Apartment security linkage method and system based on artificial intelligence

Abstract

The embodiment of the invention relates to the technical field of artificial intelligence, and provides an apartment security linkage method and system based on artificial intelligence, and the method comprises the steps: enabling initial security data obtained in real time to form a security space data set matched with an apartment space according to a space-time association relation; performing conjoint analysis on the security space data set by adopting a security intelligent decision engine to generate a corresponding linkage strategy; sending the corresponding control instruction and the security control parameter to the corresponding target security equipment based on the linkage strategy; the linkage priority of the security and protection equipment is set according to the security and protection event type and the mapping relation between the security and protection equipment and the security and protection event emergency degree, and a control array containing all levels of target security and protection equipment is constructed according to the linkage priority in combination with the current security and protection event type. The control array preferentially triggers the core target security and protection equipment to respond to the indication of the linkage strategy, so that remote control and linkage response of the security and protection equipment are realized, and the intelligent level and the emergency processing capability of apartment security and protection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an apartment security linkage method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid development of Internet of Things technology, apartment intelligent management has become an inevitable trend. However, the current apartment security field is facing many problems to be solved.

[0003] On the one hand, the related art lacks device coordination. The traditional security system is composed of various independent devices simply stacked together, lacking a unified linkage control mechanism. The intelligent camera only bears the function of picture acquisition. Even if it recognizes the abnormal situation of strangers entering, facility damage, etc., it cannot automatically trigger the response action such as closing the access control, sounding the alarm, etc. After the fire alarm is triggered, it is difficult to quickly link the fire-fighting devices such as the sprinkler system, smoke exhaust equipment, etc., and it is also impossible to activate the emergency lighting and evacuation indication system synchronously, which seriously restricts the emergency disposal efficiency.

[0004] On the other hand, the related art has a high degree of artificial dependence. In the personnel behavior monitoring and environmental safety early warning link, excessive reliance on artificial judgment of security personnel. The monitoring room personnel need to closely monitor dozens of monitoring pictures at the same time, and long-time work is easy to cause fatigue, leading to an increase in abnormal behavior omission rate. The identification of environmental hazards such as temperature and humidity abnormalities, gas leaks, etc. is often affected by long artificial inspection intervals and subjective judgment biases, which may cause safety risks.

[0005] Therefore, it is urgent to design a technical solution to solve at least one of the above technical problems. SUMMARY

[0006] The main purpose of the embodiments of the present application is to provide an apartment security linkage method and system based on artificial intelligence, which aims to solve the technical problems of lack of device coordination and high degree of artificial dependence in the related art.

[0007] In a first aspect, the embodiments of the present application provide an apartment security linkage method based on artificial intelligence, comprising: real-time acquisition of initial security data collected by various monitoring devices in the apartment, wherein the initial security data includes video image data, door and window state data, environmental safety data, and personnel access data; forming a security space data set matched with the apartment space according to the spatio-temporal correlation relationship of the real-time acquired initial security data; adopting a security intelligent decision engine to jointly analyze the security space dataset to determine the type of security event and the security risk level existing in the apartment, and generate a corresponding linkage strategy according to the determination result; wherein the security intelligent decision engine comprises a personnel behavior recognition model and an environmental safety risk assessment model; the personnel behavior recognition model distinguishes between normal behavior and abnormal behavior of personnel by real-time analysis of personnel activity characteristics in the security space dataset through a deep learning algorithm; the environmental safety risk assessment model combines the risk assessment threshold dynamically optimized by historical security risk data, quantitatively analyzes the diffusion trend and influence range of personnel abnormal behavior and environmental abnormal phenomena in the current environment based on the security space dataset, to determine the security risk level of the current environment; According to the mapping relationship between the security event type, the security device and the emergency degree of the security event, the linkage priority of the security device is set, and the control array comprising the target security device at all levels is constructed according to the type of security event in which the apartment is currently located according to the linkage priority, and the core target security device is triggered to respond to the instruction of the linkage strategy through the control array, so as to realize the remote control and linkage response of the target security device.

[0008] In a second aspect, the embodiments of the present application provide an apartment security linkage system based on artificial intelligence, comprising: The acquisition module is configured to acquire initial security data collected by various monitoring devices in the apartment in real time, wherein the initial security data comprises video image data, door and window state data, environmental safety data and personnel access data. The construction module is configured to form a security space dataset matched with the apartment space according to the spatio-temporal correlation relationship of the initial security data acquired in real time. The decision module is configured to adopt a security intelligent decision engine to jointly analyze the security space dataset to determine the type of security event and the security risk level existing in the apartment, and generate a corresponding linkage strategy according to the determination result; wherein the security intelligent decision engine comprises a personnel behavior recognition model and an environmental safety risk assessment model; the personnel behavior recognition model distinguishes between normal behavior and abnormal behavior of personnel by real-time analysis of personnel activity characteristics in the security space dataset through a deep learning algorithm; the environmental safety risk assessment model combines the risk assessment threshold dynamically optimized by historical security risk data, quantitatively analyzes the diffusion trend and influence range of personnel abnormal behavior and environmental abnormal phenomena in the current environment based on the security space dataset, to determine the security risk level of the current environment; The execution module is configured to send corresponding control instructions and security control parameters to corresponding target security devices based on the linkage strategy; set linkage priorities of the security devices according to a mapping relationship between security event types, the security devices and emergency degrees of the security events, construct a control array containing security devices at all levels according to the linkage priorities in combination with a current security event type of the apartment, and trigger the core target security device to respond to the indication of the linkage strategy in priority through the control array, thereby realizing remote control and linkage response of the target security devices.

[0009] In a third aspect, the embodiments of the present application further provide a terminal device, which comprises a processor, a memory for storing a computer program, and the processor is configured to execute the computer program and implement the artificial intelligence-based apartment security linkage method of the first aspect or any of the embodiments of the present application when executing the computer program.

[0010] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer software program, and the computer software program is executed by a processor to implement the artificial intelligence-based apartment security linkage method of the first aspect or any of the embodiments of the present application.

[0011] The embodiment of the present application provides an apartment security linkage method and system based on artificial intelligence. In the method, initial security data collected by various monitoring devices in the apartment is acquired in real time, wherein the initial security data comprises video image data, door and window state data, environment safety data and personnel access data. Then, the initial security data acquired in real time is formed into a security space data set matched with the apartment space according to a space-time correlation relationship. Next, a security intelligent decision engine is used to jointly analyze the security space data set, so as to judge the type of security event existing in the apartment and the security risk level, and generate a corresponding linkage strategy according to the judgment result. The security intelligent decision engine comprises a personnel behavior recognition model and an environment safety risk assessment model; the personnel behavior recognition model distinguishes normal behavior from abnormal behavior by analyzing personnel activity characteristics in the security space data set in real time through a deep learning algorithm; the environment safety risk assessment model combines a risk assessment threshold dynamically optimized based on historical security risk data, quantitatively analyzes the diffusion trend and influence range of personnel abnormal behavior and environment abnormal phenomenon in the current environment based on the security space data set, so as to judge the security risk level of the current environment. Finally, corresponding control instructions and security control parameters are sent to corresponding target security devices based on the linkage strategy. The linkage priority of the security device is set according to the mapping relationship among the security event type, the security device and the emergency degree of the security event, a control array comprising target security devices at all levels is constructed according to the linkage priority of the apartment currently in, the core target security device is triggered to respond to the instruction of the linkage strategy through the control array, and remote control and linkage response of the target security device are realized. Through deep integration of artificial intelligence technology and the security system, the embodiment of the present application realizes the upgrade of the apartment security from traditional decentralized monitoring and manual early warning to intelligent collaborative prevention and control, and effectively improves the intelligent level and emergency handling capacity of the apartment security. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 A flowchart of an apartment security linkage method based on artificial intelligence provided by the present application is provided. Figure 2 A structure diagram of an apartment security linkage system based on artificial intelligence provided by the present application is provided. Figure 3 A structure diagram of a terminal device provided by the present application is provided. DETAILED DESCRIPTION

[0013] Embodiments of the present application provide an apartment security linkage method and system based on artificial intelligence. The apartment security linkage method based on artificial intelligence can be applied to a terminal device, which can be a mobile terminal such as a mobile phone, a virtual reality device, a tablet computer, a notebook computer, a desktop computer, a wearable device, etc. The terminal device can be a server connected to a cloud service system, or a server cluster. The above connection mode can be realized by a hardware circuit or a communication module.

[0014] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the case of no conflict, the embodiments described below and the features in the embodiments can be combined with each other. Please refer to Figure 1 , Figure 1 A flowchart of an apartment security linkage method based on artificial intelligence provided by an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the apartment security linkage method based on artificial intelligence includes the following steps: Figure 1 Step S101, real-time acquisition of initial security data collected by various monitoring devices in the apartment.

[0015] In an embodiment of the present application, the initial security data includes video image data, door and window state data, environmental safety data, and personnel access data. Specifically, the video image data is mainly collected by intelligent cameras in public areas of the apartment (such as corridors, entrances, and elevators) and key areas (such as kitchens and balconies), and includes real-time dynamic frame sequences, personnel activity trajectories, object state changes, etc. For example, the corridor camera captures behavior pictures such as personnel loitering at night and climbing over the railing, and the elevator camera records images of personnel carrying dangerous goods. The door and window state data is collected by door and window magnetic sensors, unit door magnetic sensors, and fire passage door magnetic sensors installed on the doors and windows of the residents, and includes door and window opening and closing states (open / closed), opening and closing time stamps, abnormal opening and closing (such as unauthorized forced opening), etc. For example, the resident's door magnetic sensor records that the door was abnormally opened at 2 a.m., and synchronously triggers the state change log. The environmental safety data is collected by environmental sensors deployed in various areas of the apartment, and covers temperature, humidity, smoke concentration, gas concentration, PM2.5 value, etc. For example, the kitchen gas sensor monitors the change of gas concentration in real time, the living room smoke sensor captures smoke particles exceeding the standard, and the bedroom temperature and humidity sensor records the fluctuation of environmental temperature and humidity. The personnel access data comes from the access control system, face recognition equipment, unit door card reader, etc., and includes personnel identity information (such as resident ID, visitor registration information), access time, access location (such as unit door, underground garage entrance), etc. For example, the face recognition equipment at the main entrance records that the resident “Zhang San” entered at 7:30 by swiping the card, and the visitor “Li Si” entered after registration at 9:00.

[0016] ​Exemplarily, in step S101, initial security data is acquired in real time. For example, during the morning rush hour on weekdays, the entrance camera of an apartment building captures real-time footage of people entering and leaving (i.e., video image data), the unit door magnetic sensor records "8:05 unit door opening" (i.e., door and window status data), the kitchen gas sensor detects a gas concentration of 0.02% (i.e., environmental safety data), and the access control system simultaneously records "resident Wang Wu 8:03 card out" (i.e., personnel access data). These data are transmitted in real time to the system background through the Internet of Things gateway, ensuring the timeliness and completeness of data collection. In this way, comprehensive and real-time aggregation of multi-source heterogeneous security data is achieved, breaking the data island problem of traditional security equipment and providing full-quantity raw data support for subsequent analysis, avoiding risk omissions due to data loss.

[0017] Exemplarily, in step S101, a security space data set is formed. The initial data obtained in step S101 is processed by spatio-temporal correlation, and the door and window status data of "8:05 unit door opening" is bound to the physical coordinates of the unit door (e.g., the entrance of unit 1 in building 1, coordinates X: 120.3, Y: 30.5). The "0.02% concentration" data of the kitchen gas sensor at the same time point is associated with the space node of the kitchen (the location of "kitchen 302 in building 1" in the digital twin model). The video image frame sequence, personnel access records within the time period (8:00-8:10) are integrated according to the timestamp, forming a structured data set containing time, space, data type, and value. Through spatio-temporal anchoring and correlation fusion, the dispersed raw data is given a clear spatial context and temporal correlation, solving the problem of disorder and fragmentation of traditional data, providing accurate spatial dimension reference for subsequent risk analysis, and ensuring the knowability, traceability, and usability of data.

[0018] Exemplarily, in step S101, a personnel behavior recognition model is called to analyze the video image data, discovering "8:15 in the underground garage camera footage, a stranger tries to pick the lock without a access card" (abnormal behavior). The environmental safety risk assessment model combines historical data (no lock-picking record in the past three months in this area, and the current high humidity is prone to equipment failure), dynamically optimizes the risk threshold, and judges that "the lock-picking behavior is associated with theft risk, with a high level". At the same time, the personnel access data shows that the stranger is not registered in the access control system, further verifying the authenticity of the risk. Finally, a linkage strategy is generated: "immediately activate the sound and light alarm in the underground garage, lock the access control in the area, and push the warning information to the property security terminal simultaneously". Through deep learning and dynamic threshold optimization, accurate identification of personnel behavior and environmental anomalies is achieved, avoiding the subjectivity and lag of traditional manual judgment. Multi-model joint analysis makes risk judgment more comprehensive, reduces the probability of "single data misjudgment", and improves the accuracy of event type and risk level judgment.

[0019] Exemplarily, in step S101, based on the generated linkage strategy, the system constructs a control array according to priority: the core device (underground garage sound-light alarm, access control lock) is the first priority, and the alarm is started (volume 80 decibels) and the access control lock is dead. The auxiliary device (nearby area camera) is the second priority, and the camera is started to record in high definition and track the trajectory. The supporting device (property communication terminal) is the third priority, and the warning information containing the location coordinates is pushed. If the alarm fails to respond, the system automatically switches to the standby alarm in the adjacent area and upgrades the strategy (expands the alarm range to the entire underground garage). Through the priority linkage mechanism, the core device is ensured to respond first and dispose quickly, solving the problem of unordered start and low efficiency of traditional security devices. The functions of automatic switching of device failure and upgrading of strategy ensure the continuity and reliability of emergency response and reduce the probability of risk diffusion.

