An AI artificial intelligence machine learning system and intelligent gateway

Abstract

The present invention provides an AI machine learning system and intelligent gateway, which relate to the field of AI control technology. The system includes: a communication module that establishes a bidirectional connection with air conditioning units, security sensor arrays, and intelligent lighting networks within a building through multiple Internet of Things protocols, and collects real-time equipment operation data streams, including air conditioning compressor current waveforms, security alarm codes, and lighting equipment light intensity time series data; an edge computing module that extracts air conditioning energy consumption characteristic values, security alarm signals, and environmental temperature and humidity data sets based on the equipment operation data streams, and stores the processed structured data sets in an artificial memory area through an artificial classification algorithm. The present invention realizes real-time data collection and processing of multi-source equipment through multi-module collaboration, analyzes personnel and environmental information, intelligently generates control instructions, and adapts to various types of equipment through protocol conversion, thereby improving the comfort, safety, and energy efficiency of the building environment, and enhancing system compatibility and emergency response capabilities.

Description

Technical Field

[0001] The present invention relates to the field of AI artificial intelligence control technology, and in particular to an AI artificial intelligence machine learning system and an intelligent gateway. Background Art

[0002] Traditional IoT gateways face multiple bottlenecks in building automation scenarios. Traditional architectures employ a time-sharing, multi-tasking model and rely on a centralized data center for data collection and storage. This results in delays between data collection from building equipment (such as air conditioning units and security sensors) and the issuance of control commands.

[0003] For example, the air-conditioning system needs to dynamically adjust the temperature according to the density of people, but the traditional link delay causes the temperature response to lag, affecting comfort and energy efficiency optimization; after the smoke alarm is triggered, the security system cannot meet the building safety regulations due to the delay in cloud-based middle-end processing.

[0004] In addition, heterogeneous data such as air conditioning current waveforms and security alarm codes cannot be integrated and analyzed in real time. Digital twins rely on full data transmission, exacerbating network congestion within the building; device-level features (such as air conditioning energy efficiency ratio and personnel thermal distribution) cannot be dynamically extracted, resulting in a disconnect between control instructions and building environmental status; sensitive data (such as video surveillance streams) need to be uploaded to the cloud for processing, increasing the risk of privacy leakage, and high bandwidth usage restricts energy efficiency optimization. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an AI artificial intelligence machine learning system and an intelligent gateway, which realizes protocol adaptive conversion, efficient data compression and real-time decision-making at the edge.

[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0007] In a first aspect, an AI artificial intelligence machine learning system includes:

[0008] The communication module establishes a two-way connection with the building's air conditioning units, security sensor arrays, and smart lighting networks through multiple IoT protocols, collecting real-time data streams on equipment operation, including air conditioning compressor current waveforms, security alarm codes, and lighting equipment light intensity timing data.

[0009] The edge computing module is used to extract air conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets based on the device operation data stream, and store the processed structured data sets in the pseudo-physical memory area through the pseudo-physical classification algorithm;

[0010] The virtual memory area module is used to implement timestamp difference encoding on the historical power data of air conditioners in the structured data set to form a spatiotemporal correlation data cube;

[0011] The engine module is used to identify the thermal map pattern of people moving in the building based on the spatiotemporal correlation data cube in the simulacrum memory area and generate a dynamic temperature control parameter matrix;

[0012] The control logic module generates control instructions based on environmental comfort indicators, the distribution of crowded areas, abnormal behavior pattern recognition, and security threat levels. When an alarm signal is received, it dynamically adjusts the video surveillance focus area and access control strategy based on the signal type through a three-level priority response mechanism.

[0013] The protocol conversion module is used to dynamically convert control instructions into the target device protocol format.

[0014] Furthermore, based on the equipment operation data stream, the air conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets are extracted, and the processed structured data sets are stored in the simulacrum memory area through the simulacrum classification algorithm, including:

[0015] Perform sliding window processing on the original device data stream to split the continuous data stream into multiple data segments of equal length;

[0016] Extract air conditioner energy efficiency feature sets from multiple equal-length data segments, including real-time energy efficiency ratio, current harmonic distortion rate, and load fluctuation index. Generate an alarm frequency density matrix based on security sensor data to obtain a feature dataset containing air conditioner energy efficiency features and the spatiotemporal distribution of security events.

[0017] The feature data set is generated into device-level classification labels through the device fingerprint encoding algorithm, and the classification data is mapped to the simulacrum memory area according to the device entity ID.

[0018] Furthermore, timestamp difference encoding is implemented on the historical power data of air conditioners in the structured dataset to form a spatiotemporal correlation data cube, including:

[0019] Extract historical air conditioner power data from the structured data set, group it by device ID and arrange it by timestamp to form a time series power sequence;

[0020] For each adjacent timestamp pair, the power difference and the time interval are calculated to form a power difference encoding sequence;

[0021] The difference code sequence is associated with the device geographic coordinates to construct a three-dimensional data cube containing time dimension, space dimension and feature dimension.

[0022] Furthermore, based on the spatiotemporal correlation data cube in the simulated memory area, the thermal map pattern of people moving in the building is identified and a dynamic temperature control parameter matrix is generated, including:

[0023] Extract multi-source probe positioning data from the spatiotemporal correlation data cube in the simulated memory area, and process the probe positioning data using a long short-term memory network to predict the probability of personnel staying in each area in the future and generate a dynamic heat distribution map.

[0024] The thermal distribution map is spatially overlaid with the temperature and humidity gradient data stored in the simulated memory area to identify densely populated areas and areas with abnormal environmental parameters, and obtain coupled analysis results.

[0025] According to the coupling analysis results and the thermal response characteristics of the air-conditioning units, the regional temperature setpoint matrix is dynamically calculated.

[0026] Furthermore, the probe positioning data is extracted from the spatiotemporal correlation data cube in the simulated memory area and processed using a long short-term memory network to predict the probability of personnel staying in each area in the future and generate a dynamic heat distribution map, including:

[0027] Extract multi-source probe positioning data from the spatiotemporal correlation data cube of the simulacrum memory area, including probe connection events, Bluetooth beacon scanning records, and video analysis coordinate point clouds;

[0028] The multi-source probe positioning data are grouped by person ID and the spatiotemporal sequence of individual movement trajectories is constructed, which is then segmented into continuous trajectory segments using a sliding window.

[0029] The trajectory segments are input into the pre-trained long short-term memory network, and the probability distribution of personnel residence in each area within the future target period is predicted based on the historical movement patterns.

[0030] The residence probability distribution is spatially interpolated and optimized, and a dynamic thermal distribution map is generated based on the building topology.

[0031] Furthermore, based on the coupling analysis results and the thermal response characteristics of the air conditioning units, the regional temperature setpoint matrix is dynamically calculated, including:

[0032] Based on the historical temperature data in the building, the dynamic baseline temperature is determined, and the probability of occupancy in each area during the future target period is compared with the actual probability of occupancy at the current time point to obtain the occupancy density change rate.

[0033] Adaptive learning algorithms are used to optimize the occupant heat load adjustment factor based on the thermal response characteristics of the air conditioning equipment, including response speed and adjustment range, and historical energy efficiency data, including energy consumption and cooling capacity;

[0034] Real-time collection of outdoor environmental humidity monitoring values and meteorological forecast data, combined with the thermal inertia parameters of the air-conditioning unit, including heat capacity and heat conduction values, to generate external disturbance compensation factors;

[0035] The reference temperature, personnel heat load adjustment factor and external disturbance compensation factor are multi-dimensionally integrated to generate the temperature control parameter matrix of the spatial partition.

