify and analyze abnormal behaviors in complex scenarios, resulting in insufficiently timely and accurate responses to potential security threats; With the development of technologies such as the Internet of Things, big data, cloud computing, and artificial intelligence, the demand for intelligent and integrated fire protection and security systems is growing. The development of these technologies has made it possible to achieve deep integration and intelligent analysis of fire protection and security data, making it possible to build a smart fire protection and security integrated big data platform system. The system can integrate fire protection and security sensors and equipment, and through advanced data processing and analysis technologies, realize one-stop collection, storage, analysis, and response of fire protection and security data. However, existing technical solutions have not yet been able to fully meet this demand, especially in terms of intelligent analysis and cross-system linkage, and lack effective solutions.

[0003] In order to solve the above defects, a technical solution is now provided. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of data silos, insufficient intelligence and weak cross-system linkage capabilities in existing fire protection and security systems, and to propose an intelligent fire protection and security integrated big data platform system.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A smart fire safety integrated big data platform system, including: The data acquisition module integrates fire protection and security sensors and equipment, and transmits data to the aggregation unit through wired and wireless communication to achieve one-stop collection of basic fire protection and security data; The data storage and preprocessing module temporarily stores and preprocesses key data on the local edge device side, and classifies and persists historical data in the cloud data center to build a data warehouse; Intelligent analysis module, which builds an analysis engine based on deep learning and data mining algorithms, combines the characteristics of security video images to conduct unilateral fire prediction probability assessment, and then conducts comprehensive analysis of security video and time series data through CNN and LSTM technologies. The association rule mining algorithm analyzes the internal connection between fire protection and security data to generate risk warning reports; The linkage control module has preset emergency scenario linkage strategies and is integrated with fire protection and security hardware. It automatically links related equipment when a fire alarm occurs, and collaborates with abnormalities in daily security management and control to improve emergency response efficiency.

[0006] Furthermore, the execution process of the intelligent analysis module is as follows: Build a core analysis engine based on deep learning and data mining algorithms; use convolutional neural network models to analyze security video images frame by frame to identify smoke, open flames, and abnormal human behavior characteristics; Combine the security video image features to conduct a unilateral fire prediction probability assessment to determine the fire probability level. When the fire probability level is greater than the preset standard, a comprehensive assessment of the fire is conducted; By combining recurrent neural networks including long short-term memory networks to mine time series data associations, the probability of fire and security risks in each area is comprehensively evaluated; The association rule mining algorithm analyzes the intrinsic relationship between the status of fire protection facilities and security anomalies, and generates a risk warning report.

[0007] Furthermore, the specific operation steps of the intelligent analysis module to determine the probability level of fire occurrence are as follows: Continuously collect video streams from security cameras and extract single-frame images at a preset frame rate; perform image correction to correct problems caused by camera shooting angle tilt and lens distortion; use the chessboard calibration method to pre-acquire camera internal and external parameters, and correct the image to the normal view angle based on perspective transformation to ensure that the shape and position relationship of objects in the picture conform to the real scene proportion, reducing target recognition errors caused by distortion; A background model is constructed for the scene using a Gaussian mixture model. By continuously learning the distribution of pixel values ​​of multiple frames of images, the area that meets the statistical law of background pixels is determined as the background, and the area that differs from the background by more than the standard is considered as the foreground. Moving targets are separated, and human objects in the picture are preliminarily screened out. Static background interference is filtered out to focus on human behavior analysis. The human body key point detection algorithm is used to identify the coordinate information of the key parts of the human body, and the posture is judged according to the relative position changes of the joint points. The number of frames and frequency of abnormal postures are recorded and quantified into posture abnormality scores. Every time an abnormal action lasts for more than 5 frames, the abnormality score increases by 1. Track the overall contour movement trajectory of the person, use Kalman filtering or optical flow method to predict the position of the tracked person in the next frame, and analyze the trajectory smoothness and direction mutation; if the trajectory reversal exceeds the standard, and the acceleration towards the fire area exceeds the threshold, the corresponding abnormal indicators are also recorded. The degree of trajectory abnormality is measured by the multiple of the standard deviation of the deviation from the normal passage path. If it exceeds 2 times the standard deviation, it will be recorded as a trajectory abnormality, and the number of recorded trajectory abnormalities will be recorded as the out-of-path value; Analyze items carried by personnel in their hands or beside them based on image texture and shape feature extraction technology; extract local feature points of items through scale-invariant feature transformation and accelerated robust feature algorithm, and match them with the pre-stored common dangerous goods feature library; when the matching similarity exceeds the set threshold, identify it as a dangerous item, and count the proportion of the number of frames carrying dangerous items to the total number of observed frames as a quantitative indicator, and record it as the excess proportion. When the excess proportion exceeds 30%, it is considered a high-risk carrying situation; In the continuous video monitoring, the quantitative indicators of personnel behavior characteristics during the set time window are summarized and analyzed, and matched with the rule base one by one; when a single rule is met, it is initially marked according to the corresponding probability. If multiple rules are triggered at the same time, the abnormal score, outlier value and excess proportion obtained by analysis are marked as yif, yzz and css respectively, and the normalization process is substituted into the following formula: The disaster judgment value ZPJ is obtained, and the obtained disaster judgment value ZPJ is used as the standard for measuring the unilateral fire prediction probability evaluation combined with the security video image features. are the preset weight coefficients of anomaly score, outlier value and excess proportion respectively; the obtained disaster judgment value ZPJ is compared with several preset disaster judgment value intervals, and several disaster judgment value intervals correspond to different fire probability levels. When the disaster judgment value interval to which the disaster judgment value ZPJ belongs is determined, the fire probability level is determined. When the fire probability level is greater than the preset standard, a comprehensive assessment of the fire is performed.