[0020] Through sha, a closed-loop management from data collection, analysis and decision, device linkage, and optimization iteration is realized. Real-time data collection ensures risk monitoring without delay, spatio-temporal correlation data set provides basis for analysis, intelligent decision engine improves judgment accuracy, and dynamic device linkage improves response efficiency. Compared with traditional security, it effectively solves the problems of data dispersion, judgment lag, and unordered linkage, significantly improves the initiative, accuracy, and emergency response capability of apartment security, and provides more reliable security protection for residents.

[0021] In step S102, the initial security data acquired in real time is formed into a security space data set matched with the space of the apartment according to the spatio-temporal correlation relationship.

[0022] In the embodiment of the application, the security space data set is a structured data set formed by integrating and correlating the initial security data collected in the apartment in the time and space dimensions. The core is to assign clear spatial location attributes and time correlation information to various initial data, so that the data can be accurately corresponded to specific physical space areas (such as rooms, corridors, entrances, etc.) and time nodes of the apartment, thereby constructing a dynamic data panorama matched with the actual space layout of the apartment.

[0023] Specifically, the security space dataset is assigned a dual association identifier during integration. Spatially, each data point is bound to corresponding physical space information, for example, video image data is marked with the location of the camera that collected it (such as "3 Building 2 Unit 501 Living Room Camera", "Unit Entrance Main Camera"), door and window state data is associated with the specific space location of the door and window (such as "501 Household Entrance Door", "501 Household Balcony East Window"), environmental safety data corresponds to the sensor deployment area (such as "501 Household Kitchen Smoke Sensor", "Living Room Temperature and Humidity Sensor"), and personnel access data is bound to the entrance location (such as "2 Unit Unit Door", "5th Floor Elevator Entrance"). Temporally, all data is uniformly labeled with a timestamp or time interval, ensuring that different types of data can be aligned in the time dimension (such as "15:30:20 Living Room Camera Data" and "15:30:22 Entrance Door State Data" can be associated in time sequence). Through this integration, the security space dataset is no longer isolated video clips, switch states or numerical fluctuations, but a collection of associated data that can intuitively reflect the personnel activity, device status and environmental changes in a certain area at a certain time.

[0024] For example, assume that multiple types of initial data are generated simultaneously in an apartment at a certain time. The 501 household entrance camera captures a stranger lingering at 18:05:10, the household entrance door sensor detects that the door is not closed at 18:05:12, the living room temperature and humidity sensor returns normal values at 18:05:15, and the kitchen smoke sensor shows a slight increase in concentration at 18:05:18. Step S102 integrates these data into a security space dataset. The video data is labeled with the "501 Household Entrance Area" spatial tag and the "18:05:10" timestamp, the door state data is labeled with the "501 Household Entrance Door" spatial tag and the "18:05:12" timestamp, the temperature and humidity data is labeled with the "501 Household Living Room Area" tag and the corresponding time, and the smoke data is labeled with the "501 Household Kitchen Area" tag and the time, finally forming a dataset that includes the real-time status of each area and is time-sequential.

[0025] Thus, the above steps break the dispersion and isolation of initial security data, and through spatio-temporal association, the data is given a spatial context and temporal logic, making the originally independent device data form a data portrait corresponding to the apartment physical space. This not only allows the subsequent security intelligent decision engine to locate anomalies based on specific areas (such as determining whether the abnormal behavior occurred in the entrance area or the kitchen area), but also enables the analysis of the cause-and-effect relationship of events through time association (such as whether there is a time sequence correlation between the person lingering and the door being unlocked), avoiding misjudgment or omission due to data isolation, and providing a structured and scenario-based data foundation for subsequent accurate identification of security event types and risk assessment.

[0026] In an optional embodiment of step S102, step 201, the initial security data acquired in real time is preprocessed by an outlier detection algorithm to eliminate invalid data, noise data and repeated collection data, to obtain valid security data. Step 202, based on the unified data format standard, the heterogeneous data collected by different types of monitoring devices is converted and standardized, wherein the video image data is converted into frame sequence structured data, the door and window state data, the environmental safety data and the personnel access data are converted into key-value pair structured data, to ensure the format compatibility of multi-source data. Step 203, the physical coordinate information of each monitoring device is extracted and a digital twin mapping model of the apartment space is established, and the valid security data after preprocessing is spatio-temporally anchored according to the collection time stamp and the corresponding physical coordinate information, so that each initial security data is associated to the corresponding position node in the digital twin mapping model of the apartment space. Step 204, by a spatio-temporal association algorithm, the multi-source initial security data of the same time segment and adjacent space region are associated and fused to form a multi-dimensional index structure containing data collection time, spatial coordinates, data type and data value, and a security space data set consistent with the actual physical space of the apartment is constructed, and the dynamic security space data set is dynamically updated with the spatio-temporal index relationship of the initial security data collected in real time.

[0027] Specifically, step 201 filters the initial security data by an outlier detection algorithm, such as a statistical-based Z-score method, a quartile range method, or a machine learning-based isolation forest model, to identify and eliminate invalid data (such as blank values when the sensor is offline), noise data (such as blurred frames generated by the camera due to sudden light changes), and repeated data (such as the same state data repeatedly uploaded by the same device within a short period of time), and retains valid data reflecting the true state. For example, assume that the kitchen smoke sensor has poor line contact and continuously returns an abnormal value of "concentration = 999 mg / m³" (far exceeding the normal range) within 18:05:00-18:05:05, and the living room camera produces 3 completely blurred image frames due to lens reflection, step 201 will identify these data as noise or invalid data and eliminate them, and only retain the "concentration = 0.1 mg / m³" returned by the sensor after it resumes normal operation and the clear image frames of the camera within this period. Thus, data noise and interference are filtered from the source, ensuring that the data entering the subsequent processing is authentic and reliable, avoiding false positives (such as mistaking sensor fault values as fire risks) in subsequent analysis caused by invalid data, and providing a high-quality basis for data integration.

[0028] Step 202 converts the heterogeneous data (video stream, switch value, numerical value, etc.) output by different monitoring devices into structured data based on a unified data format standard, converting unstructured / semi-structured data into structured data. Video image data is converted into frame sequence structured data containing timestamps and image features through frame extraction technology. Door and window status, environmental safety, personnel access, and other data are converted into key-value pair structured data containing "device ID, timestamp, spatial label, data value" through field definition, solving the problem of incompatible data formats from multiple sources. For example, the H.264 format video stream output by the unit entrance camera is converted into structured data with one frame every 100 ms, with each frame labeled with "device ID: unit door camera 01, timestamp: 18:05:10.000, image features: [person outline coordinates, motion direction]". The on-off signal (originally high-low level) of the 501 household entrance sensor is converted into the key-value pair "device ID: entrance sensor 01, timestamp: 18:05:12, spatial label: 501 household entrance, data value: unlocked". Thus, the format barriers of different device data are eliminated, enabling video, sensor, status, and other multi-type data to be recognized and processed under a unified framework, providing a consistent data basis for subsequent spatio-temporal correlation and improving data integration efficiency.

[0029] Step 203 constructs a spatial digital twin mapping model (containing digital images of room, corridor, device, and other spatial nodes) consistent with the actual layout of the apartment by collecting the physical coordinate information of each monitoring device (such as the room number where the camera is installed, the specific location where the sensor is deployed). Then, the preprocessed effective data is anchored to the corresponding spatial nodes of the model through the collection of timestamps (time dimension) and corresponding physical coordinates (space dimension), realizing the precise correlation between data and physical space. For example, the "3 buildings, 2 units, 501 households" node is included in the digital twin model of the apartment, which is further divided into "living room, kitchen, and entrance door" sub-nodes; the physical coordinates of the 501 household living room temperature and humidity sensor are associated with the "living room" sub-node, and the "temperature 25°C, humidity 40%" data returned by the sensor at 18:05:15 is anchored to the "living room" sub-node of the model through the timestamp "18:05:15" and spatial coordinates "501 household living room", forming a binding relationship between space, time, and data. In this way, an intuitive mapping between abstract data and the physical space of the apartment is established, enabling each piece of data to be clearly associated with a specific area and time point, solving the problem of "no spatial attribution" of data and providing precise spatial context support for subsequent analysis of "security status in a certain area at a certain time".

[0030] Step 204, based on a spatio-temporal correlation algorithm (such as time window clustering, spatial neighborhood analysis), correlates and fuses multi-source effective data in the same time segment (such as within 10 seconds) and adjacent spatial regions (such as the adjacent region of the entrance door and the living room) to construct a multi-dimensional index structure containing "data collection time, spatial coordinates, data type, data value". At the same time, with the real-time collection of new data, the index relationship is dynamically updated, so that the data set always reflects the latest security status of the apartment. Exemplarily, within the time window of 18:05:10-18:05:20, the video data of the "501 household entrance area" (strange person loitering), the state data of the "entrance door" (not closed), and the smoke data of the "kitchen" (slight concentration rise) are associated and fused into a multi-dimensional index record because of the time continuity and spatial adjacency (the entrance area and the kitchen belong to the 501 household) : "time interval: 18:05:10-18:05:20, spatial coordinates: 501 household entrance and kitchen area, data type: video + door status + smoke, data value: [person loitering frame sequence, not closed, 0.2 mg / m³]". When the new data of the living room light anomaly is transmitted at 18:05:25, the data set updates the index in real time and includes the new data at this time point. Thus, through spatio-temporal correlation, scattered data is converted into a structured collection with logical correlation, so that the data not only contains the status of a single device, but also reflects the "multi-dimensional security scene in a certain area at a certain time period"; the dynamic updating mechanism ensures that the data set is always synchronized with the real-time state, providing a time-sequential, spatially explicit, and multi-source fused high-quality analysis basis for the subsequent security intelligent decision engine, greatly improving the accuracy of event recognition and risk assessment.

[0031] Further optionally, spatio-temporal anchoring is the process of precisely binding the preprocessed effective security data with timestamps and physical spatial coordinates to the corresponding position nodes in the apartment space digital twin mapping model, and the core is to realize the three-dimensional correlation of data, time, and space.

[0032] First, a coordinate system for apartment space is established to build a unified coordinate coding rule for the physical space of the apartment, ensuring that the installation location of each monitoring device can be uniquely identified. For example, the absolute coordinate method assigns a three-dimensional coordinate to each area in the apartment based on building surveying data, with X / Y corresponding to the horizontal position and Z corresponding to the floor height. The hierarchical coding method encodes the space according to the levels of building, unit, floor, room, and area, with each code corresponding to a unique space range. There is also a device coordinate mapping table that pre-enters the coordinate information of the installation location for each monitoring device, forming a mapping relationship library between device ID and space coordinates, ensuring that the data collected by the device can be traced back to a specific location. Second, time stamp standardization processing is performed to unify the time recording format of different devices to eliminate time confusion caused by clock bias and time zone differences. This includes time format unification, which converts the collection time of all data to a standardized format accurate to the millisecond level. Time synchronization calibration is performed through the Network Time Protocol to regularly calibrate the clocks of all monitoring devices, ensuring that the time deviation between devices does not exceed 100 milliseconds. At the same time, the time error caused by transmission delay is corrected by adding a transmission delay compensation value to the local collection time of the device. Finally, the mapping and binding of data and digital twin nodes are performed based on the apartment space digital twin mapping model to bind effective security data to the corresponding nodes. The digital twin model is a digital mirror image of the physical space of the apartment, including building structure, device location, and area division nodes. When matching the space, the corresponding space node is located in the digital twin model according to the coordinate information of the device, and the data collected by the device is associated with the node. The time anchor adds a standardized time stamp to each piece of data to ensure that the data can be sorted and associated in the time dimension. Dynamic association update automatically updates the mapping table of device ID and space coordinates when the device moves or changes location, and synchronously adjusts the association relationship between data and nodes in the digital twin model.

[0033] Taking the security data processing of 501 households in 3 buildings and 2 units of an apartment as an example, the apartment adopts a hybrid system of absolute coordinates and hierarchical coding, and the absolute coordinates of the kitchen of 501 households are X = 10.5, Y = 8.2, Z = 9.0, and the hierarchical coding is B3-U2-F5-R1-A2. The coding node in the digital twin model corresponds to the kitchen physical space, including cabinet position, smoke sensor, gas sensor and other digital mirror images. Further, the smoke sensor in the kitchen pre-records coordinate information, synchronizes time through the network time protocol, and assumes that the local collection time is 2023-10-01 18:05:15.300. Assuming that the standardized timestamp after transmission delay compensation is 2023-10-01T18:05:15.350+08:00. The smoke concentration 0.2mg / m³ effective data collected by the sensor, the system locates the corresponding digital twin node according to the device ID query mapping table, and finally binds the structured data of space node B3-U2-F5-R1-A2, timestamp 2023-10-01T18:05:15.350+08:00, data type environmental safety data, data value smoke concentration = 0.2mg / m³.

[0034] The spatio-temporal anchoring in the embodiment of the application solves the problems of no spatial attribution and time confusion of traditional security data by unique identification of spatial coordinates, precise synchronization of timestamps and binding of digital twin nodes, so that each data can be clearly corresponded to a specific location and time in the apartment. This provides a precise spatio-temporal reference for subsequent spatio-temporal correlation analysis, ensuring that the security intelligent decision engine can judge risks based on scenario data, greatly improving the accuracy of event recognition and the precision of spatial positioning.

[0035] Step S103, adopting a security intelligent decision engine, jointly analyzing the security space data set to judge the type of security events and the level of security risks existing in the apartment, and generating a corresponding linkage strategy according to the judgment result.

[0036] In the embodiments of the present application, the linkage strategy is a coordinated response scheme formed by integrating multiple security resources and systems based on the judgment results of the security intelligence decision engine on the security event types and security risk levels in the apartment. The core of the linkage strategy is to quickly contain risk spread or solve security problems through dynamic and accurate resource scheduling and operation execution. In actual application, the linkage strategy will be differentiated according to the nature of the event and the risk level. If abnormal behavior of personnel (such as illegal intrusion, fall without movement) is identified, the linkage strategy will first trigger local early warning, such as the linkage of the on-site sound and light alarm device to remind the surrounding personnel to pay attention, and at the same time, the real-time picture and location information are pushed to the property security center, so as to facilitate the security personnel to quickly locate and dispose. If the risk level is high (such as the intruder carrying dangerous goods), the access to the related area will be blocked by the linkage of the access control system to prevent the personnel from escaping or the risk from spreading, and the preset emergency contact person's phone will be automatically dialed or the alarm information will be synchronized to the police station, so as to realize the connection of internal and external response forces.