[0036] Furthermore, control instructions are generated based on environmental comfort indicators, the distribution of densely populated areas, abnormal behavior pattern recognition, and security threat levels. When an alarm signal is received, the video surveillance focus area and access control strategy are dynamically adjusted based on the signal type through a three-level priority response mechanism, including:

[0037] Generate a comprehensive threat assessment score by performing spatiotemporal alignment and confidence-weighted fusion of ambient temperature and humidity comfort indicators, heat maps of densely populated areas, abnormal behavior characteristics of video analysis, and security sensor alarm signals.

[0038] Based on the comprehensive threat assessment score, a three-level response mechanism is triggered. Based on the triggered response level, a corresponding hierarchical response strategy is generated. The hierarchical response strategy includes a set of video surveillance focus area coordinates, access control lock logic rules, and air conditioning volume adjustment parameters.

[0039] The hierarchical response strategy is converted into a control instruction set in the target device protocol format, and the control instruction set is sent to the corresponding device through the corresponding communication protocol to perform the corresponding operations, including adjusting the air volume of the air conditioner, adjusting the viewing angle of the surveillance camera, and controlling the electromagnetic lock of the access control.

[0040] Furthermore, the three-level response mechanism includes:

[0041] Level 1 response: When the comprehensive threat assessment score is greater than 90, the target area access lock and laser tracking video focus are activated;

[0042] Level 2 response: When 70 < comprehensive threat assessment score ≤ 90, strobe lighting and directional acoustic alarms are activated in the evacuation passage;

[0043] Level 3 response: When the comprehensive threat assessment score is ≤70, non-core production equipment, including cooling water circulation pump units and ventilation auxiliary units, will switch to energy-saving mode.

[0044] In a second aspect, an intelligent gateway includes:

[0045] one or more processors;

[0046] The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the system.

[0047] According to a third aspect, a computer-readable storage medium stores a program, which implements the system when executed by a processor.

[0048] The above solution of the present invention includes at least the following beneficial effects:

[0049] The communication module supports multiple IoT protocols and can quickly and stably establish two-way connections with various devices in the building. This allows the system to collect multi-source device operating data streams in real time, such as air-conditioning compressor current waveforms, security alarm codes, and lighting equipment light intensity timing data, avoiding data collection obstacles caused by protocol incompatibility and ensuring the timeliness and integrity of the data. The edge computing module uses a pseudo-physical classification algorithm to deeply process the collected data, accurately extracting key information such as air-conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets, and converting them into structured data for storage in the pseudo-physical memory area. At the same time, the pseudo-physical memory area implements timestamp difference encoding on the historical power data of the air-conditioning to form a spatiotemporal correlation data cube, which not only improves the efficiency of data storage, but also facilitates data retrieval and analysis.

[0050] Based on the spatiotemporal correlation data cube of the simulacrum's memory area, the engine module accurately identifies thermal patterns of occupant movement within the building and generates a dynamic temperature control parameter matrix. This enables the air conditioning system to intelligently adjust the temperature in each zone based on occupant distribution and activity patterns, improving occupant comfort while effectively reducing energy consumption. For example, it automatically increases cooling in crowded areas and appropriately raises the temperature in less crowded areas. The control logic module generates control commands based on environmental comfort indicators, the distribution of crowded areas, identification of abnormal behavior patterns, and security threat levels. It then implements a three-level priority response mechanism to address varying security threats. Upon receiving an alarm signal, it dynamically adjusts the video surveillance focus area and access control strategy based on the signal type. For example, the first-level response promptly locks access to dangerous areas and provides precise monitoring. The second-level response uses strobe lighting and directional sound waves to guide evacuation, enhancing building security and emergency response efficiency, ensuring the safety of people and property.

[0051] The protocol conversion module dynamically converts control commands into the target device's protocol format, ensuring that system-generated commands are compatible with the diverse brands and models of equipment within the building. Whether it's air conditioning units, security equipment, or smart lighting, commands can be accurately received and executed, preventing control failures caused by device protocol differences. This improves system versatility and compatibility, and reduces the cost and difficulty of device replacement and upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of an AI artificial intelligence machine learning system provided by an embodiment of the present invention.

[0053] Figure 2This is a flow chart of an AI artificial intelligence machine learning system provided by an embodiment of the present invention, which implements timestamp difference encoding on historical air-conditioning power data in a structured data set to form a spatiotemporal correlation data cube. DETAILED DESCRIPTION

[0054] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0055] like Figure 1 As shown, an embodiment of the present invention provides an AI artificial intelligence machine learning system, including:

[0056] Communication Module 1 establishes a two-way connection with the building's air conditioning units, security sensor arrays, and smart lighting networks through multiple IoT protocols, collecting real-time data streams on equipment operation, including air conditioning compressor current waveforms, security alarm codes, and lighting equipment light intensity time series data.

[0057] Edge computing module 2 is used to extract air conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets based on the device operation data stream, and store the processed structured data sets in the pseudo-physical memory area through the pseudo-physical classification algorithm;

[0058] The virtual memory area module 3 is used to implement timestamp difference encoding on the historical power data of the air conditioner in the structured data set to form a spatiotemporal correlation data cube;

[0059] Engine module 4, used to identify the thermal map pattern of people moving in the building based on the spatiotemporal correlation data cube in the simulated memory area and generate a dynamic temperature control parameter matrix;

[0060] Control logic module 5 is used to generate control instructions based on environmental comfort indicators, distribution of crowded areas, abnormal behavior pattern recognition, and security threat levels. When an alarm signal is received, it dynamically adjusts the video surveillance focus area and access control strategy based on the signal type through a three-level priority response mechanism;

[0061] The protocol conversion module 6 is used to dynamically convert the control instructions into the target device protocol format.

[0062] In this embodiment of the present invention, localized data processing in the edge computing module and spatiotemporal data aggregation in the emulated memory area enable end-to-end closed-loop control of building equipment data, from acquisition to decision-making. This reduces data link latency in traditional architectures and ensures real-time responsiveness for air conditioning temperature control and security linkage. Based on emulated classification algorithms and device fingerprinting technology, multi-source heterogeneous data, such as air conditioning current waveforms and security alarm codes, is stored by device entity, breaking through the inefficiency of traditional "point-based" storage and improving data read and write efficiency and multi-device collaborative analysis capabilities.

[0063] The AI / ML engine uses occupant heat map pattern recognition and combines it with ambient temperature and humidity data to dynamically generate a regional temperature control parameter matrix. This allows for differentiated temperature control strategies for densely populated and low-load areas within the building, balancing energy efficiency optimization with comfort requirements. The protocol conversion module supports dynamic adaptation of multiple protocols (such as Modbus, Zigbee, and DALI), eliminating manual configuration costs. The control logic module, based on multi-dimensional data integration such as environmental comfort and security threat levels, triggers a three-level priority response mechanism, enabling intelligent linkage between security, lighting, and air conditioning systems.

[0064] Sensitive data (such as security alarm signals) undergoes feature extraction and compression processing at the edge. Using a spatiotemporal correlation data cube, only structured analysis results are uploaded to the cloud, mitigating privacy risks and reducing network bandwidth usage. The spatiotemporal correlation data cube and quasi-physical memory area design support rapid backtracking and predictive analysis of historical building equipment data, providing a lightweight data foundation for digital twins while reducing software and hardware costs for system expansion.

[0065] In a preferred embodiment of the present invention, extracting air conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets, and storing the processed structured data sets in a pseudo-physical memory area using a pseudo-physical classification algorithm may include:

[0066] Perform sliding window processing on the original device data stream to split the continuous data stream into multiple data segments of equal length;

[0067] Extract air conditioner energy efficiency feature sets from multiple equal-length data segments, including real-time energy efficiency ratio, current harmonic distortion rate, and load fluctuation index. Generate an alarm frequency density matrix based on security sensor data to obtain a feature dataset containing air conditioner energy efficiency features and the spatiotemporal distribution of security events.

[0068] The feature data set is generated into device-level classification labels through the device fingerprint encoding algorithm, and the classification data is mapped to the simulacrum memory area according to the device entity ID.