[0008] Furthermore, the intelligent analysis module performs a comprehensive fire assessment process as follows: Combine recurrent neural networks including long short-term memory networks to mine time series data associations and comprehensively assess the probability of fire and security risks in each area; Clean the data on the operating parameters of fire-fighting equipment and the frequency of personnel movement in the concerned area; remove erroneous data points; then normalize the data to unify the scale range of data of different magnitudes; The long-term fire protection and security time series data are segmented according to fixed time windows to form data segments. At the same time, the corresponding risk labels are marked for each segment. Model construction, LSTM unit is composed of input gate, forget gate, output gate and memory unit; multiple LSTM units are connected in series in sequence to form a hidden layer. The input layer receives normalized time series data, and the dimension corresponds to the number of features in each time step; the output layer is designed according to the task. If the risk probability is simply predicted, a fully connected layer with a Sigmoid activation function is used to output the risk value; The model hyperparameters, including the number of hidden layers, the number of units in each hidden layer, and the learning rate, were determined after multiple optimizations; Model training: The labeled time series data is divided into training set, validation set, and test set. The training set is sent to the LSTM model for forward propagation to calculate the predicted risk probability, compared with the labeled real label, and the loss value is calculated using the cross entropy loss function; then the back propagation algorithm is used to calculate the gradient of each parameter according to the chain rule, and the model parameters are updated using the optimization algorithm. The training is iterated several times until the loss on the validation set no longer decreases, and the trained optimal model is saved; For the time series of fire equipment operation parameters, the trained LSTM model analyzes the new input data segment by segment and outputs the corresponding risk probability value. When the water pressure continues to drop and approaches the critical low water pressure and the model outputs a high risk probability, it indicates that there are hidden dangers in the operation of fire protection facilities. In terms of the frequency of personnel movement, if the number of personnel in the preset area increases to the threshold during non-working hours and the model warns of high risk, it indicates that there are security issues such as illegal intrusion and illegal operation; The risk probability of fire-fighting equipment operation is integrated with the risk probability of personnel flow and security-related risks, and a weighted sum is taken based on expert experience and the weight ratio mined from historical accident big data to ultimately determine the comprehensive assessment result.

[0009] Furthermore, the specific operation steps of the linkage control module are as follows: Preset multiple emergency scenario linkage strategies and integrate with fire protection and security hardware facilities; when the fire alarm is triggered, the security camera is automatically linked to focus on the alarm point and switch to thermal imaging mode to assist in observing the spread of the fire; Unlock the access control of the evacuation passage to ensure the smooth escape route, trigger the fire broadcast to prompt people to evacuate according to the preset evacuation voice, and at the same time push the building fire protection facility distribution map, real-time on-site images, and surrounding rescue force distribution to the fire command center to facilitate scientific command; In daily security management and control, when encountering access control intrusion or perimeter defense alarm, the camera is immediately instructed to track and shoot, light warnings are activated, and the status of surrounding fire-fighting facilities is checked to ensure emergency availability, thereby achieving cross-system collaborative operations for fire protection and improving overall emergency response efficiency.

[0010] Further, the determination of the security camera that automatically links the focused alarm point in the linkage control module is determined by determining the priority of the focused alarm point of the security camera; By formula: Calculate the camera focus priority , where , is the weight coefficient, It is the reciprocal of the distance from the camera to the alarm point; is the ratio of the camera's field of view covering the alarm area; the obtained camera focus priority Compare with the priority threshold, when the camera focuses on the priority When it is greater than a preset priority threshold, it is determined that the camera needs to be focused.