[0037] The linkage strategy for environmental abnormal phenomena focuses more on risk control and environmental regulation. For example, when the gas concentration is detected to be higher than the optimal evaluation threshold and the risk level is medium or high, the strategy will immediately link the gas valve to perform the closing operation, and at the same time, start the ventilation system to enhance air circulation and reduce the risk of gas accumulation. If accompanied by abnormal temperature rise (such as suspected initial fire), the fire sprinkler system will be linked to be on standby, and the real-time data will be synchronized to the fire control center to provide accurate environmental parameter support for professional rescue. In addition, for low-level environmental risks (such as slightly high humidity), the linkage strategy may only push a prompt message to the residents through the APP, suggesting manual adjustment to avoid excessive intervention.

[0038] The key of this kind of strategy lies in the integrity and timeliness of "linkage". By connecting the data channels between the monitoring system, sensing devices, execution devices, management platform and external rescue resources, the seamless connection of the whole process from risk identification to response execution is ensured, and at the same time, the response intensity will be dynamically adjusted according to the real-time risk assessment results, which can not only avoid resource waste, but also ensure that the risk is effectively controlled.

[0039] In the embodiments of the present application, the security and protection intelligent decision engine includes a personnel behavior recognition model and an environment safety risk assessment model. The security and protection intelligent decision engine is the core analysis and decision center of the entire apartment security system, responsible for integrating multi-source data, calling professional models for joint analysis, and generating accurate linkage strategies. Its core function is to receive structured security space data sets, through the collaborative calculation of the personnel behavior recognition model and the environment safety risk assessment model, to realize the full-process intelligent processing of type judgment, risk grading, and strategy generation of security events. Exemplarily, in the workflow of the security and protection intelligent decision engine, first, the multi-dimensional data such as video images, environmental parameters, and personnel trajectories in the security space data set are feature extracted, and the personnel activity features and environmental abnormality features are input into the corresponding models. Subsequently, through cross-model data interaction, such as comprehensive judgment through spatio-temporal correlation analysis of the personnel abnormal behavior and environmental risk output by each model, to determine the type of security event (such as "human-induced gas leakage" and "natural occurrence of circuit short circuit") and the corresponding safety risk level. Finally, according to the security event type, safety risk level, and device linkage rule, a linkage strategy containing target devices, response timing, and control parameters is generated. For example, when the security and protection intelligent decision engine identifies a coupled event of "kitchen gas leakage" and "personnel violation of operating the stove", it will preferentially trigger the gas valve closing instruction, simultaneously start the exhaust equipment and regional alarm, and form a progressive disposal scheme. Thus, the security and protection intelligent decision engine breaks through the analysis limitations of a single data dimension, realizes accurate judgment of complex security scenes through multi-model collaboration, and greatly shortens the decision response time.

[0040] Further optionally, the personnel behavior recognition model analyzes the personnel activity features in the security space data set in real time through a deep learning algorithm to distinguish between normal behavior and abnormal behavior. The personnel behavior recognition model is the behavior perception core of the security and protection intelligent decision engine, which analyzes the personnel activity features in the security space data set in real time based on a deep learning algorithm, and accurately distinguishes between normal behavior and abnormal behavior. Exemplarily, the algorithm principle of the personnel behavior recognition model includes three major links of target detection, behavior feature extraction, and classification reasoning. Specifically, the personnel target is located from the video image data through a target detection algorithm based on a convolutional neural network (CNN) (such as YOLO, Faster R-CNN), and key features such as human posture, motion trajectory, and action sequence are extracted; then the time series features are analyzed through a recurrent neural network (RNN) or a Transformer model, and the behavior nature is judged by a classifier trained based on historical behavior samples.

[0041] In practical applications, the normal behaviors that the personnel behavior recognition model can recognize include daily entry and exit of residents, short-term stay in the corridor, and reasonable use of public facilities. Abnormal behaviors include long-term loitering of strangers (more than 15 minutes), climbing over the railing, damaging equipment, carrying dangerous goods, etc. For example, when the corridor camera captures a person repeatedly pacing and touching the access control equipment at the entrance of the fire access door, the model will analyze the trajectory features (fixed area reciprocating motion) and action features (abnormal contact with equipment) to determine that it is an “abnormal stay + potential damage behavior”, and output the abnormal label and confidence in real time. The personnel behavior recognition model improves the behavior recognition accuracy, reduces the false negative rate of abnormal behavior, and can complete single-frame behavior analysis within a preset time, achieving early detection and early warning of risk behaviors.

[0042] Further optionally, the environmental safety risk assessment model dynamically optimizes the risk assessment threshold based on historical security risk data, quantitatively analyzes the diffusion trend and influence range of personnel abnormal behavior and environmental abnormal phenomena in the current environment, and judges the safety risk level of the current environment. Specifically, the environmental safety risk assessment model focuses on the monitoring, diffusion analysis and risk grading of environmental abnormal phenomena in the apartment, and realizes precise control of fire, gas leakage, temperature and humidity abnormalities and other environmental risks through dynamically optimized evaluation thresholds and quantitative analysis methods. The algorithm principle of the environmental safety risk assessment model includes three modules: threshold dynamic optimization, diffusion trend prediction and multi-factor risk assessment. Based on the historical security risk feature library, the particle swarm optimization algorithm is used to continuously optimize the warning thresholds of parameters such as temperature and gas concentration. Through the graph neural network (GNN) combined with the apartment space digital twin model, the diffusion speed and influence range of abnormal substances such as smoke and gas are simulated. Then, the fuzzy comprehensive evaluation method is used to quantify multiple factors such as abnormal intensity, personnel density, and ventilation conditions into risk level indexes.

[0043] For example, when the kitchen gas sensor detects an abnormal increase in concentration, the environmental safety risk assessment model first calls the threshold combination of the same scene in the historical data (such as dynamically adjusting the kitchen gas safety threshold to 0.03%), combines the current ventilation state (the diffusion coefficient is increased by 20% when the window is closed) and the personnel state (the risk weight is increased by 30% when there are people in the kitchen), and predicts that the gas may spread to the living room area within 5 minutes through GNN. Then, the risk level index is calculated comprehensively, and it is determined as high risk, and the linkage strategy of “immediately close the gas valve + start kitchen exhaust + push the resident evacuation reminder” is triggered. The environmental safety risk assessment model can reduce the risk level judgment deviation, improve the diffusion range prediction accuracy, and adapt to environmental changes such as seasons and house types through a closed-loop threshold optimization mechanism, greatly reducing the environmental risk false negative rate.

[0044] Further optionally, in the above-mentioned environmental safety risk assessment model, the step of dynamically optimizing the risk assessment threshold in combination with historical security risk data can also classify and label historical security risk data by event type, extract environmental characteristic parameters and risk evolution law data corresponding to each type of apartment security event, and construct a classified historical risk feature library; wherein the environmental characteristic parameters corresponding to each type of apartment security event at least include: temperature, humidity, gas concentration baseline value, and personnel density. Further, based on the classified historical risk feature library, real-time environmental characteristic parameters are matched with environmental characteristic parameters of historical similar apartment security events, and a particle swarm optimization algorithm is used to minimize the risk misjudgment rate and the missing rate as the objective function, to automatically search for the optimal risk assessment threshold combination matching the current apartment space area and real-time environmental state; the optimal risk assessment threshold combination at least includes: warning threshold of different environmental characteristic parameters, and risk level division threshold. Then, a time sliding window is set, and the window length is dynamically adjusted according to the historical occurrence frequency of apartment security events, so as to shorten the window length of high-frequency apartment security events to 1 hour and extend the window length of low-frequency apartment security events to 24 hours; the environmental monitoring data and event disposal results in the security space data set are real-time included through the time sliding window, and the optimal risk assessment threshold combination is dynamically calibrated. Finally, the calibrated optimal risk assessment threshold combination is input into the environmental safety risk assessment model, and the identification accuracy of the calibrated optimal risk assessment threshold combination for the current environmental abnormal phenomenon is verified in real time; if the accuracy rate of the identification accuracy is lower than the preset threshold for three times or more than three times, a new round of particle swarm optimization search is triggered, forming a closed-loop threshold optimization mechanism to ensure that the risk assessment threshold is adapted to the dynamically changing environmental characteristic parameters of the apartment.

[0045] In the environmental safety risk assessment model, the process of dynamically optimizing the risk assessment threshold begins with the systematic analysis of historical security risk data. First, the historical data is classified and labeled in detail according to the type of event, such as distinguishing between different types of security events such as fire, gas leakage, and temperature and humidity anomalies. On this basis, environmental characteristic parameters corresponding to each type of event are extracted from the historical records, including temperature, humidity, gas concentration baseline value, and personnel density, which are closely related to safety risks. The environmental safety risk assessment model also summarizes the risk evolution law of different events, such as how a certain abnormal phenomenon spreads over time and how the scope of influence changes. Through such processing, a classified historical risk feature library is constructed, providing rich historical reference for subsequent real-time analysis and threshold optimization.

[0046] Further optionally, the core of the environmental safety risk assessment model in summarizing the evolution law of risk lies in mining the internal logic of abnormal phenomena changing over time and space from historical security data, and converting these laws into quantifiable and predictable model parameters through technical means. First, the environmental safety risk assessment model will conduct in-depth analysis on the data in the classified historical risk feature library, and extract the time sequence variation characteristics of key environmental parameters for different event types such as fire, gas leakage, and water leakage. For example, the rising curve of temperature over time after a fire, the order of appearance of smoke concentration in different areas, or the accumulation speed of concentration in a closed space after a gas leak. At the same time, combined with the space-anchored security space data set, the relationship between the initial position, propagation path of abnormal phenomena and the physical structure of the apartment is associated, such as the guiding effect of the direction of ventilation ducts on smoke diffusion, and the delay effect of wall barriers on gas propagation, thereby establishing an associated feature library of “event type-environmental parameter-space structure” to provide a data basis for law summary.

[0047] In the diffusion process modeling layer, the environmental safety risk assessment model will construct a dynamic diffusion simulation framework based on the apartment space digital twin mapping model. By analyzing the spatial propagation trajectory of abnormal phenomena in historical data, the model will identify the main factors affecting diffusion in different scenarios: for example, in the case of kitchen gas leakage, the space tightness and ventilation equipment status are the key factors affecting the diffusion range. While fire smoke diffusion is more dependent on the connectivity of vertical channels such as floor height and stairwell. The environmental safety risk assessment model will convert these factors into computable topological parameters, combine with the basic laws of physical diffusion (such as concentration gradient, air flow direction), and construct a quantified model containing time decay coefficient and space propagation weight, to simulate the whole cycle evolution process of abnormal phenomena from occurrence to stability, including the core laws of diffusion speed, boundary change of affected area in different stages.

[0048] To ensure that the summarized laws can adapt to the dynamically changing actual environment, the environmental safety risk assessment model will also introduce real-time feedback and parameter self-optimization mechanism. By continuously incorporating new security event data, it will compare the deviation between the actual diffusion process and the historical law, and dynamically adjust the law parameters. For example, when the doors and windows are closed frequently in winter, the diffusion speed law of gas leakage will be corrected accordingly. After the installation of new exhaust equipment, the dissipation law of kitchen smoke will update the weight coefficient. This dynamic adaptation capability makes the evolution law summarized by the model not only contain historical experience, but also respond to changes in the apartment environment in real time. Such as equipment upgrade, spatial structure adjustment brought by house type transformation, so as to ensure that the summarized law is always consistent with the actual scene, providing a reliable logical basis for subsequent diffusion trend prediction and impact range assessment. Based on the completed classification historical risk feature library, the environmental safety risk assessment model will compare and match the real-time collected environmental feature parameters with the environmental feature parameters of the same type of security events in history. In this process, with the help of particle swarm optimization algorithm, the core goal is to reduce risk misjudgment and omission, and automatically find the optimal risk assessment threshold combination that matches the current apartment space area and real-time environmental state. These optimal threshold combinations not only include warning threshold values for different environmental feature parameters, i.e. critical values to determine when to issue a warning, but also include thresholds for dividing risk levels to ensure that the thresholds can accurately reflect the actual risk situation of the current environment, rather than relying on fixed and unchanged standards.

[0049] Further, in order to make the threshold always fit the dynamic changes of the apartment environment, the environmental safety risk assessment model will set a time sliding window to continuously calibrate the optimal threshold combination. The length of the window is not fixed, but is flexibly adjusted according to the occurrence frequency of various security events in the history of the apartment. For events with high occurrence frequency, the window length will be shortened accordingly to incorporate the latest data more timely. For events with low occurrence frequency, the window length will be appropriately extended to ensure sufficient sample support for calibration. Through this dynamically adjusted time window, the environmental safety risk assessment model will continuously incorporate real-time environmental monitoring data and event handling results in the security space dataset, continuously optimize and calibrate the determined optimal risk assessment threshold combination, so that the threshold can keep up with the subtle changes of the environment.