[0069] In this embodiment of the present invention, raw data streams generated by air conditioning, security, and lighting equipment (such as the continuous current waveform of an air conditioning compressor, the alarm code stream of a security sensor, and the time-series light intensity data of a lighting device) are dynamically segmented. The data streams are segmented using fixed time windows (e.g., 5-second windows) or adaptive event windows (e.g., every 1000 data points), ensuring that each data segment encompasses a complete event cycle (e.g., a single start / stop cycle of an air conditioning compressor). Each data segment is appended with a device ID, timestamp, and spatial tag (e.g., floor location) to ensure spatiotemporal alignment of data from multiple devices. For example, an air conditioning current segment is strictly synchronized with the temperature and humidity data for the area in which it resides.

[0070] Calculate the EER value (cooling capacity / power) based on the instantaneous power (current × voltage) of the current segment and the preset cooling capacity mapping table , reflecting the current energy efficiency status; among them, is the cooling capacity, obtained from a preset mapping table and associated with current, voltage, and device model; is the current; is the voltage. Perform Fast Fourier Transform (FFT) on the current waveform to extract the fundamental wave and each harmonic component and calculate the total harmonic distortion rate. , evaluates the current quality; where, is the highest harmonic order; is the harmonic order; It is The effective value of subharmonic voltage; This is the effective value (RMS) of the fundamental voltage, i.e., the voltage amplitude of the main frequency (power frequency, such as 50Hz / 60Hz) component in the signal. The peak-to-valley current difference within the statistical window, combined with the standard deviation of historical load data, quantifies the operating stability of the compressor. The alarm event frequency is counted by spatial region (such as room, corridor) and time window (such as every 10 minutes), constructing a two-dimensional matrix. For example, the matrix rows represent regions, the columns represent time windows, and the element value is the number of alarms in that region within the corresponding time window.

[0071] Static device parameters (such as model and installation location) are hashed and combined with dynamic features (such as mean THD and alarm frequency patterns) to generate a unique device fingerprint tag using a lightweight classification model (such as a decision tree). Feature datasets (such as air conditioner energy efficiency characteristics and security alarm matrices) are categorized and stored in independent data units based on device entity ID. For example, the data unit for air conditioner unit ID-A contains historical EER sequences, THD trends, and associated environmental parameters, forming a device-centric semantic storage structure.

[0072] For example, in the optimization of the air conditioning and security systems of a commercial building, the air conditioning unit generates a current waveform data stream every 5 seconds, and the sliding window divides it into continuous segments, each of which contains 20 current cycle data.

[0073] Security sensors report alarm codes (such as smoke alarms and intrusion detection) in real time, and the number of alarms on each floor is counted in 10-minute windows. The EER value calculated from current fragments reveals that the EER value of the air conditioner on floor 3 remains consistently below the threshold, triggering an energy efficiency alarm. The security alarm matrix reveals an abnormally high alarm frequency in Area B during peak hours, and combined with historical data, it is identified as a high-incidence area for false alarms. The THD signature of air conditioner 3 is compared with historical data to generate a unique fingerprint, which is mapped to an independent unit in the virtual memory area and associated with its installation location and maintenance records. The security alarm matrix is linked to air conditioner operating data by timestamp, forming a multi-dimensional analysis view of "area, time, and device."

[0074] Through sliding windows and edge feature extraction, air conditioner energy efficiency anomalies (such as sudden drops in EER) can be detected and responded to within seconds, avoiding the minute-long delays associated with traditional architectures. Device fingerprinting maps discrete current waveforms and alarm codes into device-centric semantic datasets, supporting cross-device correlation analysis (e.g., air conditioners and security systems). Based on a coupled analysis of the security alarm matrix and air conditioner energy efficiency characteristics, the system automatically switches the air conditioner to low-power mode in high-security areas, balancing energy efficiency and emergency response requirements. The spatiotemporal correlation data cube compresses historical data through differential encoding, reducing storage redundancy. Device-level classification tags simplify data query logic and reduce database load. Device fingerprinting is compatible with multi-protocol devices (e.g., Modbus air conditioners and Zigbee sensors), eliminating the need for manual rule configuration and adapting to complex building equipment environments.

[0075] In a preferred embodiment of the present invention, timestamp difference encoding is performed on the historical power data of the air conditioner in the structured data set to form a spatiotemporal correlation data cube, which may include:

[0076] Extract historical air conditioner power data from the structured data set, group it by device ID and arrange it by timestamp to form a time series power sequence;

[0077] For each adjacent timestamp pair, the power difference and the time interval are calculated to form a power difference encoding sequence;

[0078] The difference code sequence is associated with the device geographic coordinates to construct a three-dimensional data cube containing time dimension, space dimension and feature dimension.

[0079] In this embodiment of the present invention, historical power data of air conditioners is extracted from a structured database. The data fields include device ID, timestamp, power value, and device geographic coordinates (such as floor, room number). The data is grouped by device ID, and each group of data is sorted in ascending order by timestamp to form a time series power sequence for each device. For example, the time series sequence of device ID-A is: , ),( , ),…,( , ). For each device's timing power sequence, calculate the power difference between adjacent timestamp pairs ( ) and time interval ( ).

[0080] Power difference = ( is the time point index);

[0081] Time interval = (Unit: seconds).

[0082] Generate difference code sequence: ( , ),( , ),…,( , Only the initial power value is stored ( ) and difference sequences, replacing the original full data to achieve data compression. The difference coding sequence is associated with the device's geographic coordinates (such as latitude and longitude, floor coordinates) to form a data unit with a spatial tag.

[0083] Time dimension: Divide by time window (such as hours, days) and record the time distribution of the difference series;

[0084] Spatial dimension: Divide the area (such as room, floor) by device coordinates and aggregate the difference data of devices in the same area;

[0085] Feature dimension: including power difference ( ), time interval ( ), equipment type (such as variable frequency air conditioner, fixed frequency air conditioner).

[0086] Columnar storage (such as Parquet format) is used to support fast queries by time, space, or device ID range.

[0087] Assume that the energy consumption analysis of the air conditioner in an office building is as follows: the original power sequence of device ID-A (the air conditioner in the conference room on the 3rd floor) is (09:00, 1500W), (09:15, 1800W), (09:30, 1650W)... The original power sequence of device ID-B (the air conditioner in the corridor on the 5th floor) is (09:00, 800W), (09:15, 750W), (09:30, 820W)... The difference sequence of device ID-A is ( =300W, =900s),( =-150W, =900s).... The device ID-B difference sequence is ( =-50W, =900s),( =70W, =900s)....

[0088] Time dimension: divided by hour, storing hourly difference series;

[0089] Spatial dimensions: 3rd floor meeting room (ID-A), 5th floor corridor (ID-B);

[0090] Feature dimension: power change trend (such as "rapid rise and fall" for ID-A and "smooth fluctuation" for ID-B).

[0091] By replacing full storage with differential coding, only the initial value and the change are retained, which significantly reduces the redundancy of historical data storage. For example, the original 1,000 power records can be compressed into 1 initial value + 999 difference values. The data cube supports multi-dimensional joint queries, for example: spatial dimension: statistics on the power fluctuation pattern of all air conditioners on a certain floor; time dimension: analysis of energy efficiency change trends during peak hours (such as 12:00-14:00); feature dimension: identification of abnormal fluctuations of specific equipment types (such as variable frequency air conditioners). Differential coding allows for rapid restoration of power values at any point in time, meeting the needs of real-time monitoring and historical data backtracking. Through the distribution of power difference in the space-time cube, abnormal areas (such as a certain floor) can be quickly located. Consistently negative values), combined with device coordinates, determine the source of the fault. Spatiotemporal correlation analysis reveals regional energy consumption patterns (for example, air conditioning in a conference room continues to run at high power even when no one is present), providing a data foundation for automated scheduling strategies (such as adjusting temperature based on visitor flow).