[0011] Furthermore, the specific operation steps for determining whether the security camera needs to switch to the thermal imaging mode in the linkage control module are as follows: By analyzing the smoke concentration change rate and the perceived ambient temperature change rate in the collected image data and normalizing them, the smoke concentration change rate and the perceived ambient temperature change rate are used as the two sides of a right triangle, the other side of the triangle is connected, and a triangular pyramid model is established with a preset correction coefficient as the height. The surface area of ​​the triangular pyramid model is calculated and recorded as the thermal judgment value. When the thermal judgment value is greater than the preset thermal judgment threshold, it is determined that the security camera needs to switch to thermal imaging mode.

[0012] Furthermore, the specific steps of the data acquisition module are as follows: Integrate various fire protection and security sensors and equipment to build a comprehensive data collection network; In terms of fire protection, smoke sensors, temperature sensors, combustible gas detectors, and electrical fire monitoring devices are installed in various areas of the building, and data is transmitted to the data aggregation unit through a combination of wired and wireless communications; In terms of security, surveillance cameras, access control systems, and electronic patrol equipment are deployed and connected to the data aggregation unit to achieve "one-stop" collection of basic fire and security data, laying a solid foundation for subsequent analysis.

[0013] Furthermore, the specific steps of the data storage and preprocessing module are as follows: The local edge device is equipped with a server or intelligent gateway with built-in storage media to temporarily store real-time and critical data. The judgment standard of critical data is to judge the level of data cache, obtain the time freshness of the data, that is, the time difference from the current time and quantify it into a score. The higher the score, the shorter the time difference. The data criticality index is quantified with a value between 0 and 1. The frequency of data generation, that is, the number of times it is generated per unit time. After normalizing the obtained time freshness, data criticality index and data generation frequency, the optimal judgment value is obtained by weighted summation, and the obtained optimal judgment value is compared with the preset optimal judgment threshold. When the optimal judgment value is greater than the preset optimal judgment threshold, the data is judged to be critical data and needs to be temporarily stored. Use lightweight algorithm preprocessing to ensure data quality and reduce the burden of cloud transmission and processing; The cloud data center relies on the collaborative architecture of distributed file systems, relational databases and non-relational databases to classify and persist historical data, build a data warehouse for retrieval, query and correlation analysis, and reserve materials for in-depth mining of safety hazards.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention realizes one-stop data collection and storage by integrating fire protection and security sensors and equipment. It uses advanced data storage and preprocessing modules to temporarily store and preprocess key data on local edge devices, while the cloud data center classifies and persists historical data to build a data warehouse. This design greatly improves data integration capabilities and provides rich data materials for in-depth mining of safety hazards. The intelligent analysis module conducts comprehensive analysis of security video and time series data based on deep learning and data mining algorithms, thereby improving the accuracy of fire risk prediction and response speed. (2) The present invention presets the emergency scenario linkage strategy through the linkage control module and integrates it with the fire protection and security hardware, thus realizing the automatic linkage of related equipment when a fire alarm is triggered, as well as the collaborative operation when encountering abnormalities in daily security control. This cross-system linkage mechanism significantly improves the emergency response efficiency and ensures that effective measures can be taken quickly in an emergency, such as automatically unlocking the evacuation channel access control, triggering the fire broadcast, etc., thereby ensuring the safety of personnel and reducing property losses. (3) In the present invention, the intelligent analysis module can combine the security video image features to evaluate the fire prediction probability, and use CNN and LSTM technology to conduct a comprehensive analysis of the security video and time series data. The association rule mining algorithm analyzes the internal connection between fire protection and security data to generate a risk warning report. The intelligent analysis method improves the accuracy and timeliness of security warnings, so that potential security threats can be discovered and handled in a timely manner, effectively preventing the occurrence of safety accidents such as fires. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 This is the overall system block diagram of the present invention. DETAILED DESCRIPTION