[0050] The calibrated optimal risk assessment threshold combination will be input into the environmental safety risk assessment model for real-time verification of its recognition accuracy for current environmental abnormal phenomena. The environmental safety risk assessment model will continuously monitor the performance of the threshold in actual application. If the accuracy rate of recognition accuracy is always lower than the preset standard in continuous multiple verifications, a new round of particle swarm optimization search will be automatically triggered to find a more suitable threshold combination. This closed-loop threshold optimization mechanism, through continuous verification, feedback and adjustment, ensures that the risk assessment threshold can always adapt to the dynamic changes of the environmental feature parameters of the apartment, thereby providing a reliable basis for accurate judgment of risk levels. In the embodiments of the present application, the optimal risk assessment threshold combination is a set of threshold values that are adapted to the current environmental state after dynamic optimization in the environmental safety risk assessment for various environmental characteristics and safety risk levels of the apartment. Its core role is to accurately define the normal and abnormal limits of environmental parameters and the division standards of different risk levels, thereby reducing false positives or false negatives of safety events and making risk assessment more consistent with the dynamic changes of actual scenarios. This combination includes two types of key thresholds: one is the warning threshold of environmental characteristic parameters, which corresponds to various core environmental monitoring indicators in the apartment, such as the critical value of temperature (exceeding which may indicate a potential fire), the safety upper limit of gas concentration (exceeding which triggers a warning), the abnormal fluctuation limit of humidity, and the crowd density warning standard, etc. Each environmental parameter has a matching warning threshold for quickly identifying the abnormal state of a single environmental parameter. The other is the risk level division threshold, which is used to distinguish the severity of different safety risks, such as when environmental abnormal parameters are superimposed with personnel activity characteristics, the risk is divided into different levels of low, medium, and high by setting limits, providing clear level basis for subsequent response strategies.

[0051] The verification method of the optimal risk assessment threshold combination mainly focuses on dynamic adaptability and recognition accuracy. First, dynamic calibration verification is performed by continuously incorporating the latest environmental monitoring data and actual event handling results through time sliding window to constantly adjust the threshold combination to ensure its adaptability to the real-time changes of the apartment environment, such as shortening the window length in high-frequency event scenarios to respond quickly, and extending the window in low-frequency scenarios to accumulate enough data to support calibration. Second, real-time accuracy verification is performed by applying the calibrated threshold combination to the identification of current environmental abnormal phenomena to monitor its judgment accuracy in real time to ensure the ability to capture real risks. Finally, closed-loop feedback verification is performed, if the identification accuracy is lower than the preset standard for several times in a row, a new round of threshold optimization process is automatically triggered to search for a more optimal threshold combination through algorithm, forming a closed-loop mechanism of evaluation, verification, and optimization to ensure that the threshold combination is always adapted to the dynamic changes of the environmental characteristics of the apartment.

[0052] Exemplarily, the application of the optimal risk assessment threshold combination in different actual scenarios is to dynamically adapt the threshold parameters to achieve accurate risk identification and warning by combining the scene risk characteristics, environmental parameter characteristics, and actual safety needs, avoiding false positives or false negatives. The optimal risk assessment threshold combination needs to focus on the core risk types of the scene, the fluctuation rules of environmental parameters, and the patterns of personnel activities, which can be embodied in the following example scenarios: First, in high-risk scenarios such as the kitchen, the core risks are concentrated on gas leakage, smoke diffusion, and temperature abnormalities (such as fire hazards). At this time, the optimal threshold combination will focus on adapting to three types of environmental characteristic parameters: gas concentration, smoke concentration, and temperature. Considering that daily cooking in the kitchen will produce temporary fluctuations in gas or smoke, the threshold will be dynamically adjusted in combination with historical data during cooking peak hours (such as early and late hours): during non-cooking hours, the warning threshold for gas concentration is set more strictly, and slight leakage can trigger a warning. During cooking hours, the tolerance for short-term fluctuations is appropriately relaxed to avoid frequent false alarms, but the sensitivity to sustained over-standard conditions (such as a gas valve that is not tightly closed) is maintained. At the same time, the temperature threshold will be associated with the smoke parameter to form a linked judgment: if the temperature rises sharply accompanied by an over-standard smoke concentration, the risk level division threshold will automatically increase to accelerate the triggering of emergency response.

[0053] Second, for living scenarios such as the living room and bedroom where people are densely active, risks are more related to abnormal behavior of people (such as falling, staying) and environmental comfort abnormalities (such as sudden temperature changes, humidity imbalances). The optimal threshold combination will focus on the threshold adaptation of personnel density, activity frequency, and temperature and humidity parameters. For example, in the bedroom during the night, the normal threshold of personnel activity frequency will be significantly reduced, and if abnormally high-frequency activities (such as multiple get-ups) are detected, combined with sudden temperature drop data, it may trigger a risk warning of "elderly discomfort at night"; in the living room during holidays when people gather, the safe threshold of personnel density will be relaxed according to the space carrying capacity, but if the density continuously exceeds the maximum space carrying threshold and is accompanied by temperature rise, it will trigger a risk prompt of "insufficient ventilation" to avoid environmental risks caused by crowded people.

[0054] Third, in public passage scenarios such as corridors and elevators, the core risks are people congestion, abnormal stay, and equipment operation abnormalities (such as elevator failure). The optimal threshold combination will focus on personnel density, stay duration, and equipment operation parameters (such as elevator temperature, operating noise). As an evacuation passage, the safe threshold of personnel density in the corridor will strictly match the passage width, and if the short-time density exceeds the standard, combined with the stay duration threshold (such as more than 10 minutes without evacuation), it will trigger a congestion risk warning. In the elevator, the temperature threshold will be bound with the running state, and if the temperature rises sharply during operation and exceeds the equipment tolerance threshold, even if the personnel discomfort standard is not met, it will also trigger an equipment abnormality warning in linkage, to troubleshoot the fault in advance.

[0055] Fourthly, the risks of closed scenes such as underground garages mainly come from tail gas accumulation (such as carbon monoxide concentration), equipment aging (such as circuit temperature) and humid environment (such as humidity corrosion of metal components). The optimal threshold combination can set the carbon monoxide concentration warning threshold, the circuit temperature safety threshold and the humidity upper threshold. Due to the low ventilation efficiency of the garage, the warning threshold of carbon monoxide concentration will be lower than that of open space, and will be dynamically adjusted according to the running state of the ventilation system. If the ventilation equipment fails, the threshold will be automatically lowered to trigger the warning in advance. At the same time, the humidity threshold will be associated with the risk of metal component corrosion. If the humidity is out of standard for a long time, even if it does not directly affect the personnel, it will also increase the threshold through risk level division, prompting the maintenance requirement.

[0056] Through this scenario-based adaptive logic, the optimal risk assessment threshold combination can dynamically adjust the warning thresholds and risk level division standards of various parameters according to the core risk points, environmental parameter fluctuation rules and actual safety requirements of different scenarios, ensuring accurate and efficient risk identification in various scenarios, and providing reliable basis for the generation of linkage strategy.

[0057] Further optionally, in the step of dynamically optimizing the risk assessment threshold, based on the classified historical risk feature library, the real-time environmental feature parameters are matched with the environmental feature parameters of historical security events of the same type of apartment, and a particle swarm optimization algorithm is used to minimize the risk misjudgment rate and the missed judgment rate as the objective function, to automatically search for the optimal risk assessment threshold combination matching the current apartment space area and real-time environmental state, including: The risk assessment threshold combination including temperature warning threshold, smoke concentration threshold, gas leakage threshold and personnel density threshold is encoded as particle position vector, and the vector dimension corresponds to the type of environmental feature parameter one by one; based on the threshold distribution range of historical security events of the same type of apartment in the classified historical risk feature library, the particle swarm position and velocity are randomly initialized, and the number of particles is dynamically set according to the dimension of environmental feature parameters to ensure that the search space covers the historical optimal threshold interval; The fitness function is constructed with risk misjudgment rate, risk missed judgment rate, threshold sensitivity deviation and their corresponding weight coefficients, wherein the weight coefficients are determined by historical event data training; the risk misjudgment rate is calculated as the proportion of samples in normal state being misjudged as abnormal, the risk missed judgment rate is calculated as the proportion of samples in abnormal state not being identified, and the threshold sensitivity deviation reflects the response delay of the threshold to small changes in the environment, which is obtained by comparing the gradient of real-time environmental parameters with historical data; According to the functional attribute of the physical space of the apartment, a preset number of independent optimization subgroups are divided, each subgroup corresponds to a specific area and loads the historical risk feature subset exclusive to the specific area; the particle initialization range in the subgroup is limited within the historical optimal threshold interval of the specific area, ensuring that the search starting point matches the regional characteristics; In each iteration, the particle updates the position and velocity by tracking the threshold combination corresponding to the individual optimal solution of the historical minimum fitness and the threshold combination corresponding to the global optimal solution of the population historical minimum fitness; the initial value of the dynamic inertia weight is 0.9, which linearly decreases to 0.4 with the iteration number to enhance the local search ability in the later period, the learning factor is 2.0, and the random number range is 0 to 1; the upper and lower limits of each threshold are determined based on the historical data by threshold boundary constraint to prevent the particle position from exceeding the physically reasonable range; for the spatial differentiation demand, the regional correction coefficient is additionally introduced in the iteration of the subgroup to calibrate the particle position with regional characteristics; The global optimal solution sharing between subgroups is performed once every 10 generations of iteration, the fitness value of the optimal solution of each subgroup is calculated, and the global optimal solution is broadcast to each subgroup as a reference benchmark; at the same time, the regional compatibility constraint is set, when the threshold difference between adjacent regions exceeds the preset safety threshold, the boundary region particle collaborative optimization is triggered to ensure the reasonable connection of the threshold values of adjacent regions; When the iteration number reaches the preset value dynamically adjusted according to the complexity of the event or the fitness value fluctuation is less than 0.001 for 10 consecutive generations, the iteration is terminated, and the optimal particle position vector corresponding to the subgroups in each region is output to form the optimal risk assessment threshold combination matched with the functional characteristics of each space region of the apartment.

[0058] Specifically, by dynamically optimizing the risk assessment threshold combination through the particle swarm optimization algorithm, the threshold is deeply adapted to the apartment environment characteristics, regional functions, and real-time state. The technical effect is reflected in the coordinated improvement of multiple dimensions. First, by encoding the threshold combination as a particle position vector and initializing the search space based on historical data, the optimization process starts from the historical optimal threshold interval, avoiding blind search. At the same time, the number of particles is dynamically adjusted with the parameter dimension, ensuring that the search range covers potential optimal solutions without efficiency decline due to dimension expansion, laying a foundation for subsequent precise optimization. Second, the fitness function combines risk misjudgment rate, missed judgment rate, and threshold sensitivity deviation, and the weight coefficients are trained through historical data, making the optimization target more consistent with actual security needs. This reduces the interference of normal state being misjudged as abnormal and the risk of real risk being missed, while ensuring that the threshold can respond to small environmental changes in time and avoid delay due to insufficient sensitivity, fundamentally improving the practical value of the threshold combination. In fact, the design of independent optimization subgroups according to spatial functions allows the threshold optimization to adapt to different regional characteristics. For example, the environmental parameter fluctuation laws of kitchens and bedrooms differ significantly. By loading only the historical feature data of the corresponding region and limiting the particle initialization range to the regional historical optimal interval, the optimization starting point is highly matched with the regional functional characteristics, avoiding one-size-fits-all threshold adaptation deviations in complex spaces, and allowing each region's threshold to accurately reflect its risk characteristics. The adjustment of dynamic inertia weight in the iteration process balances global exploration and local convergence. The larger inertia weight in the early stage supports the particle to widely search for potential optimal solutions, and the decreasing weight in the later stage focuses on local fine optimization. Combined with the calibration of particle position by the regional correction coefficient, the threshold's adaptation ability to regional differentiation needs is further strengthened. The global optimal solution sharing between subgroups and regional compatibility constraints allow each subgroup to learn from global optimal experience and avoid risk assessment faults caused by large differences in threshold between adjacent regions, ensuring that the threshold system of the entire apartment space is coherent and reasonable. Finally, the regionalized optimal threshold combination output based on the iteration termination condition ensures the reliability of the optimization results through stable fitness fluctuations and achieves precise matching with regional functional characteristics through subgroup division. This dynamic optimization mechanism makes the risk assessment threshold no longer a fixed standard, but an intelligent scale that can be adjusted in real time with changes in the apartment environment and regional characteristics, providing a more accurate and adaptive judgment basis for environmental safety risk assessment models, and improving the accuracy and robustness of risk identification from the source.

[0059] As an optional embodiment, in step S103, first, an apartment space topology graph is constructed based on a graph neural network, room areas and device positions in the security space data set are mapped to nodes, the weight values of the edges between the nodes are defined in a physical connection relationship, a dynamic risk propagation path network is formed; through the message passing mechanism of the graph neural network, the real-time environmental feature parameters in the security space data set and the spatial attributes of the apartment space digital twin mapping model are fused, the diffusion speed, diffusion direction and influence range of abnormal phenomena in the apartment space are predicted in real time; the historical diffusion data of the computational fluid dynamics simulation are introduced to train the model parameters, the diffusion deviation under different ventilation conditions and house type structures is corrected, and the diffusion prediction result is matched with the actual physical environment characteristics of the apartment.

[0060] Exemplarily, assuming that gas leakage occurs in the kitchen of apartment 3, 2 units, 501 households, the system first constructs a space topology graph based on a graph neural network: the kitchen, living room, hallway, gas sensor, exhaust equipment, etc. of 501 households are mapped to nodes, and the weight values of the edges between the nodes are defined in the door and window connection relationship (such as the door between the kitchen and the living room, and the door between the living room and the hallway). The weight value of the edge is increased when the door is opened, representing the decrease of diffusion resistance. Through the message passing mechanism, the real-time gas concentration (0.05% in the kitchen), ventilation state (kitchen window closed) in the security space data set and the house type structure (kitchen is closed) in the digital twin model are fused, and the diffusion trend is predicted in real time: the gas may diffuse to the living room (concentration 0.03%) through the kitchen door within 10 minutes, and diffuse to the hallway (concentration 0.01%) within 30 minutes. At the same time, the computational fluid dynamics simulation data of similar house types (such as closed kitchen, window closed) are introduced to correct the diffusion speed deviation caused by insufficient ventilation, so that the prediction result is more matched with the actual physical environment characteristics. Through the space correlation modeling of the graph neural network and the correction of the physical diffusion law, the accuracy of the abnormal phenomenon diffusion prediction is improved, the risk spread path can be predicted in advance, which provides accurate spatial decision basis for subsequent regional isolation and equipment linkage, and avoids the waste of resources caused by blind expansion of prevention and control range.