[0092] In a preferred embodiment of the present invention, based on the spatiotemporal correlation data cube in the simulacrum memory area, identifying the thermal map pattern of personnel movement in the building and generating a dynamic temperature control parameter matrix may include:

[0093] Extract multi-source probe positioning data from the spatiotemporal correlation data cube in the simulated memory area, and process the probe positioning data using a long short-term memory network to predict the probability of personnel staying in each area in the future and generate a dynamic heat distribution map.

[0094] The thermal distribution map is spatially overlaid with the temperature and humidity gradient data stored in the simulated memory area to identify densely populated areas and areas with abnormal environmental parameters, and obtain coupled analysis results.

[0095] According to the coupling analysis results and the thermal response characteristics of the air-conditioning units, the regional temperature setpoint matrix is dynamically calculated.

[0096] In this embodiment of the present invention, Wi-Fi probes (device connection time, signal strength), Bluetooth beacons (device scanning records), and video coordinate point clouds (personnel real-time location) are extracted from a spatiotemporal correlation data cube. Noise data (such as signal drift points) is aligned by timestamp and cleaned. Multi-source data is linked using device IDs or anonymous identifiers (such as hashed MAC addresses) to generate time-stamped individual movement trajectories (e.g., a user's path from the lobby to the conference room and the duration of their stay). The trajectories are segmented into segments using a sliding window (e.g., 15-minute intervals), and features (such as dwell point density and rate of change in movement direction) are extracted. These features are then fed into a pre-trained LSTM model to predict the probability of a person staying in each area in the future (e.g., an 85% probability of staying in the conference room within the next hour).

[0097] The predicted probabilities were mapped onto a 1m×1m building grid. Kriging interpolation was used to fill in blank areas. The distribution was corrected by combining topological structures such as walls and access control points to generate a color gradient heat map (red indicates high density, blue indicates low density). Real-time temperature and humidity gradient data stored in the simulated memory area (sampled every 5 minutes) was extracted and aligned with the heat map at the same grid granularity to form a multidimensional data layer (heat, temperature, humidity, and air conditioning status). Using Geographic Information System (GIS) technology, the dynamic heat distribution map and temperature and humidity gradient data were spatially overlaid. During the overlay process, the occupancy probability of each area in the heat map was correlated with the temperature and humidity data at the corresponding location. By setting a occupancy probability threshold (e.g., 0.7) and a normal temperature and humidity range (temperature 22-26°C, humidity 40-60%), densely populated areas (areas with occupancy probability exceeding the threshold) and areas with abnormal environmental parameters (areas with temperature or humidity outside the normal range) were identified.

[0098] For each area within the building (e.g., offices, conference rooms, corridors, etc. on different floors), the appropriate temperature setpoint is calculated using the formula (dynamic base temperature × occupant heat load adjustment factor × external disturbance compensation factor). The temperature setpoints for all areas are organized into a matrix, forming a regional temperature setpoint matrix, which provides specific parameters for precise temperature control of the air conditioning system.

[0099] Suppose there is a large shopping mall with three floors, each floor has multiple stores and public areas.

[0100] On a weekend afternoon, Wi-Fi probes collected a large amount of customer connection information, Bluetooth beacons also scanned the movement paths of many customers, and the video surveillance system recorded the coordinate point clouds of customers in various areas. After cleaning and integrating this data, it was grouped by customer ID to construct a spatiotemporal sequence of individual movement paths. For example, between 2:00 and 3:00 PM, customer A entered from the first-floor entrance, passed through several clothing stores, and then went to the restaurant on the second floor. A sliding window was used to segment this spatiotemporal sequence into segments, such as the movement path from 2:00 PM to 2:15 PM. These segments were input into an LSTM network, which predicted that within the next hour, the probability of occupancy would be higher in the cinema on the third floor and the dessert area on the first floor, while the probability of occupancy in some shops on the second floor would be lower.

[0101] The prediction results were spatially interpolated and optimized, and a dynamic thermal distribution map was generated based on the mall's floor plan. In this map, the cinema and dessert area are colored red, indicating dense crowds; some shops are colored blue, indicating sparse crowds. Temperature and humidity gradient data within the mall were obtained from the simulated memory area. The cinema's high density resulted in a 3°C higher temperature and increased humidity than the surrounding area. The dessert area also experienced slightly higher temperatures and humidity due to frequent traffic. Spatial overlay analysis identified the cinema and dessert areas as areas with high crowd density and abnormal environmental parameters. Coupling analysis results indicated that these areas required temperature and humidity adjustments. Based on the areas of high crowd density and abnormal environmental parameters identified by the coupling analysis, combined with the mall's air conditioning units' thermal response characteristics, such as cooling and heating capacity and response speed, and considering the changing trends in occupancy density across different zones, the appropriate temperature setpoints for each zone were dynamically calculated. This ultimately generated a regional temperature setpoint matrix, providing a basis for precise control of the mall's air conditioning system to meet the comfort needs of occupants in different zones and achieve energy savings.

[0102] By accurately predicting the probability of occupancy, temperature control parameters can be dynamically adjusted based on the density of different areas. In sparsely populated areas, the temperature setpoint can be appropriately raised or lowered to reduce air conditioning energy consumption. In densely populated areas, a suitable temperature environment is maintained to avoid energy waste, thereby improving energy efficiency for the entire building. By combining the results of coupled analysis of densely populated areas with areas with abnormal environmental parameters, temperature and humidity can be adjusted more precisely. Timely adjustments to temperature and humidity in densely populated areas provide a comfortable environment, reduce health issues and discomfort caused by environmental discomfort, and improve employee satisfaction. Utilizing an LSTM network and adaptive learning algorithm, the system learns and adjusts based on historical and real-time data, automatically adapting to varying occupant activity patterns and environmental changes. For example, the temperature control strategy can be automatically adjusted based on the different distribution and activity patterns of occupants on weekdays and weekends, enhancing the system's intelligence and adaptability.

[0103] In another preferred embodiment of the present invention, the probe positioning data is extracted from the spatiotemporal correlation data cube in the simulated memory area, and the probe positioning data is processed using a long short-term memory network to predict the probability of personnel staying in each area in the future period of time and generate a dynamic heat distribution map, which may include:

[0104] Extract multi-source probe positioning data from the spatiotemporal correlation data cube of the simulacrum memory area, including probe connection events, Bluetooth beacon scanning records, and video analysis coordinate point clouds;

[0105] The multi-source probe positioning data are grouped by person ID and the spatiotemporal sequence of individual movement trajectories is constructed, which is then segmented into continuous trajectory segments using a sliding window.

[0106] The trajectory segments are input into the pre-trained long short-term memory network, and the probability distribution of personnel residence in each area within the future target period is predicted based on the historical movement patterns.

[0107] The residence probability distribution is spatially interpolated and optimized, and a dynamic thermal distribution map is generated based on the building topology.

[0108] In an embodiment of the present invention, the simulated memory area is an area for storing processed data, and the spatiotemporal correlation data cube integrates multi-dimensional information such as time, space, and device. The probe connection event records the time and related information when the device (such as a mobile phone) establishes a connection with the probe, which can be used to determine the presence of a person within the probe coverage area; the Bluetooth beacon scan record contains Bluetooth signal strength, scan time, etc., which can roughly determine the relative position of the person and the Bluetooth beacon; the video analysis coordinate point cloud obtains the precise coordinate information of the person in the scene by performing image analysis on the surveillance video. The system uses preset data extraction rules and interfaces to filter out these multi-source probe positioning data within a specific time period from the spatiotemporal correlation data cube. For example, set the extraction of relevant data from 9 to 11 a.m. on a certain weekday, and accurately obtain the required probe connection events, Bluetooth beacon scan records, and video analysis coordinate point cloud data according to the time range and data type identifier.