[0016] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0018] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0019] like Figure 1 As shown, a smart fire safety integrated big data platform system includes a data acquisition module, a data storage and preprocessing module, an intelligent analysis module and a linkage control module; The data acquisition module integrates fire protection and security sensors and equipment, and transmits data to the aggregation unit through wired and wireless communication, realizing one-stop collection of basic fire protection and security data; Integrate various fire protection and security sensors and equipment to build a comprehensive data collection network; in terms of fire protection, install smoke sensors, temperature sensors, combustible gas detectors, and electrical fire monitoring devices in various areas of the building (such as electrical fire monitoring detectors in the electrical distribution room to monitor leakage and overload; smoke and temperature alarms in key fire protection areas such as warehouses and kitchens), and transmit data to the data aggregation unit through a combination of wired (RS485 bus and CAN bus to ensure stable transmission and strong anti-interference, suitable for fixed points and areas with convenient wiring) and wireless (such as low-power Bluetooth and ZigBee networking for the renovation of old buildings or temporary construction sites that are difficult to wire, and 4G / 5G modules to ensure remote and high-speed data return) communication; In terms of security, surveillance cameras are deployed (covering entrances, corridors, public areas, and some special places are equipped with thermal imaging cameras to enhance the monitoring capabilities of day and night and smoke environments), access control systems (integrated with card swiping, fingerprint, and face recognition functions, recording people's entry and exit time, identity information), and electronic patrol equipment (supervising security patrol trajectories and time), and connected to data aggregation units to achieve "one-stop" collection of basic fire and security data, laying a solid foundation for subsequent analysis.

[0020] The data storage and preprocessing module uses the local edge device to temporarily store and preprocess key data, and the cloud data center classifies and persists massive historical data to build a data warehouse; The local edge device is equipped with a server or intelligent gateway with built-in storage media (such as solid-state hard disk, memory card) to temporarily store real-time and critical data (fire sensor values ​​in the past few minutes, current access control records); the judgment standard of critical data is to judge the level of data cache, obtain the time freshness of the data, that is, the time difference from the current time and quantify it into a score. The higher the score, the shorter the time difference; data criticality indicators, and quantify them with values ​​between 0 and 1; data generation frequency, that is, the number of times generated per unit time; after normalizing the obtained time freshness, data criticality indicators and data generation frequency, the optimal judgment value is obtained by weighted summation, and the obtained optimal judgment value is compared with the preset optimal judgment threshold. When the optimal judgment value is greater than the preset optimal judgment threshold, the data is judged to be critical data and needs to be temporarily stored; Pre-processing with lightweight algorithms (data filtering to remove noise interference and preliminary screening of outliers) ensures data quality and reduces the burden of cloud transmission and processing. Even a brief network outage will not affect the acquisition of local emergency operation basis. The cloud data center relies on a collaborative architecture of distributed file systems (HDFS to ensure reliable storage of massive videos and log files), relational databases (MySQL stores structured equipment parameters, personnel permissions, etc.) and non-relational databases (MongoDB stores semi-structured security video annotation information and fire inspection records), classifies and persists massive historical data, and builds a data warehouse for efficient retrieval, query, and correlation analysis, which reserves materials for in-depth mining of the laws of safety hazards.