[0061] Further, in step S103, the multi-dimensional risk factors are quantitatively coupled and analyzed by using the fuzzy comprehensive evaluation method, and the abnormal intensity, environmental characteristics, and personnel state are extracted from the security space data set as core evaluation indexes; based on the combination of historical risk event handling data and optimal risk assessment threshold, the weights of each index are dynamically determined by the analytic hierarchy process; the safety risk level is calculated and divided into four safety risk intervals, the key factors that significantly affect the risk level are identified through sensitivity analysis, the corresponding index weight is adjusted in real time, and the interference of local data deviation on overall evaluation is reduced. In the above gas leakage scene, the system uses the fuzzy comprehensive evaluation method to analyze the multi-dimensional risk factors: the abnormal intensity (gas concentration 0.05%, exceeding the warning threshold by 2 times), environmental characteristics (kitchen closed without ventilation, high diffusion coefficient), and personnel state (1 resident in the kitchen) are extracted from the security space data set as core indexes. Based on the combination of historical data and optimal threshold, the weights are dynamically determined by the analytic hierarchy process: the abnormal intensity weight is 0.4 (directly reflecting the risk level), the environmental characteristic weight is 0.3 (affecting the diffusion speed), and the personnel state weight is 0.3 (associated with personnel exposure risk). The risk level index is calculated as 7.8, corresponding to the high risk interval (four risk intervals: low <3, medium 3-5, high 5-8, and urgent >8). Through sensitivity analysis, it is found that the personnel state has a significant impact on the risk level (sensitivity coefficient 0.8), and the weight is temporarily adjusted to 0.35, reducing the evaluation deviation caused by personnel data fluctuations. The quantitative coupling analysis of multi-dimensional factors and dynamic weight adjustment solve the one-sidedness problem of single index evaluation, and the risk level division error is reduced. Even if there is a deviation in local data (such as temporary fluctuation of environmental sensors), the overall evaluation reliability can be ensured through key factor weight optimization, providing a scientific basis for risk classification and disposal.

[0062] Finally, in step S103, a Bayesian probability graph is constructed by fusing the personnel behavior recognition model and the environmental safety risk assessment model, the nodes of the Bayesian probability graph cover the personnel behavior state, the environmental parameter abnormal value, the historical event type, and the risk level label; the real-time personnel activity feature data and environmental monitoring data in the security space data set are used, combined with the associated event data in the historical learning library, to update the prior probability distribution of the probability graph; the security event type probability and the corresponding risk level posterior probability under the joint action of multiple factors are iteratively calculated by the Markov chain Monte Carlo method; according to the maximum probability event type and the risk level, combined with the event strategy mapping rules associated with the linkage priority of the security equipment, a linkage strategy containing the target equipment type, the linkage timing, and the control parameters is generated, and the strategy parameters are dynamically adjusted according to the safety risk level.

[0063] For example, the Bayesian probability graph constructed by the system contains nodes such as personnel behavior state (violation operation of gas stove), environmental parameter abnormal value (gas concentration 0.05%), and historical event type (similar gas leakage is mostly caused by operation error). Combining real-time data to update the prior probability assumption, the confidence level of the violation operation of the gas stove is 95%, the confidence level of the gas concentration exceeding the standard is 100%, and the co-occurrence probability of operation error -> gas leakage in historical data is 80%. Through Markov chain Monte Carlo method, the posterior probability is calculated iteratively to determine the probability of human operation error causing gas leakage event as 92%, corresponding to the high risk level probability of 90%. According to the event type and risk level, the device linkage priority rule is matched to generate the linkage strategy: the first-level device (gas valve) is immediately closed, the second-level device (kitchen exhaust) is started within 5 seconds, the third-level device (corridor sound and light alarm) is triggered within 10 seconds, and the control parameters are dynamically adjusted according to the risk level (exhaust intensity is set to maximum, alarm volume is adjusted to 80 decibels). The multi-factor probability reasoning of Bayesian probability graph and the fusion of historical data improve the accuracy of event type judgment and shorten the response time of linkage strategy generation to seconds. The strategy parameters are dynamically adapted according to the risk level, realizing precise prevention and control with higher risk, faster response, and stronger control. Compared with traditional fixed strategy, the disposal efficiency is improved, and the loss caused by risk diffusion is minimized.

[0064] Through spatial diffusion prediction by graph neural network, risk quantification by fuzzy comprehensive evaluation method, and multi-factor reasoning by Bayesian probability graph, the security intelligent decision engine realizes the whole-process intelligentization including risk identification, diffusion prediction, level evaluation, and strategy generation. The whole process not only combines physical laws and data-driven models to ensure evaluation accuracy, but also adapts to environmental changes through dynamic weight and probability update mechanisms. The final linkage strategy has the characteristics of spatial precision, reasonable timing, and strength adaptation, significantly improving the emergency disposal capability of apartment security events.

[0065] Further optionally, in step S103, the Bayesian probability graph of the fusion of the personnel behavior recognition model and the environmental safety risk assessment model is constructed, including: first, defining a multi-level node structure, the core node layer is divided into an input layer, a correlation layer and an output layer: the input layer includes personnel behavior feature nodes and environmental parameter nodes, the personnel behavior feature nodes are obtained based on normal behavior labels, abnormal behavior labels, behavior duration, behavior area output by the personnel behavior recognition model, and the environmental parameter nodes are obtained based on real-time monitoring values such as temperature, smoke concentration, and gas concentration of the environmental safety risk assessment model; the correlation layer includes space-time correlation nodes and historical correlation nodes; the output layer includes event type nodes and risk level nodes corresponding to four-level safety risk intervals. Then, the joint probability distribution of the personnel behavior feature nodes and the environmental parameter nodes is obtained by calculating the corresponding conditional probability through the co-occurrence samples of abnormal behaviors and environmental abnormalities in the historical data; the probability parameters of the space-time correlation nodes are calibrated in combination with the space-time anchoring relationship of the security space data set. Further, the real-time output of the personnel behavior recognition model is input as evidence into the personnel behavior feature nodes, and the marginal probability is dynamically updated; the threshold judgment result of the environmental safety risk assessment model is input as evidence into the environmental parameter nodes, and the conditional probability of the correlation layer nodes is updated through the evidence propagation algorithm. When new event handling data is added to the historical learning library, if the deviation of the association mode from behavior to environment event in the new event handling data from the existing probability table exceeds a preset threshold, the conditional probability table is automatically triggered for iterative updating, and the probability parameters of the correlation layer nodes are corrected by using the maximum likelihood estimation method. Finally, the input nodes that have a significant impact on the output layer nodes are identified through sensitivity analysis, the weight proportion of the input nodes in the probability reasoning is increased, and the recognition accuracy of the Bayesian probability graph for complex coupled events is improved.

[0066] Specifically, the Bayesian probability graph of the fusion of the personnel behavior recognition model and the environmental safety risk assessment model is constructed, first, starting from the hierarchical design of the node structure, and building a basic framework by clearly defining the functions and data sources of each layer of nodes. The core node layer is divided into an input layer, a correlation layer and an output layer, forming a complete logical chain from data access to result output. The input layer directly interfaces the outputs of the two basic models, the personnel behavior feature nodes carry the behavior labels, duration, region and other information parsed by the behavior recognition model, and the environmental parameter nodes integrate real-time data such as temperature, smoke concentration monitored by the environmental assessment model, ensuring that the original data can enter the probability reasoning process in the form of structured nodes. The correlation layer serves as an intermediate hub, the space-time correlation nodes focus on the time and space correlation of the data, and the historical correlation nodes link the regular features of past events, while the event type nodes and risk level nodes of the output layer directly correspond to the final decision-making needs, realizing hierarchical mapping from raw data to decision results.

[0067] In the construction and calibration phase of the probability parameters, the core is to convert historical data into quantifiable probability relationships to provide a basis for reasoning. The joint probability distribution between the personnel behavior characteristic nodes and the environmental parameter nodes is determined by mining the sample rules of abnormal behavior and environmental anomalies appearing simultaneously in historical data, so that the correlation strength of the two is quantitatively characterized. The probability parameters of the space-time correlation nodes are adjusted in combination with the space-time anchoring relationship of the security space data set to ensure that the probability correlation between nodes accurately reflects the space-time coupling relationship of behavior and environmental anomalies in the actual space, avoiding probability bias caused by space-time misalignment, and laying a reliable probability foundation for subsequent reasoning.

[0068] The dynamic input of real-time evidence and the probability updating mechanism enable the Bayesian probability graph to respond to new data in real time. When the personnel behavior recognition model outputs real-time results, these information will be directly input into the personnel behavior characteristic nodes as evidence, and the marginal probability of the nodes will be dynamically adjusted to reflect the changes in the current behavior state in a timely manner. The threshold judgment results of the environmental safety risk assessment model are input into the environmental parameter nodes as evidence, triggering the evidence propagation algorithm to spread and update in the entire network, driving the synchronous adjustment of the conditional probability of the correlation layer nodes, so that the probability graph always keeps consistent with the latest security state, ensuring the timeliness of the reasoning process.

[0069] In order to enable the model to continuously adapt to new scenarios and new data, the probability graph designs a closed-loop probability table iteration updating mechanism. When new event handling data is added to the historical learning library, the system will automatically compare the differences between the association mode of behavior and environmental events in the new data and the existing probability table. If the deviation exceeds the preset range, the conditional probability table updating process will be triggered immediately, and the probability parameters of the correlation layer nodes will be corrected through the maximum likelihood estimation method, so that the probability table can continuously absorb new experience rules, avoid reasoning bias caused by outdated historical data, and maintain the long-term adaptability of the model.

[0070] Finally, through sensitivity analysis, the model's ability to capture key factors is strengthened, further improving the recognition accuracy of complex coupled events. The system will identify input nodes that have a significant impact on the output layer nodes (event type and risk level), and by increasing the weight proportion of these nodes in the probability reasoning, the model can focus more on key influencing factors when dealing with complex scenarios with multiple interwoven factors, reducing the interference of secondary factors, and thus more accurately identifying complex events coupled with behavior and environmental anomalies, providing more reliable reasoning support for the generation of joint security strategies.

[0071] The Bayesian probability graph of the fusion of personnel behavior recognition model and environmental safety risk assessment model is constructed. First, the hierarchical design of node structure is started, and the basic framework is built by clarifying the functions and data sources of each layer node. The core node layer is divided into input layer, correlation layer and output layer, forming a complete logical chain from data access to result output. The input layer directly connects the output of the two basic models, the personnel behavior feature node carries the behavior label, duration, area and other information analyzed by the behavior recognition model, and the environmental parameter node integrates real-time data such as temperature and smoke concentration monitored by the environmental assessment model, ensuring that the original data can enter the probability reasoning process in the form of structured nodes. The correlation layer serves as an intermediate hub, with the time and space correlation node focusing on the time and space correlation of the data, and the historical correlation node linking the regular characteristics of past events. The event type node and risk level node of the output layer directly correspond to the final decision-making needs, realizing the hierarchical mapping from raw data to decision results.

[0072] In the construction and calibration stage of probability parameters, the key is to convert historical data into quantifiable probability relationships to provide a basis for reasoning. The joint probability distribution between the personnel behavior feature node and the environmental parameter node is determined by mining the sample rules of abnormal behavior and environmental anomalies appearing at the same time in historical data, so that the correlation strength of the two is quantitatively represented. The probability parameters of the time and space correlation node are adjusted in combination with the time and space anchoring relationship of the security space data set, ensuring that the probability correlation between nodes accurately reflects the time and space coupling relationship of behavior and environmental anomalies in the actual space, avoiding probability bias caused by time and space misalignment, and laying a reliable probability foundation for subsequent reasoning.

[0073] The dynamic input of real-time evidence and the probability updating mechanism enable the Bayesian probability graph to respond to new data in real time. When the personnel behavior recognition model outputs real-time results, these information will be directly input into the personnel behavior feature node as evidence, and by dynamically adjusting the marginal probability of the node, the changes in the current behavior state are timely reflected. The threshold judgment result of the environmental safety risk assessment model is input into the environmental parameter node as evidence, triggering the evidence propagation algorithm to spread and update in the entire network, driving the synchronous adjustment of the conditional probability of the correlation layer nodes, so that the probability graph always keeps consistent with the latest security state, ensuring the timeliness of the reasoning process.

[0074] In order to enable the model to continuously adapt to new scenarios and new data, the probability graph designs a closed-loop probability table iterative updating mechanism. When new event handling data is added to the historical learning library, the system will automatically compare the differences between the association mode of behavior and environmental events in the new data and the existing probability table. If the deviation exceeds the preset range, the probability parameter updating process of the conditional probability table will be triggered immediately, and the probability parameters of the correlation layer nodes are corrected by the maximum likelihood estimation method, so that the probability table can continuously absorb new experience rules, avoid reasoning deviation caused by outdated historical data, and maintain the long-term adaptability of the model.

[0075] Finally, the sensitivity analysis strengthens the model's ability to capture key factors, further improving the accuracy of identifying complex coupled events. The system identifies input nodes that significantly affect the output layer nodes (event type and risk level), and by increasing the weight of these nodes in the probability reasoning, the model can focus on key influencing factors and reduce the interference of secondary factors when dealing with complex scenarios with multiple intertwined factors, thus more accurately identifying complex events with abnormal coupling of behavior and environment, providing more reliable reasoning support for the generation of linked security strategies.

[0076] Step S104, based on the linkage strategy, sends corresponding control instructions and security control parameters to the corresponding target security device.