[0109] After acquiring multi-source probe positioning data, the data is grouped according to the person ID (e.g., a unique device identifier) contained in the data. Each person ID corresponds to a set of data, which is then arranged in chronological order to construct a spatiotemporal sequence of the individual's movement trajectory. This sequence details the location information of each person at different points in time, reflecting their movement trajectory. The spatiotemporal sequence is then segmented using a sliding window technique. The sliding window has a fixed time length (e.g., 10 minutes) and a sliding step size (e.g., 1 minute). Starting from the start time of the sequence, the sliding window moves one step at a time, capturing data within a fixed time period to form continuous trajectory segments. For example, if a person's spatiotemporal sequence is segmented using a 10-minute window length and a 1-minute step size, the first segment might contain movement trajectory data from 9:00 to 9:10, the second segment from 9:01 to 9:11, and so on.

[0110] A long short-term memory (LSTM) network is a neural network specifically designed for processing time series data. It can learn long-term dependencies and complex patterns in data. Before use, the LSTM network must be trained with a large amount of historical trajectory data. During training, the network learns the movement patterns and patterns of people in different time periods and areas. Once training is complete, the segmented trajectory segments are input into the pre-trained LSTM network. Based on the learned historical movement patterns, the network analyzes and predicts the input trajectory segments, outputting the probability distribution of people in each area over a target future time period (e.g., the next 30 minutes). For example, the network might predict that within the next 30 minutes, the probability of people being present in the elevator area on a certain floor is 0.6, and the probability of people being present in the restaurant area is 0.3.

[0111] The predicted occupancy probability distribution represents discrete data points. To present a more intuitive and continuous representation of occupancy distribution, spatial interpolation optimization is required. Spatial interpolation algorithms (such as inverse distance weighted interpolation and kriging) use known discrete probability data points to estimate the probability values for other locations across the entire spatial range, making the probability distribution smoother and more continuous. Furthermore, building topology information (such as floor layout, room location and size, and corridor orientation) is incorporated to map the probability values to actual spatial locations. Different colors (such as red for high probability and blue for low probability) and color depths are used to represent the occupancy probability of different areas, generating a dynamic heat map. This heat map clearly illustrates the density and distribution trends of occupancy in different areas of the building over a specific time period.

[0112] Consider a large office building with five floors, each floor containing multiple offices, meeting rooms, and common areas.

[0113] From 10:00 AM to 12:00 AM on a weekday, data was extracted from the spatiotemporal correlation data cube in the simulated memory area. Probe connection events captured information about the connections between numerous employee devices and probes distributed across various floors at different times. Bluetooth beacon scan records revealed employee movements near the beacons, and video analysis provided a precise record of employee locations within the monitoring area. For example, at 10:15 AM, employee A's phone established a connection with a probe located near the elevator entrance on the third floor. The Bluetooth beacon scan also detected the employee's presence in the area, and the video analysis provided a precise coordinate point cloud. This data was grouped according to the employee's work ID (serving as a person ID) to construct a spatiotemporal sequence of each employee's individual movement trajectory. For example, from 10:00 AM to 10:30 AM, employee A's trajectory showed him starting from his office on the third floor, moving through the corridor, and arriving at the meeting room. This spatiotemporal sequence was segmented using a sliding window with a window length of 10 minutes and a step size of 1 minute. The first segment is the trajectory from 10:00 to 10:10, the second segment is the trajectory from 10:01 to 10:11, and so on.

[0114] These segmented trajectory segments are fed into a pre-trained LSTM network. This network has been trained on a large amount of historical employee movement data from the office building, learning employee movement patterns. Network analysis predicts that within the next 30 minutes, the probability of a person being present in the pantry on the 4th floor is 0.7, as this period is typically employee break time and a high number of people visit the pantry. The probability of a person being present in a small conference room on the 5th floor is 0.2, as there are no meetings scheduled that day. Spatial interpolation is performed on the predicted probability distribution to smooth the distribution. Based on the office building's floor plan, the probability values are mapped to the corresponding areas. On the resulting dynamic heat map, the pantry on the 4th floor appears dark red, indicating a high probability of a person being present; the small conference room on the 5th floor appears light blue, indicating a low probability of a person being present. Other areas are also displayed in varying shades of color based on their respective probability values, visually demonstrating the distribution of people within each area of the office building over the next 30 minutes.

[0115] Dynamic heat maps allow managers to clearly understand future traffic trends and density levels in different areas. For office buildings, this allows them to rationally schedule cleaning and maintenance work schedules based on the heat map, avoiding disruptions to staff during peak hours. For shopping malls, promotional activities can be planned ahead of time, targeting high-traffic areas to maximize effectiveness and commercial returns.

[0116] Based on the probability distribution of occupancy, resources can be more effectively allocated. In densely populated areas, the power or number of devices like air conditioners and lighting can be increased to ensure a comfortable environment. In less crowded areas, the power of these devices can be appropriately reduced to conserve energy. For example, in an office building, based on a heat map, air conditioning can be pre-activated and adjusted to the appropriate temperature in crowded conference rooms, while some lighting can be turned off on less crowded floors. Dynamic heat maps facilitate security management.

[0117] If an area's probability of occupancy is abnormally high or low, prompt inspection and action can be taken. For example, in an office building, if a normally sparsely populated area is suddenly predicted to be crowded, it could pose a safety hazard, allowing security personnel to conduct a preliminary inspection. Similarly, in shopping malls, areas of potential congestion or unusual crowding can be promptly identified, allowing for proactive mitigation measures to ensure safety. This provides strong support for operational decision-making. For example, shopping malls can use heat maps to analyze the attractiveness of different areas and customer flow patterns, adjusting store layouts and merchandise displays to enhance customer shopping experiences and boost purchasing desire. Office building managers can also rationally plan new office areas or adjust the use of existing space based on the distribution of personnel, improving space utilization.

[0118] In another preferred embodiment of the present invention, based on the coupling analysis results and in combination with the thermal response characteristics of the air conditioning unit, the regional temperature setting value matrix is dynamically calculated, which may include:

[0119] Based on the historical temperature data in the building, the dynamic baseline temperature is determined, and the probability of occupancy in each area during the future target period is compared with the actual probability of occupancy at the current time point to obtain the occupancy density change rate.

[0120] Adaptive learning algorithms are used to optimize the occupant heat load adjustment factor based on the thermal response characteristics of the air conditioning equipment, including response speed and adjustment range, and historical energy efficiency data, including energy consumption and cooling capacity;

[0121] Real-time collection of outdoor environmental humidity monitoring values and meteorological forecast data, combined with the thermal inertia parameters of the air-conditioning unit, including heat capacity and heat conduction values, to generate external disturbance compensation factors;

[0122] The reference temperature, personnel heat load adjustment factor and external disturbance compensation factor are multi-dimensionally integrated to generate the temperature control parameter matrix of the spatial partition.

[0123] In an embodiment of the present invention, temperature data of a building in different seasons and time periods over a long period of time (e.g., several years) is collected. This data is classified by season, and the average temperature of the same time period every day in each season (e.g., 14:00 - 15:00 every day) is calculated. For example, in summer, the average temperature of the same time period every day in multiple summers is calculated to obtain the summer benchmark temperature value; the same method is used for winter, spring, and autumn. The benchmark temperatures of different seasons obtained in this way are , which adjusts with the change of seasons, reflecting the base temperature in the building in different seasons.

[0124] Get the future target period from the previously generated dynamic heat distribution map prediction data Each area ( )’s probability of staying At the same time, the actual monitoring data at the current time is used to obtain the current time point of each region ( )’s actual personnel stay probability The area here ( ) can be a specific spatial location divided by floors, rooms, etc. Subtract the actual probability of personnel staying at the current time point from the probability of personnel staying at the future target time period, and then divide it by the time interval ,Right now , we can get the population density change rate. This change rate reflects the speed of change of the population in each area in the future.