[0021] The intelligent analysis module builds a core analysis engine based on deep learning and data mining algorithms. It analyzes security video and time series data through technologies such as CNN and LSTM. The association rule mining algorithm analyzes the internal connection between fire protection and security data and generates risk warning reports. The core analysis engine is built based on deep learning and data mining algorithms. The convolutional neural network (CNN) model is used to analyze security video images frame by frame to identify the characteristics of smoke, open flames, and abnormal human behavior (deliberate destruction of fire protection facilities, illegal use of fire). Combined with the characteristics of security video images, a unilateral fire prediction probability assessment is performed, specifically: Continuously collect video streams from security cameras and extract single-frame images at a preset frame rate (e.g. 10-25 frames per second, based on the balance between the dynamics of the scene and computing power). Since the camera may have problems such as tilted shooting angle and lens distortion, image correction must be performed first. Use the checkerboard calibration method to pre-acquire the camera's internal and external parameters, and correct the image to the normal view based on perspective transformation to ensure that the shape and position relationship of the objects in the picture conform to the real scene proportions, reducing target recognition errors caused by distortion; A background model is constructed for the scene using a Gaussian mixture model (GMM). By continuously learning the distribution of pixel values ​​of multiple frames of images, the area that conforms to the statistical laws of background pixels (such as a specific mean value and variance range) is determined as the background, and the area with large differences is considered as the foreground. Moving targets are separated, and human objects in the picture are preliminarily screened out. Static background interference is filtered out to focus on human behavior analysis. Use a human key point detection algorithm (such as the OpenPose algorithm principle, which extracts joint points based on a deep learning pre-training model. Here, we only draw on the idea and do not evaluate the complete model) to identify the coordinate information of key parts of the person's body (head, shoulders, elbows, hands, hips, knees, feet, etc.), and judge the posture based on the relative position changes of the joints. For example, if the hand position is continuously close to fire-fighting facilities (by comparing with the pre-marked positions of fire boxes, fire hydrants, etc.) and the elbow bending angle is in a grasping state, record the number of frames and frequency of occurrence of this abnormal posture, and quantify it into a posture anomaly score. Each time an abnormal posture lasts for more than 5 frames, the anomaly score increases by 1; track the overall contour movement trajectory of the person, use Kalman filtering or optical flow method to predict the position of the tracked person in the next frame, and analyze the trajectory smoothness and sudden change in direction; if the trajectory frequently turns back and suddenly accelerates to the fire-fighting area (distribution room, flammable and explosive storage room), the corresponding abnormal indicators are also recorded, and the degree of trajectory abnormality is measured by the multiple of the standard deviation of the deviation from the normal passage path. If the number of abnormal trajectories exceeds 2 times the standard deviation, it will be recorded as an abnormal trajectory. The number of abnormal trajectory records will be recorded as the outlier value. The items carried by the personnel in their hands or beside them are analyzed based on image texture and shape feature extraction technology. The local feature points of the items are extracted through the scale-invariant feature transform (SIFT) and speeded up robust feature (SURF) algorithms, and matched with the pre-stored feature library of common dangerous items (lighters, gasoline barrels, blowtorches, etc.). When the matching similarity exceeds the set threshold (such as 70%, optimized through a large number of sample tests), it is identified as a dangerous item. The proportion of the number of frames carrying dangerous items to the total number of observed frames is counted as a quantitative indicator and recorded as the excess proportion. When the excess proportion exceeds 30%, it is considered a high-risk carrying situation. In the continuous video monitoring, the quantitative indicators of personnel behavior characteristics during the set time window (such as every 30 seconds) are summarized and analyzed, and matched with the rule base one by one; when a single rule is met, it is initially marked according to the corresponding probability. If multiple rules are triggered at the same time, the abnormal score, outlier value and excess proportion obtained by analysis are marked as yif, yzz and css respectively, and the normalization process is substituted into the following formula: To obtain the disaster judgment value ZPJ, where are preset weight coefficients for abnormal score, outlier value and excess ratio respectively; the obtained disaster judgment value ZPJ is compared with several preset disaster judgment value intervals, and several disaster judgment value intervals correspond to different fire probability levels. When the disaster judgment value interval to which the disaster judgment value ZPJ belongs is determined, the fire probability level is determined. When the fire probability level is greater than the preset standard, a comprehensive assessment of the fire is performed; The comprehensive fire assessment process is as follows: Combined with recurrent neural networks (RNN) including long short-term memory networks (LSTM), the association of time series data (fluctuations in operating parameters of firefighting equipment over time, periodic changes in the frequency of personnel flow in specific areas) is mined to comprehensively evaluate the probability of fire and security risks in each area. The specific process is as follows: Cleaning operations are performed on the operating parameters of fire-fighting equipment (such as the water pressure value of fire pumps, the water flow rate of fire-fighting pipe networks, the leakage current value of electrical fire monitoring devices, etc.) and the frequency of personnel movement in the concerned areas; erroneous data points are removed; the data is then normalized to unify the scale range of data of different magnitudes; long-term fire-fighting and security time series data are segmented according to fixed time windows (for example, every 5 minutes, 10 minutes, etc., which can be flexibly set according to the frequency of data changes and actual scene requirements) to form data segments; at the same time, each segment is labeled with a corresponding risk label; Model construction: LSTM units are composed of input gates, forget gates, output gates, and memory units; multiple LSTM units are connected in series in sequence to form a hidden layer. The input layer receives normalized time series data (firefighting equipment operating parameters or personnel flow frequency data), and the dimension corresponds to the number of features in each time step (water pressure value, flow value, personnel count, etc.); the output layer is designed according to the task. If the risk probability is simply predicted, a fully connected layer is used with a Sigmoid activation function (the output is mapped to the 0-1 interval to represent the probability) to output the risk value; model hyperparameters such as the number of hidden layers (1-3 layers can be tried, and the overfitting and underfitting risks are weighed according to the complexity of the data), the number of units in each hidden layer (selected between 32-256 through experiments, and more units are used in scenarios with large data volumes and complex patterns), and the learning rate (initial setting is 0.001-0.01, and adjusted according to the convergence situation during training) are determined after multiple tuning; Model training: The labeled time series data is divided into training set (such as 70%-80%), validation set (10%-15%), and test set (10%-15%). The training set is sent to the LSTM model for forward propagation to calculate the predicted risk probability, and compared with the labeled real label. The cross entropy loss function (applicable to binary or multi-classification probability prediction scenarios, measuring the difference between the prediction and the real distribution) is used to calculate the loss value; then the back propagation algorithm is used to calculate the gradient of each parameter according to the chain rule, and the optimization algorithm (such as Adam, Adagrad and other adaptive learning rate algorithms) is used to update the model parameters. The training is repeated hundreds to thousands of times until the loss on the validation set no longer decreases (to prevent overfitting), and the trained optimal model is saved; For the time series of fire equipment operation parameters, the trained LSTM model analyzes the new input data segment by segment (such as the real-time collected water pressure fluctuation data of the continuous operation of the fire pump) and outputs the corresponding risk probability value. When the water pressure continues to drop and approaches the critical low water pressure (which may lead to ineffective fire extinguishing) and the model outputs a high risk probability (such as exceeding 0.7), it indicates that there are hidden dangers in the operation of fire protection facilities. In terms of the frequency of personnel flow, the number of people in a specific area (such as the night warehouse area) suddenly increases during non-working hours and the model warns of high risk, indicating that there may be security problems such as illegal intrusion and illegal operations. The risk probability of fire equipment operation is integrated and analyzed with the risk probability of personnel flow and security-related risks. The weighted sum is calculated based on the expert experience and the weight ratio mined from historical accident big data to finally determine the comprehensive evaluation result. Association rule mining algorithms (such as the Apriori algorithm) analyze the intrinsic connection between the status of fire-fighting facilities (insufficient water pressure, aging and failure of components) and security anomalies (frequent intrusion of unknown persons into a certain area, long-term stay in blind spots of surveillance) to generate risk warning reports.