[0077] As an optional embodiment, in step S104, first, based on the target device type in the linkage strategy and the priority of the control array, structured control instructions are generated, including device unique identifier, operation type, precise control parameters and linkage timing requirements, and are associated with the corresponding spatio-temporal coordinates of the security space data set, ensuring that the instructions match the event occurrence area. Then, a pre-verification mechanism based on the digital twin mapping model of the apartment space is used to simulate the execution effect of the control instructions before sending, to verify whether there is a conflict in the device linkage logic. If a conflict is found in the simulation, the linkage timing parameters are automatically adjusted, and the control instructions are generated again after optimization. Finally, the control instructions are sent to the target security device through a standardized encrypted transmission protocol, and device identity authentication information and instruction check code are embedded in the transmission process to prevent tampering or misexecution of the instructions. Based on the node connectivity relationship of the security space topology graph, the instruction transmission path of the control instructions is optimized, and low-delay communication links are preferentially selected to ensure that the response delay of the core target security device does not exceed the preset time length.

[0078] Specifically, in the process of sending control instructions to target security devices based on linkage strategies, the generation of structured control instructions is first required. This process is based on the target device type and the priority of the control array in the linkage strategy, and converts abstract strategy requirements into specific executable instruction content. The instructions will include device unique identifier for precise positioning of target devices, clear operation type (such as start, stop or parameter adjustment), setting of precise control parameters adapted to the scene, and specification of linkage timing requirements (such as start interval and execution sequence) between devices. At the same time, these instructions will be associated with the corresponding spatio-temporal coordinates in the security space data set, ensuring that each instruction can accurately match the specific area where the event occurs, avoiding execution deviation due to spatial misalignment, and providing clear action guidelines for subsequent device collaborative response.

[0079] To ensure the accuracy of instruction execution and the coordination of device linkage, a pre-verification mechanism based on the digital twin mapping model of the apartment space will be introduced. Before the formal sending of control instructions, the digital twin model is used to simulate the execution effect of the instructions, virtually restoring the linkage process of the devices in the actual space. Through this simulation, potential logical conflicts between device actions can be identified in advance, such as timing conflicts or functional overlap interference between different devices in the same area. If conflicts are found, the linkage timing parameters will be automatically adjusted, and the final control instructions will be generated after optimization. This step avoids potential problems in actual execution through digital space pre-verification, ensuring the smoothness of device linkage and the effectiveness of strategy execution.

[0080] The transmission of control instructions focuses on the dual protection of safety and reliability. Standardized encryption transmission protocols are used to ensure data security during network transmission. Device identity authentication information is embedded during transmission, and only the target device with the corresponding identity can receive and analyze the instructions. Additionally, an instruction check code is attached to verify the integrity of the instructions, preventing tampering or data corruption during transmission that could lead to misexecution. This encryption and authentication mechanism builds a security barrier, ensuring the trustworthiness of the entire process from instruction issuance to reception, and avoiding unauthorized operations or instruction abnormalities that could interfere with the security system.

[0081] To further improve the efficiency of instruction response, the transmission path will be optimized based on the node connectivity relationship of the security space topology graph. By analyzing the connectivity state and communication link performance of device nodes in the topology graph, low-latency communication paths are preferentially selected to send control instructions. This path optimization mechanism ensures that core target security devices can receive instructions and respond as quickly as possible, avoiding transmission delays that could affect emergency response effectiveness. Especially in critical security incidents, efficient instruction transmission paths can ensure that core devices respond in the shortest time, gaining valuable time for risk control and improving the real-time performance and reliability of the overall security system.

[0082] Further optionally, the digital twin model simulates the execution effect of the control instruction in the pre-performance verification. The core is to reproduce the dynamic process of equipment linkage and check the logical rationality through the accurate mapping of the digital mirror image of the physical space. This process begins with real-time data loading of the apartment space digital twin mapping model. The model synchronizes the current environment state (such as device operating parameters, space layout features) in the security space data set and the details of the control instruction to be executed, converts the device attributes, space relationships, and environmental conditions of the physical world into computable parameters in the digital space, and builds a virtual test environment consistent with the actual scene. In the simulation execution phase, the model will generate virtual action sequences for each target security device according to the linkage timing requirements in the control instruction. For example, when simulating the action of closing the gas valve, the location coordinates of the valve in the digital twin model are associated, the time consumption of the closing process and the virtual influence on the surrounding gas concentration are calculated; when simulating the start of the exhaust device, the ventilation path in the space topology is combined to deduce the effect of air flow changes on environmental parameters. At the same time, the model will introduce physical rules and device characteristic parameters, such as simulating response delay based on device performance curve and simulating signal propagation range based on space structure, so that the effect of virtual action is close to the actual physical law, avoiding logical disconnection in the pure digital level. During the simulation process, the system will focus on monitoring the coordination and potential conflicts of the equipment linkage logic. By tracking the action interaction of different devices in the virtual space, such as whether there is functional interference between the start sequence of the sprinkler and the exhaust device in the same area, or whether there is logical contradiction between the lock of the access control and the opening of the emergency passage, the model can capture problems such as action overlap and timing conflict in real time. If it is found that the start of a device will affect the monitoring accuracy of another device (such as the camera field of view blocked by the sprinkler water mist), the model will record the conflict node and feed back to the system, providing a basis for subsequent adjustment of the linkage timing parameters. After the simulation is completed, the digital twin model will generate a complete execution effect report, including the virtual action trajectory of each device, the environmental parameter change curve, and the conflict detection result. These information are used to verify the feasibility of the control instruction: if the simulation result meets the expected linkage strategy and there is no conflict, the instruction will be confirmed to be sent. If there is a conflict or the execution effect deviates from the expectation, the model will mark the problem link to assist the system in automatically optimizing the linkage timing or parameter settings until the virtual execution effect meets the safety requirements. Through this pre-performance in the digital space, the digital twin model realizes the pre-performance, verification, and optimization closed loop of the equipment linkage process in the physical world, ensuring that the device actions are coordinated and the strategy is effectively implemented in the actual execution.

[0083] Further optionally, after step S104, the state data of the target security device can also be collected in real time during the process and fed back to the security intelligent decision engine, and the control parameters are dynamically adjusted in combination with the expected effect in the linkage strategy and the real-time state data. If the risk spreading speed exceeds the expectation, the spraying intensity is automatically increased, and the alarm coverage range is expanded. If the personnel evacuation is slow, the flashing frequency and voice guidance frequency of the escape indication device are increased. Further, a cross-device linkage state synchronization mechanism is established, and the execution states of multiple devices in the same event area are associated to the same space-time index of the security space data set through a space-time correlation algorithm, ensuring the coordination of device actions. When a time sequence deviation is detected in the device linkage, a calibration instruction is automatically sent to correct the action timing, avoiding the decline of emergency efficiency caused by different synchronization of device responses. After the execution of the instruction is completed, the device actual execution parameters, response time, linkage effect data are associated to the historical learning library for optimizing the control parameter configuration and device priority weight in the subsequent linkage strategy, forming a closed-loop control mechanism including strategy generation, instruction execution, effect feedback, and parameter optimization.

[0084] Specifically, after the control instruction is sent to the target security device, it will enter the real-time monitoring and dynamic adjustment stage to ensure that the execution effect of the linkage strategy is accurately matched with the risk disposal demand. During this process, the state data of the target security device will be collected in real time, including device running parameters, action execution progress, environmental feedback results, etc. These data are continuously fed back to the security intelligent decision engine through the Internet of Things link. The decision engine compares and analyzes the real-time state data with the expected effect preset in the linkage strategy. If it is found that the risk spreading speed is faster than the expectation, such as the smoke spreading range exceeding the simulation prediction, the control parameters will be automatically adjusted to strengthen the response intensity of the corresponding device. If it is monitored that the personnel evacuation progress is lagging behind, such as the personnel staying time in the escape passage being too long, the flashing frequency and voice guidance frequency of the escape indication device are increased to optimize the evacuation guidance effect. This dynamic adjustment mechanism makes the strategy execution no longer limited to preset instructions, but can be flexibly adapted according to the actual risk changes, improving the pertinence of emergency disposal.

[0085] To ensure the coordination of multi-device linkage, the system establishes a cross-device linkage state synchronization mechanism. With the help of space-time correlation algorithm, the execution states of all devices participating in linkage in the same event area will be accurately correlated to the same space-time index of the security space data set, forming a unified dynamic execution view. This means that the action progress and response state of different devices can be aligned in time and space dimensions, such as the start time of the sprinkler device and the running state of the smoke exhaust device will be bound to the same event period and regional coordinates. When the system detects a timing deviation in device linkage, such as a regional alarm device starting later than the preset time, causing the actions of other devices to be out of sync, it will immediately send a calibration instruction to adjust the action time of the relevant devices, ensuring that each device is consistent in time and space coverage, avoiding the prevention and control loopholes or resource waste caused by different response speeds.

[0086] After the completion of the instruction execution, the device actual execution parameters, response time data, linkage effect feedback and other information generated during the entire process will be automatically associated by the system to the historical learning library, becoming an important basis for optimizing subsequent strategies. These data will be used to analyze the effects of different control parameters in actual scenarios, such as the actual effect of a specific sprinkler intensity on fire control, or the impact of different alarm frequencies on the efficiency of personnel evacuation, and then adjust the control parameter configuration in future linkage strategies. At the same time, the response performance of the device in the event will also be fed back to the device priority weight calculation model, allowing more accurate determination of the priority order of core devices and auxiliary devices when generating subsequent strategies. This complete closed loop from strategy generation, instruction execution to effect feedback and parameter optimization enables the security system to continuously accumulate experience through continuous learning, gradually improving the adaptability and execution efficiency of linkage strategies, and achieving dynamic evolution of intelligent level.

[0087] Further optionally, after step S104, execution feedback information of the target security device can also be received in real time. If it is detected that the target security device does not execute according to the control instruction, the control instruction is automatically resent or the standby device is switched to execute the corresponding control instruction. The execution feedback information returned by the target security device, including execution status, execution progress and exception code, is received in real time, and the execution feedback information is associated with the space-time coordinate corresponding to the security space data set. If it is detected that the target security device does not execute according to the control instruction, based on the spatial correlation relationship of the security space data set, the standby security device closest to the event occurrence area in space-time distance is selected from the control array, a secondary control instruction containing device switching reason and adjusted control parameter is generated; the target security device does not execute according to the control instruction includes: execution timeout, execution failure, or execution parameter deviates from the preset threshold. If the standby security device still does not meet the requirements, or the core target security device fails continuously, the automatic escalation linkage strategy level is triggered, the device linkage response of higher priority is triggered, and the device failure information and event handling trajectory are synchronized to the apartment management center; the device linkage response includes: expanding the alarm range, increasing the spraying coverage area. Finally, the device execution feedback data, failure type and handling result of this time are input into the historical learning library of the security intelligent decision engine, which is used to optimize the judgment logic of the personnel behavior recognition model and the environmental safety risk assessment model and dynamically adjust the device priority configuration of the control array.

[0088] Specifically, after sending control instructions to the target security device, the system receives real-time execution feedback information returned by the device, thereby dynamically monitoring the execution status of the instructions. These feedback information includes the real-time execution status, current execution progress, and possible abnormal codes of the device, which can comprehensively reflect the response of the device to the instructions. At the same time, the system accurately associates these execution feedback information to the corresponding space-time coordinates in the security space data set, ensuring that the feedback data corresponds to the specific area and time node of the event occurrence, providing a clear spatial and temporal reference for subsequent fault analysis and strategy adjustment, so that each feedback can be traced back to a specific event scene. When the system detects that the target security device does not execute according to the control instructions, it will immediately start the emergency handling mechanism. The failure to execute according to the instructions includes various situations, such as the device executing the operation beyond the time limit, explicitly returning an execution failure signal, or the execution parameters deviating from the preset reasonable range. At this time, based on the spatial association relationship of the security space data set, the system quickly retrieves the standby security device closest to the event occurrence area in time and space in the control array. After selecting the standby device, a secondary control instruction is generated, which clearly marks the reason for the device switching and the control parameters adjusted according to the characteristics of the new device, ensuring that the standby device can seamlessly take over the task of the original device and avoid the interruption of security response due to the failure of a single device. If the standby security device still does not achieve the expected effect after execution, or if the core target security device has continuous failures, the system will automatically trigger the upgrade of the linkage strategy level. This means starting a higher priority device linkage response, such as expanding the coverage range of the alarm to alert more people in the area, or increasing the coverage area of the sprinkler system to enhance the risk control effect. At the same time, the system synchronizes the detailed information of the device failure (such as failure type, occurrence time) and the complete trajectory of event handling to the apartment management center, so that the management personnel can timely grasp the on-site situation and intervene in the processing, thereby reducing the possibility of risk spreading caused by device failure through multi-level response mechanism. After each event handling is completed, the system will input the execution feedback data of the device, the specific failure type, and the final handling result into the historical learning library of the security intelligent decision engine. These data will become an important basis for the judgment logic of the optimization model, helping the personnel behavior recognition model and the environmental safety risk assessment model to more accurately identify abnormalities and assess risks. At the same time, based on the performance of the device in the event, the system will dynamically adjust the device priority configuration in the control array, so that devices with higher reliability and faster response can obtain higher priority in future linkage strategies, thereby improving the fault tolerance and response efficiency of the entire security system through continuous experience accumulation.

[0089] In step S105, the linkage priority of the security device is set according to the mapping relationship between the security event type, the security device and the emergency degree of the security event, a control array containing target security devices at all levels is constructed according to the linkage priority in combination with the current security event type of the apartment, and the core target security device is triggered to respond to the indication of the linkage strategy through the control array, so as to realize remote control and linkage response of the target security device.

[0090] In the embodiment of the application, the mapping relationship between the security device and the emergency degree of the security event is a corresponding association rule pre-established by the system according to the emergency degree, risk type of the security event and the functional characteristics of the security device. The mapping relationship is constructed based on the threat degree of the event to personnel safety and property protection and the core role of the device in handling the event, so as to ensure that events of different emergency degrees can be matched to the most suitable device resources. For example, for a high-emergency-degree gas leakage event, the mapping relationship will preferentially associate core devices such as gas valve closing devices and exhaust devices that can directly contain the risk spread. For a low-emergency-degree door / window closing event, the main association is a local sound-light reminding device or a resident APP push function, so as to avoid excessive investment of resources. Through this mapping, the system can quickly locate the key devices required by the event and provide a basis for subsequent linkage response.