[0125] Based on the thermal response characteristics of the air conditioning equipment (such as response speed and adjustment range) and historical energy efficiency data (energy consumption and cooling capacity), a reinforcement learning algorithm is used for optimization. In this algorithm, we set the state space to be the historical operating data of the air conditioning (including energy consumption, cooling capacity, indoor and outdoor temperature difference, etc. at different time periods), the current indoor occupancy density, and outdoor environmental parameters (such as humidity and temperature); the action space is the occupant heat load adjustment coefficient. The value range is assumed to be [0.5, 2]. The reward function is set based on comfort and energy consumption. If the indoor temperature quickly reaches the comfortable range and energy consumption is low, a higher reward is given; conversely, if the temperature adjustment effect is poor or energy consumption is too high, a lower reward is given.

[0126] After extensive iterative training, for a certain area A, assuming that its air conditioner has a fast response speed, under ideal operating conditions, the response time from receiving the temperature adjustment command to reaching the set temperature is 5 minutes, and the cooling capacity is sufficient, it can reduce the indoor temperature by 5°C per hour. Using a reinforcement learning algorithm, we analyzed and optimized the historical operating data of this area over the past year (including data on energy consumption and temperature adjustment effects under different occupancy densities), and found that under the current increase in occupancy density, =1.8. This indicates that when the population density in the area increases, in order to adjust the temperature more quickly to maintain comfort, the cooling capacity of the air conditioner needs to be adjusted more significantly, so The value is relatively large.

[0127] For the other area B, the air conditioner has a slower response speed of 15 minutes, a relatively low cooling capacity, and can only reduce the indoor temperature by 3°C per hour, and has a high energy consumption. Similarly, after the reinforcement learning algorithm optimizes and calculates its historical operating data, it finds that when the population density increases, =1.2. Due to the performance limitations of the air conditioner in this area, it cannot adjust the temperature as quickly and efficiently as in area A. In order to balance comfort and energy consumption, The value of is smaller than that of area A, and more attention will be paid to controlling energy consumption when adjusting the temperature.

[0128] By connecting with meteorological monitoring equipment or meteorological data service platform, real-time collection of outdoor environmental humidity monitoring values , and obtain future time periods in weather forecast data Predicted outdoor humidity Subtract the current outdoor humidity monitoring value from the outdoor humidity forecast value for the future period, and then divide it by the time interval. ,Right now , get the outdoor humidity change rate. This change rate reflects the trend of outdoor humidity changes in the future. Combined with the thermal inertia parameters (heat capacity, heat conductivity) of the air conditioning unit to determine the external disturbance compensation coefficient ,in, , is the thermal conductivity value; The heat capacity and thermal conductivity of the air conditioner determine the air conditioner's ability to resist the impact of outdoor environmental changes on indoor temperature. For example, if the heat capacity of the air conditioner is large and the thermal conductivity is small, it means that it can better maintain a stable indoor temperature. On the contrary, if the air-conditioning unit is more sensitive to changes in the outdoor environment, The value of will increase accordingly. By comprehensively considering these factors, the corresponding calculation model or empirical formula is used to determine value.

[0129] The dynamic reference temperature calculated above , related to the population density change rate and the part related to the external disturbance compensation factor Add up and get each area ( ) Temperature setting value Arranging the temperature setting values of all zones in a matrix according to the zone layout generates the temperature control parameter matrix of the spatial partition.

[0130] ,This matrix provides specific parameter basis for the temperature regulation of the air-conditioning system in different areas.

[0131] By adjusting the temperature setpoint based on the rate of change in occupancy density, the indoor temperature can be adjusted promptly based on actual occupancy changes. When the number of people in a particular area gradually increases, the temperature setpoint is appropriately lowered to ensure comfort in a crowded environment. When the number of people decreases, the temperature setpoint is increased, avoiding energy waste while maintaining a comfortable temperature in the area. Furthermore, adjusting the temperature setpoint based on changes in outdoor humidity comprehensively considers the impact of the outdoor environment on indoor comfort, further enhancing the overall comfort experience. The occupancy heat load adjustment factor is optimized using the thermal response characteristics of air conditioning equipment and historical energy efficiency data to avoid over-cooling or over-heating. A faster air conditioning response allows for more precise temperature adjustment in response to changes in occupancy density, reducing energy consumption. A disturbance compensation factor, generated based on the rate of change in outdoor humidity and the thermal inertia of the air conditioning unit, enables the air conditioning system to better adapt to changes in the outdoor environment, avoiding over-adjustment of indoor temperature due to fluctuations in the outdoor environment, thereby reducing energy consumption and achieving efficient energy utilization.

[0132] This calculation process integrates a variety of real-time and historical data, enabling the temperature control parameter matrix to dynamically adjust to seasonal changes, human activity, and the outdoor environment. Whether it's differences in occupancy distribution between weekdays and weekends or seasonal variations in outdoor climate, it automatically adapts and provides appropriate temperature setpoints, improving the adaptability and stability of the entire air conditioning system. Appropriate temperature regulation avoids frequent and drastic temperature adjustments on the air conditioning units, reducing equipment wear and tear. For example, when occupancy density fluctuates minimally and the outdoor environment is relatively stable, the air conditioning system doesn't need to frequently change its operating state, reducing component fatigue, thereby extending the life of the air conditioning equipment and reducing maintenance and replacement costs.

[0133] In a preferred embodiment of the present invention, control instructions are generated based on environmental comfort indicators, distribution of densely populated areas, recognition of abnormal behavior patterns, and security threat levels. When an alarm signal is received, the video surveillance focus area and access control strategy are dynamically adjusted based on the signal type through a three-level priority response mechanism, which may include:

[0134] Generate a comprehensive threat assessment score by performing spatiotemporal alignment and confidence-weighted fusion of ambient temperature and humidity comfort indicators, heat maps of densely populated areas, abnormal behavior characteristics of video analysis, and security sensor alarm signals.

[0135] Based on the comprehensive threat assessment score, a three-level response mechanism is triggered. Based on the triggered response level, a corresponding hierarchical response strategy is generated. The hierarchical response strategy includes a set of video surveillance focus area coordinates, access control lock logic rules, and air conditioning air volume adjustment parameters. The three-level response mechanism includes:

[0136] Level 1 response: When the comprehensive threat assessment score is greater than 90, the target area access lock and laser tracking video focus are activated;

[0137] Level 2 response: When 70 < comprehensive threat assessment score ≤ 90, strobe lighting and directional acoustic alarms are activated in the evacuation passage;

[0138] Level 3 response: When the comprehensive threat assessment score is ≤70, non-core production equipment, including cooling water circulation pump units and ventilation auxiliary units, will switch to energy-saving mode;

[0139] The hierarchical response strategy is converted into a control instruction set in the target device protocol format, and the control instruction set is sent to the corresponding device through the corresponding communication protocol to perform the corresponding operations, including adjusting the air volume of the air conditioner, adjusting the viewing angle of the surveillance camera, and controlling the electromagnetic lock of the access control.

[0140] In this embodiment of the present invention, temperature and humidity sensors are deployed in key areas of a building (such as offices, corridors, and conference rooms). These sensors collect ambient temperature and humidity data at regular intervals (e.g., every minute) and transmit the data in real time to a central control system. For example, in an office, the temperature and humidity sensors record a current temperature of 26°C and a humidity of 50%. This is achieved through the use of various positioning technologies. Wi-Fi probes estimate the location of individuals by detecting the device's MAC address and signal strength; Bluetooth beacons use Bluetooth signals for more accurate close-range positioning; and video surveillance systems use image recognition algorithms to detect and count individuals in the surveillance footage. This location information from various sources is integrated and a specialized algorithm is used to generate a heat map of densely populated areas, visually displaying the density of people in different areas. For example, in a shopping mall, if these technologies reveal a high density of people in the dining area on a certain floor, this area will appear darker on the heat map.