[0022] The linkage control module presets emergency scenario linkage strategies and integrates with fire protection and security hardware. It automatically links related equipment when a fire alarm occurs. It can also work together when anomalies occur during daily security control, thus improving emergency response efficiency. Preset multiple emergency scene linkage strategies and integrate with fire protection and security hardware facilities; when the fire alarm is triggered, the security camera is automatically linked to focus on the alarm point and switch to thermal imaging mode to assist in observing the spread of the fire; the priority of the security camera focusing on the alarm point is determined by the formula: Calculate the camera focus priority , where , is the weight coefficient, It is the reciprocal of the distance from the camera to the alarm point; is the ratio of the camera's field of view covering the alarm area (value range 0-1); the obtained camera focus priority Compare with the priority threshold, when the camera focuses on the priority When it is greater than the preset priority threshold, it is determined that the camera needs to be focused; Whether the security camera needs to switch to thermal imaging mode is determined by analyzing the smoke concentration change rate and the perceived ambient temperature change rate in the collected image data. After normalization, the smoke concentration change rate and the perceived ambient temperature change rate are used as the two sides of a right triangle, and the other side of the triangle is connected. A triangular pyramid model is established with a preset correction coefficient as the height, and the surface area of ​​the triangular pyramid model is calculated and recorded as the thermal judgment value. When the thermal judgment value is greater than the preset thermal judgment threshold, it is determined that the security camera needs to switch to thermal imaging mode; wherein the preset correction coefficient value is set between 5.6-8.2; Unlock the access control of the evacuation passage (the electromagnetic lock is powered off to open) to ensure the smooth escape route, trigger the fire broadcast to prompt people to evacuate according to the preset evacuation voice, and at the same time push the building fire protection facility distribution map (fire hydrant, fire extinguisher location), real-time on-site images, and surrounding rescue force distribution (internal micro-fire station personnel and equipment, external fire brigade distance and estimated arrival time) to the fire command center to facilitate scientific command; In daily security control, when encountering forced intrusion at the access control or perimeter defense alarm, we will immediately instruct nearby cameras to track and shoot, set off light warnings, and check the status of surrounding fire-fighting facilities to ensure their availability for emergency situations, thus achieving cross-system collaboration for fire-fighting and security and improving overall emergency response efficiency.

[0023] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A smart fire safety integrated big data platform system, characterized in that: include: The data acquisition module integrates fire protection and security sensors and equipment, and transmits data to the aggregation unit through wired and wireless communication to achieve one-stop collection of basic fire protection and security data; The data storage and preprocessing module temporarily stores and preprocesses key data on the local edge device side, and classifies and persists historical data in the cloud data center to build a data warehouse; Intelligent analysis module, which builds an analysis engine based on deep learning and data mining algorithms, combines the characteristics of security video images to conduct unilateral fire prediction probability assessment, and then conducts comprehensive analysis of security video and time series data through CNN and LSTM technologies. The association rule mining algorithm analyzes the internal connection between fire protection and security data to generate risk warning reports; The linkage control module has preset emergency scenario linkage strategies and is integrated with fire protection and security hardware to automatically link related equipment when a fire alarm occurs.