[0091] The linkage priority is a sorting rule for the security devices participating in the response according to importance and disposal time sequence on the basis of the mapping relationship. The core is to clarify which devices should be executed first and which devices should be followed up in the multi-device collaborative response, so as to ensure that the key operation is completed before the auxiliary operation. The setting of the priority is usually based on the irreplaceability of the device in the event handling: the core device (such as a gas valve and a fire sprinkler starting device) is given the highest priority because it directly determines the risk control effect. The auxiliary device (such as a sound-light alarm and a ventilation device) is used to expand the warning range or strengthen the handling effect, and the priority is second. The support device (such as a property terminal notification and an environmental monitoring device) is used for information synchronization and state feedback, and the priority is relatively low. For example, in a fire event, the priority of the device starting the fire sprinkler is higher than that of the corridor alarm device, so as to ensure that the fire is controlled first and then the warning is expanded, and the handling delay caused by the reverse order is avoided.

[0092] The control array is an ordered device execution sequence based on linkage priority, which is a carrier for converting mapping relationship and priority rules into specific execution scheme. It screens out adaptive devices from the mapping relationship according to the type and emergency level of the current security event, and arranges them in order from high to low according to the linkage priority to form a structured set containing target security devices at all levels. The control array not only clearly defines the execution order of the devices, but also contains the control parameters of the devices, the linkage timing requirements and the fault backup scheme. For example, in the high-emergency level illegal intrusion event, the first echelon of the control array is the access control locking device and the core area camera, which ensures that the area is isolated first and the evidence is preserved. The second echelon is the sound and light alarm device and the security terminal notification, which expands the warning range and links to manual disposal. This array structure ensures that the core devices respond first, and through hierarchical division, step-by-step disposal is achieved. Even if some devices fail, the system can quickly switch to backup devices based on the priority order in the array, ensuring the continuity and reliability of the response.

[0093] As an optional embodiment, in step S105, a security event-device linkage mapping rule library is constructed in advance, and is divided into a first-level classification according to event types (such as fire, gas leakage, abnormal intrusion, and personnel accident, etc.), each type of event is associated with a corresponding core security target (such as “fire extinguishing + evacuation” for fire, and “interception + warning” for abnormal intrusion). The emergency degree index of each type of event is quantified based on historical emergency disposal data (with a value of 0-10, and the higher the value, the higher the emergency degree), and the basic priority weight of the associated device is matched for each type of event (such as the weight of the sprinkler system in the fire event is 0.9, and the weight of the fire alarm is 0.85). The device-event correlation weight matrix is constructed, and the matrix dimension includes event type, device function type, and spatial correlation degree: the device function type weight is determined according to the control efficiency of the device on the event (such as the weight of the gas valve closing device is higher than that of the alarm device in the gas leakage event). The spatial correlation degree weight is calculated based on the physical coordinates of the security space data set, and the weight of the device within 3 meters of the event occurrence area is increased by 20%, the weight of the device within 5-10 meters is increased by 10%, and the weight of the device beyond 10 meters is reduced by 5%, to ensure that the devices in close proximity are preferentially responded. The priority is dynamically adjusted in combination with the real-time health status of the device: the device operation state data (such as online status, response delay, and historical fault record) is called, and the priority weight of the device with a health degree lower than a preset threshold (such as health index < 60) is reduced (by 10%-30%), and the weight of the standby device is increased (by 15%-20%), to avoid the influence of the faulty device on the linkage efficiency. The control array is hierarchically constructed according to the priority from high to low: the first-level array includes the core target device (weight ≥ 0.8), which is responsible for directly disposing the event (such as the sprinkler system in the fire, and the valve closing device in the gas leakage). The second-level array includes the auxiliary device (weight 0.5-0.8), which is responsible for environmental regulation and warning (such as smoke exhaust device, and sound and light alarm). The third-level array includes the support device (weight < 0.5), which is responsible for data acquisition and background synchronization (such as enhanced camera, and management center communication device). The priority real-time calibration mechanism is established: in the event disposal process, according to the risk level change (such as the risk level from medium to emergency) output by the security intelligent decision engine, the priority weight of the core device is automatically increased, and the second-level array device is triggered to start in advance. When it is detected that the first-level array device performs abnormally, the second-level array device corresponding to the function is upgraded to the first-level array, to ensure that the linkage strategy is continuously effective.

[0094] Further, in step S105, the control array realizes the priority response of the core target security equipment to the linkage strategy through hierarchical structure design and dynamic priority management, and further achieves remote control and multi-device collaborative linkage. The core logic starts from the hierarchical construction of the control array. Based on the mapping relationship between the security event type and the emergency degree, the first array is clearly defined as the core target device set, and the weight of such devices is ≥0.8. The function is directly directed to the key disposal demand of the event. For example, the sprinkler system in case of fire and the valve closing device in case of gas leakage. They occupy the highest priority in the array and provide a structural basis for priority response. The system accurately selects the core device that adapts to the current event from the mapping rule library through the preset device event correlation weight matrix, combined with the spatial correlation degree of the event occurrence area (device weight promotion within 3 meters) and the real-time health status of the device (healthy index standard device priority), to ensure that the devices in the first array have functional adaptability and execution reliability.

[0095] In the generation and sending link of remote control instructions, the priority mechanism of the control array dominates the execution order of the instructions. When the linkage strategy is triggered, the system first analyzes the core disposal target in the strategy and takes the first array device as the primary object of instruction sending. The generated structured control instruction contains the unique identifier of the core device, the accurate operation type (such as “immediately close”, “maximum strength start”) and the linkage timing starting point, ensuring that the instruction is accurately matched with the functional characteristics of the core device. Through the standardized encryption transmission protocol, the instruction is preferentially sent to the core device through the low-delay communication link. The device identity authentication and check code are embedded in the transmission process to ensure the safety and uniqueness of the instruction from generation to reception. This core device priority transmission mechanism ensures that the first array device receives the instruction and starts the response in the shortest time, realizing the instantaneity of remote control.

[0096] The synergy of linkage response is realized through the priority timing within the array. After the core target device responds to the instruction, the subsequent linkage is triggered based on the hierarchical relationship of the control array: the first array device executes the key operation at the same time, and the second array device (such as smoke exhaust equipment, sound and light alarm) starts to delay according to the preset timing, auxiliary strengthening the disposal effect of the core device; the third array device (such as enhanced camera, management center communication equipment) synchronously starts data collection and information uploading, providing real-time feedback for the decision engine. This linkage logic of core first and auxiliary follow-up avoids the disorderly superposition of device actions, ensures that the key operation of the core device is preferentially landed, and the function of the auxiliary device is effectively supplemented.

[0097] In view of the dynamic changes in event handling, the real-time calibration mechanism of the control array further guarantees the continuous priority response of the core equipment. When the security intelligent decision engine detects that the risk level is upgraded (such as from medium risk to emergency), the system automatically increases the priority weight of the first-level array equipment, strengthens its control parameters (such as increasing the spraying intensity), and compresses the response delay. If it is detected that the first-level array equipment performs abnormally (such as timeout without response, parameter deviation), immediately select the equipment with replacement function from the second-level array, automatically upgrade it to the first-level array, generate secondary control instructions and send them preferentially, and maintain the continuity of the core handling capability through dynamic replacement. This priority dynamic calibration mechanism ensures that the control array can still guarantee the priority response of the core target when the equipment fails or the risk changes, and ultimately realizes the precise triggering of remote control of core equipment and efficient cooperation of multi-device linkage.

[0098] In the embodiments of the present application, the real-time collected multi-type security data forms a data set accurately matched with the physical space after spatio-temporal correlation processing, providing a unified and structured basis for subsequent analysis, breaking down the barriers between different device data. The security intelligent decision engine analyzes personnel behavior characteristics with the help of deep learning algorithms, quantifies environmental abnormal diffusion trends combined with dynamically optimized risk assessment thresholds, can more accurately distinguish between normal and abnormal states, accurately judge the safety risk level, and reduce the deviation and lag of traditional manual judgment. Based on the device linkage priority mechanism constructed according to the event type and the degree of urgency, the core equipment can respond preferentially, and through the control array, the orderly cooperation of various security devices is realized, avoiding the low efficiency problem of isolated operation of devices. Even if some devices perform abnormally, the system can automatically switch to backup devices or upgrade strategies to ensure the continuity of the handling process. At the same time, the entire process forms a complete closed loop from data collection, analysis and decision-making to device linkage and effect feedback. The model parameters and linkage strategies can be continuously optimized based on historical data, adapting to changes in the apartment environment, improving the safety of tenants while providing more efficient and intelligent security management for apartment managers, and promoting the transformation of apartment security from passive response to active prevention.

[0099] Please refer to Figure 2 , Figure 2 The application provides an apartment security linkage system based on artificial intelligence, which comprises a collection module, a construction module, and an intelligent decision engine. The construction module is used for forming security space data sets matched with the apartment space according to the spatio-temporal correlation of the real-time collected initial security data. The decision-making module employs a security intelligent decision-making engine to jointly analyze the security spatial dataset to determine the types of security events and security risk levels within the apartment, and generates corresponding response strategies based on the judgment results. The security intelligent decision-making engine includes a personnel behavior recognition model and an environmental security risk assessment model. The personnel behavior recognition model uses deep learning algorithms to analyze the characteristics of personnel activities in the security spatial dataset in real time to distinguish between normal and abnormal behaviors. The environmental security risk assessment model combines historical security risk data with dynamically optimized risk assessment thresholds, and quantitatively analyzes the spread trend and impact range of abnormal personnel behaviors and environmental anomalies in the current environment based on the security spatial dataset to determine the current security risk level. The execution module is used to send corresponding control commands and security control parameters to the corresponding target security devices based on the linkage strategy; set the linkage priority of security devices according to the mapping relationship between security event type, security device and security event urgency; construct a control array containing target security devices at all levels according to the linkage priority based on the current security event type of the apartment; and trigger the core target security devices to respond to the instructions of the linkage strategy through the control array to realize remote control and linkage response of target security devices.

[0100] In some implementations, the AI-based apartment security linkage system can be applied to terminal devices. It should be noted that, for the sake of convenience and brevity, the specific working process of the AI-based apartment security linkage system described above can be referred to the corresponding process in the aforementioned embodiments of the AI-based apartment security linkage method, and will not be repeated here.

[0101] Please see Figure 3 , Figure 3 This is a schematic block diagram of the structure of a terminal device provided in an embodiment of this application.

[0102] like Figure 3 As shown, the terminal device 300 includes a processor 301 and a memory 302, which are connected via a bus 303, such as an I2C bus. Specifically, the processor 301 provides computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit, or it can be other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Those skilled in the art will understand that... Figure 3The structure shown in the figure is only a block diagram of part of the structure related to the embodiment of the present application, and does not constitute a limitation on the terminal device to which the embodiment of the present application is applied. The specific server can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement. The processor is used to run the computer program stored in the memory, and when the computer program is executed, any one of the artificial intelligence-based apartment security linkage methods provided by the embodiments of the present application is implemented. It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the terminal device described above can refer to the foregoing artificial intelligence-based apartment security linkage method embodiments, which will not be described here.

[0103] The embodiment of the present application also provides a storage medium for computer readable storage, the storage medium stores one or more programs, the one or more programs can be executed by one or more processors to implement the steps of any one of the artificial intelligence-based apartment security linkage methods provided in the specification of the embodiments of the present application.

Claims

1. An apartment security linkage method based on artificial intelligence, characterized by, The method comprises: Real-time acquisition of initial security data collected by various monitoring devices in the apartment, wherein the initial security data includes video image data, door and window status data, environmental safety data, and personnel access data; Forming a security space data set matching the apartment space according to the real-time acquired initial security data according to the space-time correlation relationship; Using a security intelligent decision engine to jointly analyze the security space data set to determine the type of security event and the safety risk level existing in the apartment, and generate a corresponding linkage strategy according to the judgment result; wherein the security intelligent decision engine includes a personnel behavior recognition model and an environmental safety risk assessment model; the personnel behavior recognition model analyzes the personnel activity characteristics in the security space data set in real time through a deep learning algorithm to distinguish between normal behavior and abnormal behavior; the environmental safety risk assessment model combines the risk assessment threshold dynamically optimized based on historical security risk data, quantitatively analyzes the diffusion trend and influence range of personnel abnormal behavior and environmental abnormal phenomena in the current environment based on the security space data set, to determine the safety risk level of the current environment; Based on the linkage strategy, corresponding control instructions and security control parameters are sent to the corresponding target security device; the linkage priority of the security device is set according to the mapping relationship between the security event type, the security device and the emergency degree of the security event, and a control array including target security devices at all levels is constructed according to the linkage priority based on the current security event type of the apartment, and the core target security device is triggered to respond to the instruction of the linkage strategy through the control array, realizing remote control and linkage response of the target security device.

2. The method of claim 1, wherein, After the corresponding control instructions and security control parameters are sent to the corresponding target security device based on the linkage strategy, the method further comprises: Real-time receiving of execution feedback information returned by the target security device, and if it is detected that the target security device does not execute the control instruction, automatically re-sending the control instruction or switching to a backup device to execute the corresponding control instruction; Real-time receiving of execution feedback information returned by the target security device, and if it is detected that the target security device does not execute the control instruction, automatically re-sending the control instruction or switching to a backup device to execute the corresponding control instruction; If it is detected that the target security device does not execute the control instruction, based on the spatial correlation relationship of the security space data set, the backup security device closest to the space-time distance of the event occurrence area is selected from the control array, and a secondary control instruction including the device switching reason and the adjusted control parameter is generated; the target security device does not execute the control instruction, including: execution timeout, execution failure, or execution parameter deviating from the preset threshold; If the backup security device still does not meet the requirements or the core target security device fails continuously, the linkage strategy level is automatically upgraded, triggering a higher priority device linkage response, and synchronizing the device failure information and event handling trajectory to the apartment management center; the device linkage response includes: expanding the alarm range, increasing the spraying coverage area. The feedback data, fault type and treatment result of the equipment are input into the historical learning library of the security intelligent decision engine to optimize the judgment logic of the personnel behavior recognition model and the environmental safety risk assessment model and dynamically adjust the equipment priority configuration of the control array.