[0141] Video surveillance systems use computer vision technology to analyze surveillance video frame by frame. Using pre-set behavioral models, they can identify unusual behaviors such as people running, remaining stationary in sensitive areas for extended periods, and large gatherings. For example, if someone suddenly starts running in a bank teller area, the video analysis system extracts relevant features of the behavior, including the time, location, and number of people involved, and marks it as unusual. Various security sensors, such as intrusion detection sensors, smoke sensors, and emergency buttons, are installed at strategic locations within the building (such as entrances and exits, computer rooms, and warehouses). Once a sensor detects an anomaly, it immediately sends an alarm signal to the central control system. For example, if an intrusion detection sensor detects an illegal intrusion, it sends a signal containing information such as the intrusion location and the type of alarm.

[0142] Because this data comes from different devices and systems, their collection times and frequencies may be inconsistent, necessitating spatiotemporal alignment. Using a high-precision clock source (such as an atomic clock or network time server) as a benchmark, the data timestamps from different data sources are calibrated to ensure temporal consistency across all data. Furthermore, different types of data are matched to specific physical locations based on the device's installation location and monitoring range. For example, data from temperature and humidity sensors in a room on a certain floor, data on abnormal behavior captured by the room's surveillance camera, and alarm signals from security sensors in the same room can all be associated with the room's specific coordinates, ensuring that these data correspond to the same area in both time and space.

[0143] A confidence level is determined for each data source based on factors such as data reliability, device performance, and the accuracy of historical data. High-precision and stable temperature and humidity sensors are assigned a higher confidence level, such as 0.9; wireless security sensors, which are susceptible to environmental interference, might receive a confidence level of 0.7. Video analytics systems that have been trained with a large number of samples and demonstrate high accuracy in real-world applications can receive a confidence level of 0.8. The confidence level of heat maps of densely populated areas depends on the accuracy of positioning technology and the timeliness of data updates. For accurate positioning and real-time updates, a confidence level of 0.85 can be used. The system dynamically adjusts the confidence level based on factors such as the device's operating status and the accuracy of historical data.

[0144] According to the importance and reliability of each data, assign corresponding weights to them. Assume that the weight of the ambient temperature and humidity comfort index is 0.2, the weight of the heat map of the densely populated area is 0.3, the weight of the abnormal behavior feature of the video analysis is 0.3, and the weight of the security sensor alarm signal is 0.2. Before performing the fusion calculation, the data is first standardized and converted into a unified numerical range (such as 0-1) to eliminate the dimensional differences of different data types. Standardization can be done by normalization method, such as converting the temperature data through the formula Normalize, where is the original data, and The values are the minimum and maximum values for the data type, respectively. The comprehensive threat assessment score is then calculated using the weighted formula: Comprehensive Threat Assessment Score = 0.2 × Standardized Value of Ambient Temperature and Humidity Comfort Index + 0.3 × Standardized Value of Crowded Area Heat Map + 0.3 × Standardized Value of Abnormal Behavior Characteristics from Video Analysis + 0.2 × Standardized Value of Security Sensor Alarm Signals.

[0145] Threshold setting:

[0146] Level 1 response (>90): For high-risk incidents (such as armed break-ins and fires), immediate physical isolation and precise tracking are required.

[0147] Level 2 response (70-90): corresponds to medium-risk events (such as conflicts involving gatherings of people, localized smoke), requiring evacuation guidance and warnings.

[0148] Level 3 response (≤70): Low risk or energy efficiency optimization scenarios (such as off-peak hours), switch the device to energy-saving mode.

[0149] First-level response strategy:

[0150] Call the preset laser tracking algorithm to control the PTZ camera to focus on the target coordinates (such as the intruder's position) and update the focus area in real time through the SLAM algorithm.

[0151] Access control: Locks all electromagnetic access control systems in the target area (such as exit doors) and sends a "forced lock" command to the central control.

[0152] Air conditioning adjustment: Turn off the fresh air system to prevent smoke from spreading, and increase the exhaust air volume to the maximum level.

[0153] Secondary response strategy:

[0154] Evacuation guidance: triggers strobe lighting (500Hz flashing frequency) and directional sound waves (145dB directional alarm) in the evacuation passage.

[0155] Video surveillance: Adjust the camera to the preset evacuation route perspective (such as corridors and stairwells) and enable crowd density analysis.

[0156] Three-level response strategy:

[0157] Equipment energy saving: non-core equipment (such as cooling water pumps) are switched to low-frequency operation mode (30% power), and ventilation units are operated at the minimum air exchange rate.

[0158] Convert focus coordinates (such as XYZ coordinates) into PTZ control commands in ONVIF protocol format ( <continuousmove>Order).

[0159] Access control system: Generate Wiegand protocol or IP instructions.

[0160] Air conditioning system: Converted to Modbus RTU commands (such as writing to register address 0x4001 and setting the air volume value to 80%).

[0161] Send commands asynchronously via the MQTT message queue, ensuring that high-priority commands (such as level 1 responses) preempt the transmission channel. Use transaction logs to record command status and automatically retry on failure (e.g., three retries + manual alert).

[0162] Assume that in a large shopping mall:

[0163] During the weekend afternoon shopping rush, the temperature and humidity sensors on the second floor of the mall detected a temperature of 30°C and a humidity of only 30%, far exceeding the human comfort range. After normalization, the ambient temperature and humidity comfort index was 0.8 (assuming the comfort range corresponds to a normalized value of 0-1, with values closer to 1 indicating less comfort). The personnel positioning system indicated that the clothing area on the second floor was densely populated, appearing dark red on the heat map, with a normalized heat map value of 0.9. The video analysis system detected abnormal behavior in the clothing area, with a customer lingering in front of a store and repeatedly looking inside. This was identified as abnormal behavior, and the normalized value of the abnormal behavior characteristic in the video analysis was 0.7. At this time, the anti-theft sensor in the jewelry store on the first floor suddenly issued an intrusion alarm signal, and the normalized value of the security sensor alarm signal was 1.

[0164] Based on the pre-set confidence level and the weight of each data, the comprehensive threat assessment score is assumed to be 72.9.

[0165] Since the score of 72.9 falls within the range of 70 < Comprehensive Threat Assessment Score ≤ 90, the secondary response mechanism is triggered. Upon receiving the command, the lighting in all evacuation routes in the mall begins flashing at a specific frequency (e.g., three times per second), creating a clear visual guidance signal. The mall's public address system activates a directional acoustic alarm, broadcasting a voice message to the second-floor clothing area and surrounding areas: "An unusual situation has been detected within the mall. Please remain calm and follow the evacuation signs to exit the area in an orderly manner." The secondary response strategy is converted into protocol-formatted instructions for the corresponding devices. For lighting devices, the strobe command is converted into a specific control code using the DALI protocol and then transmitted. For the public address system, the voice alarm command is transmitted to the corresponding area's speakers via the IP network protocol. Upon receiving the command, the lighting begins flashing, and the public address system plays a voice alarm, completing the response.