2. The intelligent fire safety integrated big data platform system according to claim 1 is characterized in that: The execution process of the intelligent analysis module is as follows: Build a core analysis engine based on deep learning and data mining algorithms; use convolutional neural network models to analyze security video images frame by frame to identify smoke, open flames, and abnormal human behavior characteristics; Combine the security video image features to conduct a unilateral fire prediction probability assessment to determine the fire probability level. When the fire probability level is greater than the preset standard, a comprehensive assessment of the fire is conducted; By combining recurrent neural networks including long short-term memory networks to mine time series data associations, the probability of fire and security risks in each area is comprehensively evaluated; The association rule mining algorithm analyzes the intrinsic relationship between the status of fire protection facilities and security anomalies, and generates a risk warning report.

3. The intelligent fire safety integrated big data platform system according to claim 2 is characterized in that: The specific operation steps of the intelligent analysis module to determine the probability level of fire are as follows: Continuously collect video streams from security cameras, extract single-frame images at a preset frame rate, and perform image correction; use the chessboard calibration method to pre-acquire the camera's internal and external parameters, and correct the image to the normal view based on perspective transformation; A background model is constructed for the scene using a Gaussian mixture model. By continuously learning the distribution of pixel values ​​of multiple frames of images, the area that meets the statistical law of background pixels is determined as the background, and the area that differs from the background by more than the standard is considered as the foreground. Moving targets are separated, and the human objects in the picture are preliminarily screened out. The human body key point detection algorithm is used to identify the coordinate information of the key parts of the human body, and the posture is judged according to the relative position changes of the joint points. The number of frames and frequency of abnormal postures are recorded and quantified into posture abnormality scores. Every time an abnormal action lasts for more than 5 frames, the abnormality score increases by 1. Track the overall contour movement trajectory of the person, use Kalman filtering or optical flow method to predict the position of the tracked person in the next frame, and analyze the trajectory smoothness and direction mutation; if the trajectory reversal exceeds the standard, and the acceleration towards the fire area exceeds the threshold, the corresponding abnormal indicators are also recorded. The degree of trajectory abnormality is measured by the multiple of the standard deviation of the deviation from the normal passage path. If it exceeds 2 times the standard deviation, it will be recorded as a trajectory abnormality, and the number of recorded trajectory abnormalities will be recorded as the out-of-path value; Analyze items carried by personnel in their hands or beside them based on image texture and shape feature extraction technology; extract local feature points of items through scale-invariant feature transformation and accelerated robust feature algorithm, and match them with the pre-stored common dangerous goods feature library; when the matching similarity exceeds the set threshold, identify it as a dangerous item, and count the proportion of the number of frames carrying dangerous items to the total number of observed frames as a quantitative indicator, and record it as the excess proportion. When the excess proportion exceeds 30%, it is considered a high-risk carrying situation; In the continuous video monitoring, the quantitative indicators of personnel behavior characteristics during the set time window are summarized and analyzed, and matched with the rule base one by one; when a single rule is met, it is initially marked according to the corresponding probability. If multiple rules are triggered at the same time, the abnormal score, outlier value and excess proportion obtained by analysis are marked as yif, yzz and css respectively, and the normalization process is substituted into the following formula: To obtain the disaster judgment value ZPJ, where are the preset weight coefficients of anomaly score, outlier value and excess proportion respectively; the obtained disaster judgment value ZPJ is compared with several preset disaster judgment value intervals, and several disaster judgment value intervals correspond to different fire probability levels. When the disaster judgment value interval to which the disaster judgment value ZPJ belongs is determined, the fire probability level is determined. When the fire probability level is greater than the preset standard, a comprehensive assessment of the fire is performed.

4. The intelligent fire safety integrated big data platform system according to claim 2 is characterized in that: The intelligent analysis module performs a comprehensive fire assessment process as follows: Combine recurrent neural networks including long short-term memory networks to mine time series data associations and comprehensively assess the probability of fire and security risks in each area; Clean the data on the operating parameters of fire-fighting equipment and the frequency of personnel movement in the concerned area; remove erroneous data points; then normalize the data to unify the scale range of data of different magnitudes; The long-term fire protection and security time series data are segmented according to fixed time windows to form data segments. At the same time, the corresponding risk labels are marked for each segment. Model construction, LSTM unit is composed of input gate, forget gate, output gate and memory unit; multiple LSTM units are connected in series in sequence to form a hidden layer. The input layer receives normalized time series data, and the dimension corresponds to the number of features in each time step; the output layer is designed according to the task. If the risk probability is simply predicted, a fully connected layer with a Sigmoid activation function is used to output the risk value; The model hyperparameters, including the number of hidden layers, the number of units in each hidden layer, and the learning rate, were determined after multiple optimizations; Model training: The labeled time series data is divided into training set, validation set, and test set. The training set is sent to the LSTM model for forward propagation to calculate the predicted risk probability, compared with the labeled real label, and the loss value is calculated using the cross entropy loss function; then the back propagation algorithm is used to calculate the gradient of each parameter according to the chain rule, and the model parameters are updated using the optimization algorithm. The training is iterated several times until the loss on the validation set no longer decreases, and the trained optimal model is saved; For the time series of fire equipment operation parameters, the trained LSTM model analyzes the new input data segment by segment and outputs the corresponding risk probability value. When the water pressure continues to drop and approaches the critical low water pressure and the model outputs a high risk probability, it indicates that there are hidden dangers in the operation of fire protection facilities. In terms of the frequency of personnel movement, if the number of personnel in the preset area increases to the threshold during non-working hours and the model warns of high risk, it indicates that there are security issues such as illegal intrusion and illegal operation; The risk probability of fire-fighting equipment operation is integrated with the risk probability of personnel flow and security-related risks, and a weighted sum is taken based on expert experience and the weight ratio mined from historical accident big data to ultimately determine the comprehensive assessment result.