3. The method of claim 1, wherein, The initial security data obtained in real time is formed into a security space data set matched with the apartment space according to the space-time correlation relationship, which comprises: The initial security data obtained in real time is preprocessed to remove invalid data, noise data and repeated collection data through an outlier detection algorithm to obtain valid security data; Based on the unified data format standard, the heterogeneous data collected by different types of monitoring equipment are format-converted and standardized, wherein the video image data are converted into frame sequence structured data, the door and window state data, environmental safety data and personnel access data are converted into key-value pair structured data, and the multi-source data format compatibility is ensured; The physical coordinate information of each monitoring equipment is extracted and an apartment space digital twin mapping model is established, the valid security data after preprocessing is time and space anchored according to the collection time stamp and the corresponding physical coordinate information, and each initial security data is associated to the corresponding position node in the apartment space digital twin mapping model; Through a space-time correlation algorithm, the multi-source initial security data of the same time segment and adjacent space region are associated and fused to form a multi-dimensional index structure containing data collection time, space coordinates, data type and data value, and a security space data set consistent with the actual physical space of the apartment is constructed, and the dynamic security space data set dynamically updates the space-time index relationship with the initial security data collected in real time.

4. The method of claim 1, wherein, The step of dynamically optimizing the risk assessment threshold combined with the historical security risk data further comprises: Classify and label the historical security risk data according to event types, extract the environmental characteristic parameters and risk evolution rule data corresponding to each type of apartment security event, and construct a classified historical risk feature library; wherein the environmental characteristic parameters corresponding to each type of apartment security event at least include temperature, humidity, gas concentration reference value and personnel density; Based on the classified historical risk feature library, match the real-time environmental characteristic parameters with the environmental characteristic parameters of the same type of historical apartment security event, and use a particle swarm optimization algorithm to minimize the risk misjudgment rate and the missing rate as the objective function to automatically search for the optimal risk assessment threshold combination matching the current apartment space region and real-time environmental state; the optimal risk assessment threshold combination at least includes the warning threshold of different environmental characteristic parameters and the risk level division threshold; Set a time sliding window, and dynamically adjust the window length according to the historical occurrence frequency of the apartment security event to shorten the window length of high-frequency apartment security events to 1 hour and extend the window length of low-frequency apartment security events to 24 hours; through the time sliding window, the environmental monitoring data and event treatment results in the security space data set are included in real time, and the optimal risk assessment threshold combination is dynamically calibrated; The calibrated optimal risk assessment threshold combination is input into the environmental safety risk assessment model to verify the identification accuracy of the calibrated optimal risk assessment threshold combination for the current environmental abnormal phenomenon in real time; if the accuracy of the identification accuracy is lower than a preset threshold for three or more times in succession, a new round of particle swarm optimization search is triggered to form a closed-loop threshold optimization mechanism to ensure that the risk assessment threshold is adapted to the dynamically changing environmental characteristic parameters of the apartment.

5. The method of claim 4, wherein, The real-time environmental characteristic parameters are matched with the environmental characteristic parameters of the same type of apartment security events in history based on the classified historical risk feature library, and a particle swarm optimization algorithm is used to minimize the risk misjudgment rate and the missed judgment rate as the objective function to automatically search for the optimal risk assessment threshold combination that matches the current apartment space area and the real-time environmental state, including: The risk assessment threshold combination including the temperature warning threshold, the smoke concentration threshold, the gas leakage threshold, and the personnel density threshold is encoded as a particle position vector, and the vector dimension corresponds to the type of environmental characteristic parameters one by one; based on the threshold distribution range of historical security events of the same type of apartment in the classified historical risk feature library, the particle swarm position and velocity are randomly initialized, and the number of particles is dynamically set according to the dimension of the environmental characteristic parameters to ensure that the search space covers the historical optimal threshold interval; The risk misjudgment rate, the risk missed judgment rate, the threshold sensitivity deviation, and their respective weight coefficients are used to construct the fitness function, wherein the weight coefficients are determined by historical event data training; the risk misjudgment rate is calculated as the proportion of samples that are misjudged from normal state to abnormal state, the risk missed judgment rate is calculated as the proportion of samples that are not identified from abnormal state, and the threshold sensitivity deviation reflects the response delay of the threshold to small changes in the environment, which is obtained by gradient comparison of real-time environmental parameters and historical data; According to the function attribute of the physical space of the apartment, a preset number of independent optimization subgroups are divided, each subgroup corresponds to a specific area and loads the historical risk feature subset exclusive to the specific area; the particle initialization range in the subgroup is limited within the historical optimal threshold interval of the specific area to ensure that the search starting point matches the regional characteristics; In each iteration, the particle takes the threshold combination corresponding to the historical lowest fitness as the individual optimal solution, and the threshold combination corresponding to the historical lowest fitness of the population as the global optimal solution to update the position and velocity; the dynamic inertia weight initial value is 0.9, which linearly decreases to 0.4 with the iteration number to enhance the local search ability in the later stage, the learning factor is 2.0, and the random number range is 0 to 1; the upper and lower limits of each threshold are determined based on historical data through threshold boundary restriction to prevent the particle position from exceeding the physical reasonable range; for the differentiated needs of space, a regional correction coefficient is additionally introduced during iteration in the subgroup to calibrate the particle position according to the regional characteristics; Global optimal solution sharing between subgroups is performed once every 10 generations, the fitness values of the optimal solutions of each subgroup are calculated, and the global optimal solution is broadcast to each subgroup as a reference benchmark; at the same time, regional compatibility constraints are set, and when the threshold difference between adjacent regions exceeds a preset safety threshold, boundary region particle collaborative optimization is triggered to ensure the reasonableness of the threshold connection between adjacent regions; When the number of iterations reaches the preset value dynamically adjusted according to the complexity of the event or the fitness value fluctuation is less than 0.001 for 10 consecutive generations, the iteration is terminated, and the optimal particle position vector corresponding to the subpopulation in each region is output to form an optimal risk assessment threshold combination matched with the functional characteristics of each space region of the apartment.

6. The method of claim 1, wherein, The security and protection space data set is jointly analyzed by using the security and protection intelligent decision engine to determine the type of security and protection event and the security risk level existing in the apartment, and a linkage strategy is generated according to the determination result, including: A graph neural network is used to construct an apartment space topology graph, the room regions and device positions in the security and protection space data set are mapped as nodes, the weight values of the edges between the nodes are defined according to the physical connection relationship, and a dynamic risk propagation path network is formed; through the message passing mechanism of the graph neural network, the real-time environmental feature parameters in the security and protection space data set and the spatial attributes of the apartment space digital twin mapping model are fused to predict the diffusion speed, diffusion direction and influence range of abnormal phenomena in the apartment space in real time; historical diffusion data of computational fluid dynamics simulation are introduced to train model parameters, correct the diffusion deviation under different ventilation conditions and house type structures, and match the diffusion prediction result with the actual physical environmental characteristics of the apartment; A fuzzy comprehensive evaluation method is used to quantitatively couple and analyze multi-dimensional risk factors, and abnormal intensity, environmental features and personnel state are extracted from the security and protection space data set as core evaluation indexes; based on historical risk event handling data and the optimal risk assessment threshold combination, the weights of each index are dynamically determined by an analytic hierarchy process; the security risk level is calculated and divided into four security risk intervals, the key factors that significantly affect the risk level are identified through sensitivity analysis, the corresponding index weights are adjusted in real time, and the interference of local data deviation on the overall evaluation is reduced; A Bayesian probability graph integrating a personnel behavior recognition model and an environmental security risk assessment model is constructed, the nodes of the Bayesian probability graph cover personnel behavior state, environmental parameter abnormal value, historical event type and risk level label; real-time personnel activity feature data and environmental monitoring data in the security and protection space data set are used to update the prior probability distribution of the probability graph in combination with the associated event data in the historical learning library; the security event type probability and the corresponding risk level posterior probability under the joint action of multiple factors are iteratively calculated by a Markov chain Monte Carlo method; according to the maximum probability event type and the risk level, the event strategy mapping rule associated with the linkage priority of the security and protection device is combined to generate a linkage strategy including the target device type, linkage timing and control parameter, and the strategy parameter is dynamically adjusted according to the security risk level.

7. The method of claim 6, wherein, The Bayesian probability graph integrating the personnel behavior recognition model and the environmental security risk assessment model includes: The multi-level node structure is defined, and the core node layer is divided into an input layer, a correlation layer and an output layer: the input layer includes personnel behavior characteristic nodes and environment parameter nodes, the personnel behavior characteristic nodes are obtained based on normal behavior labels, abnormal behavior labels, behavior duration and behavior area output by a personnel behavior recognition model, and the environment parameter nodes are obtained based on real-time monitoring values such as temperature, smoke concentration and gas concentration of an environment safety risk assessment model; the correlation layer includes space-time correlation nodes and historical correlation nodes; and the output layer includes event type nodes and risk level nodes corresponding to four-level safety risk intervals. The joint probability distribution of the personnel behavior characteristic nodes and the environment parameter nodes is obtained by calculating the corresponding conditional probability through co-occurrence samples of abnormal behaviors and environmental abnormalities in historical data; and the probability parameters of the space-time correlation nodes are calibrated in combination with the space-time anchoring relationship of the security space data set; The real-time output of the personnel behavior recognition model is taken as evidence to input the personnel behavior characteristic nodes, and the marginal probability thereof is dynamically updated; and the threshold judgment result of the environment safety risk assessment model is taken as evidence to input the environment parameter nodes, and the conditional probability of the correlation layer nodes is updated through an evidence propagation algorithm; When new event handling data is added to the historical learning library, if the correlation mode from behavior to environment event in the new event handling data deviates from the existing probability table by more than a preset threshold, the conditional probability table is automatically triggered for iterative updating, the probability parameters of the correlation layer nodes are corrected by using a maximum likelihood estimation method, and through sensitivity analysis, the input nodes that have a significant impact on the output layer nodes are identified, the weight proportion of the input nodes in the probability reasoning is increased, and the recognition accuracy of the Bayesian probability graph for complex coupled events is improved.

8. The method of claim 1, wherein, The corresponding control instructions and security control parameters are sent to the corresponding target security devices based on the linkage strategy, which includes: Based on the target device type in the linkage strategy and the priority of the control array, a structured control instruction is generated, the instruction contains a device unique identifier, an operation type, a precise control parameter and a linkage time sequence requirement, and is associated with a corresponding space-time coordinate of the security space data set to ensure that the instruction matches the event occurrence area; A pre-play verification mechanism based on an apartment space digital twin mapping model is used to simulate the execution effect of the control instruction before the instruction is sent, to verify whether there is a conflict in the device linkage logic, and if a conflict is found in the simulation, the linkage time sequence parameter is automatically adjusted, and the control instruction is generated again after optimization; The control instruction is sent to the target security device through a standardized encrypted transmission protocol, and device identity authentication information and instruction check codes are embedded in the transmission process to prevent the instruction from being tampered with or misexecuted; based on the node connectivity relationship of the security space topology graph, the instruction transmission path of the control instruction is optimized, and low-delay communication links are preferentially selected to ensure that the instruction response delay of the core target security device is not greater than a preset time length.

9. The method of claim 8, wherein, After the corresponding control instructions and security control parameters are sent to the corresponding target security devices based on the linkage strategy, the method further includes: Real-time collection of state data of target security equipment during execution, and feedback to the security intelligent decision engine, combining the expected effect in the linkage strategy and real-time state data, dynamically adjusting the control parameters: if the risk spreading speed exceeds the expectation, automatically increasing the spraying intensity and expanding the alarm coverage range; if the personnel evacuation is slow, increasing the flicker frequency and voice guidance frequency of the escape indication equipment; A cross-device linkage state synchronization mechanism is established, and a spatio-temporal correlation algorithm is used to associate the execution states of multiple devices in the same event area to the same spatio-temporal index of the security space data set, ensuring that the devices act in coordination; when a timing deviation is detected in the device linkage, a calibration instruction is automatically sent to correct the action timing, avoiding the decrease in emergency efficiency caused by different responses of the devices; After the execution of the instruction, the actual execution parameters, response time, and linkage effect data of the device are associated to the historical learning library, which is used to optimize the control parameter configuration and device priority weight in the subsequent linkage strategy, forming a closed-loop control mechanism including strategy generation, instruction execution, effect feedback, and parameter optimization.

10. An apartment security linkage system based on artificial intelligence, characterized by, The system comprises: A collection module for real-time acquisition of initial security data collected by various monitoring devices in the apartment, wherein the initial security data includes video image data, door and window state data, environmental safety data, and personnel access data; A construction module for forming a security space data set matching the apartment space according to the spatio-temporal correlation of the real-time acquired initial security data; A decision module for using a security intelligent decision engine to jointly analyze the security space data set to determine the type of security event and the safety risk level in the apartment, and generate a linkage strategy according to the determination result; wherein the security intelligent decision engine includes a personnel behavior recognition model and an environmental safety risk assessment model; the personnel behavior recognition model analyzes the personnel activity features in the security space data set in real time through a deep learning algorithm to distinguish between normal and abnormal behaviors; the environmental safety risk assessment model combines the risk assessment threshold dynamically optimized based on historical security risk data to quantitatively analyze the diffusion trend and influence range of personnel abnormal behavior and environmental abnormal phenomenon in the current environment, so as to determine the safety risk level of the current environment; An execution module for sending corresponding control instructions and security control parameters to corresponding target security equipment based on the linkage strategy; setting the linkage priority of the security device according to the mapping relationship between the security event type, the security device, and the emergency degree of the security event, and constructing a control array including target security devices at all levels according to the linkage priority based on the current security event type of the apartment, and triggering the core target security device to respond to the instruction of the linkage strategy through the control array, realizing remote control and linkage response of the target security device.

Images