[0166] By integrating multi-source data for comprehensive threat assessment, various potential security risks can be identified promptly and accurately. From environmental anomalies to unusual human behavior and security alarms, a three-tiered response mechanism ensures appropriate handling of threats of varying severity. For example, Level 1 quickly locks access to the target area and conducts precise monitoring to prevent the spread of danger. Level 2 uses lighting and alarms to guide evacuation, reducing the risk of casualties. Level 3 optimizes equipment operation during low-risk situations, providing a stable foundation for security protection and comprehensively ensuring the safety of people and property within the building. Equipment operation strategies are dynamically adjusted based on threat levels to ensure optimal resource allocation. Level 3 switches non-core production equipment to energy-saving mode to avoid unnecessary energy consumption during low-risk conditions. Targeted adjustments to video surveillance, access control, lighting, and other equipment are made at different response levels to avoid resource waste, reducing operating costs and improving resource efficiency while ensuring safety and comfort. A clear three-tiered response mechanism and automated command generation and issuance process enable the system to react quickly to alarm signals and threats. From threat assessment to response strategy execution, excessive human intervention is eliminated, significantly shortening emergency response time. For example, in the event of an emergency, evacuation corridor lighting and directional acoustic alarms can be quickly activated to guide personnel to evacuate in an orderly manner, effectively improving the building's overall emergency response capabilities and efficiency. In addition to safety-related responses, this mechanism also considers environmental comfort indicators. During daily operations, by adjusting air conditioning air volume and aligning it with the distribution of crowded areas, we ensure that people in different areas are in a comfortable environment. Furthermore, refined control and management of equipment makes building operations management more scientific and intelligent, improving overall management level and service quality.

[0167] An embodiment of the present invention further provides an intelligent gateway, comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, executes the system described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0168] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, cause the computer to execute the system described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0169] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.< / continuousmove>

Claims

1. An AI artificial intelligence machine learning system, characterized in that: include: The communication module establishes a two-way connection with the building's air conditioning units, security sensor arrays, and smart lighting networks through multiple IoT protocols, collecting real-time data streams on equipment operation, including air conditioning compressor current waveforms, security alarm codes, and lighting equipment light intensity timing data. The edge computing module is used to extract air conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets based on the device operation data stream, and store the processed structured data sets in the pseudo-physical memory area through the pseudo-physical classification algorithm; The quasi-physical memory area module is used to implement timestamp difference encoding on the historical power data of air conditioners in the structured data set to form a spatiotemporal correlation data cube. This module extracts the historical power data of air conditioners from the structured data set, groups it by device ID, and arranges it by timestamp to form a time-series power sequence. For each adjacent timestamp pair, the power difference and time interval are calculated to form a power difference code sequence. The difference code sequence is then associated with the geographic coordinates of the device to construct a three-dimensional data cube containing time, space, and feature dimensions. The engine module is used to identify the thermal map pattern of human movement within the building based on the spatiotemporal correlation data cube in the simulated memory area and generate a dynamic temperature control parameter matrix. This includes extracting multi-source probe positioning data from the spatiotemporal correlation data cube in the simulated memory area and processing the probe positioning data using a long-short-term memory network to predict the probability of human presence in each area over a period of time in the future and generate a dynamic thermal distribution map. The thermal distribution map is then spatially overlaid with the temperature and humidity gradient data stored in the simulated memory area to identify densely populated areas and areas with abnormal environmental parameters, thereby obtaining coupled analysis results. The control logic module generates control instructions based on environmental comfort indicators, the distribution of crowded areas, abnormal behavior pattern recognition, and security threat levels. Upon receiving an alarm signal, it dynamically adjusts the video surveillance focus area and access control strategy based on the signal type through a three-level priority response mechanism. This involves temporally and spatially aligning and confidence-weighting the environmental temperature and humidity comfort indicators, heat maps of crowded areas, abnormal behavior characteristics analyzed through video analysis, and security sensor alarm signals to generate a comprehensive threat assessment score. A three-level response mechanism is triggered based on the comprehensive threat assessment score. The three-level response mechanism includes: Level 1 response: When the comprehensive threat assessment score is greater than 90, the target area access lock and laser tracking video focusing are activated; Level 2 response: When the comprehensive threat assessment score is 70 < Comprehensive threat assessment score ≤ 90, the evacuation channel strobe lighting and directional acoustic alarm are activated; Level 3 response: When the comprehensive threat assessment score is ≤ 70, non-core production equipment, including cooling water circulation pumps and ventilation auxiliary units, are switched to energy-saving mode; The protocol conversion module is used to dynamically convert control instructions into the target device protocol format.

2. The AI artificial intelligence machine learning system according to claim 1, characterized in that: Based on the equipment operation data stream, the air conditioning energy consumption characteristic values, security alarm signals, and ambient temperature and humidity data sets are extracted. The processed structured data sets are stored in the simulacrum memory area through the simulacrum classification algorithm, including: Perform sliding window processing on the original device data stream to split the continuous data stream into multiple data segments of equal length; Extract air conditioner energy efficiency feature sets from multiple equal-length data segments, including real-time energy efficiency ratio, current harmonic distortion rate, and load fluctuation index. Generate an alarm frequency density matrix based on security sensor data to obtain a feature dataset containing air conditioner energy efficiency features and the spatiotemporal distribution of security events. The feature data set is generated into device-level classification labels through the device fingerprint encoding algorithm, and the classification data is mapped to the simulacrum memory area according to the device entity ID.

3. The AI artificial intelligence machine learning system according to claim 2, characterized in that: Based on the spatiotemporal correlation data cube in the simulacrum, the thermal map pattern of occupant movement in the building is identified and a dynamic temperature control parameter matrix is generated, including: According to the coupling analysis results and the thermal response characteristics of the air-conditioning units, the regional temperature setpoint matrix is dynamically calculated.

4. The AI artificial intelligence machine learning system according to claim 3, characterized in that: Extract probe location data from the spatiotemporal correlation data cube in the simulated memory area and process it using a long short-term memory network to predict the probability of personnel staying in each area in the future and generate a dynamic heat distribution map, including: Extract multi-source probe positioning data from the spatiotemporal correlation data cube of the simulacrum memory area, including probe connection events, Bluetooth beacon scanning records, and video analysis coordinate point clouds; The multi-source probe positioning data are grouped by person ID and the spatiotemporal sequence of individual movement trajectories is constructed, which is then segmented into continuous trajectory segments using a sliding window. The trajectory segments are input into the pre-trained long short-term memory network, and the probability distribution of personnel residence in each area within the future target period is predicted based on the historical movement patterns. The residence probability distribution is spatially interpolated and optimized, and a dynamic thermal distribution map is generated based on the building topology.

5. The AI artificial intelligence machine learning system according to claim 4, characterized in that: Based on the coupling analysis results and the thermal response characteristics of the air conditioning units, the regional temperature setpoint matrix is dynamically calculated, including: Based on the historical temperature data in the building, the dynamic baseline temperature is determined, and the probability of occupancy in each area during the future target period is compared with the actual probability of occupancy at the current time point to obtain the occupancy density change rate. Adaptive learning algorithms are used to optimize the occupant heat load adjustment factor based on the thermal response characteristics of the air conditioning equipment, including response speed and adjustment range, and historical energy efficiency data, including energy consumption and cooling capacity; Real-time collection of outdoor environmental humidity monitoring values and meteorological forecast data, combined with the thermal inertia parameters of the air-conditioning unit, including heat capacity and heat conduction values, to generate external disturbance compensation factors; The reference temperature, personnel heat load adjustment factor and external disturbance compensation factor are multi-dimensionally integrated to generate the temperature control parameter matrix of the spatial partition.

6. The AI artificial intelligence machine learning system according to claim 5, characterized in that: Generate control instructions based on environmental comfort indicators, distribution of crowded areas, identification of abnormal behavior patterns, and security threat levels. When an alarm signal is received, dynamically adjust the video surveillance focus area and access control strategy based on the signal type through a three-level priority response mechanism, including: Generate a corresponding hierarchical response strategy based on the triggered response level. The hierarchical response strategy includes a set of video surveillance focus area coordinates, access control lock logic rules, and air conditioning volume adjustment parameters. The hierarchical response strategy is converted into a control instruction set in the target device protocol format, and the control instruction set is sent to the corresponding device through the corresponding communication protocol to perform the corresponding operations, including adjusting the air volume of the air conditioner, adjusting the viewing angle of the surveillance camera, and controlling the electromagnetic lock of the access control.

7. An intelligent gateway, characterized in that: include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the system according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the system according to any one of claims 1 to 6.

Images