5. The intelligent fire safety integrated big data platform system according to claim 1 is characterized in that: The specific operation steps of the linkage control module are as follows: Preset multiple emergency scenario linkage strategies and integrate with fire protection and security hardware facilities; when the fire alarm is triggered, the security camera is automatically linked to focus on the alarm point and switch to thermal imaging mode to assist in observing the spread of the fire; Unlock the access control of the evacuation passage to ensure the smooth escape route, trigger the fire broadcast to prompt people to evacuate according to the preset evacuation voice, and at the same time push the building fire protection facility distribution map, real-time on-site images, and surrounding rescue force distribution to the fire command center to facilitate scientific command; In daily security management and control, when encountering access control intrusion or perimeter defense alarm, immediately instruct the camera to track and shoot, light warning, and check the status of surrounding fire protection facilities.

6. The intelligent fire safety integrated big data platform system according to claim 5 is characterized in that: The determination of the security camera that automatically links the focused alarm point in the linkage control module is determined by determining the priority of the focused alarm point of the security camera; By formula: Calculate the camera focus priority , where , is the weight coefficient, It is the reciprocal of the distance from the camera to the alarm point; is the ratio of the camera's field of view covering the alarm area; the obtained camera focus priority Compare with the priority threshold, when the camera focuses on the priority When it is greater than a preset priority threshold, it is determined that the camera needs to be focused.

7. The intelligent fire safety integrated big data platform system according to claim 5 is characterized in that: The specific operation steps in the linkage control module for determining whether the security camera needs to switch to the thermal imaging mode are as follows: By analyzing the smoke concentration change rate and the perceived ambient temperature change rate in the collected image data and normalizing them, the smoke concentration change rate and the perceived ambient temperature change rate are used as the two sides of a right triangle, the other side of the triangle is connected, and a triangular pyramid model is established with a preset correction coefficient as the height. The surface area of ​​the triangular pyramid model is calculated and recorded as the thermal judgment value. When the thermal judgment value is greater than the preset thermal judgment threshold, it is determined that the security camera needs to switch to thermal imaging mode.

8. The intelligent fire safety integrated big data platform system according to claim 1 is characterized in that: The specific steps of the data acquisition module are as follows: Integrate various fire protection and security sensors and equipment to build a comprehensive data collection network; In terms of fire protection, smoke sensors, temperature sensors, combustible gas detectors, and electrical fire monitoring devices are installed in various areas of the building, and data is transmitted to the data aggregation unit through a combination of wired and wireless communications; In terms of security, surveillance cameras, access control systems, and electronic patrol equipment are deployed and connected to the data aggregation unit.

9. The intelligent fire safety integrated big data platform system according to claim 1 is characterized in that: The specific steps of the data storage and preprocessing module are as follows: The local edge device is equipped with a server or intelligent gateway with built-in storage media to temporarily store real-time and critical data. The judgment criteria for critical data are based on the level of data cache, and the time freshness of the data is obtained, that is, the time difference from the current time and quantified into a score. The higher the score, the shorter the time difference; data criticality indicators are quantified with values ​​between 0 and 1; data generation frequency, that is, the number of times it is generated per unit time; After normalizing the obtained time freshness, data criticality index and data generation frequency, the optimal judgment value is obtained through weighted summation, and the obtained optimal judgment value is compared with the preset optimal judgment threshold. When the optimal judgment value is greater than the preset optimal judgment threshold, the data is judged to be critical data and needs to be temporarily stored; Use lightweight algorithms for preprocessing; The cloud data center relies on the collaborative architecture of distributed file systems, relational databases and non-relational databases to classify and persist historical data and build a data warehouse for retrieval, query and correlation analysis.

Images