Multi-device linkage control method for building intelligent system

Abstract

The invention relates to the technical field of building intelligent control, in particular to a multi-device linkage control method for a building intelligent system, which comprises the following steps: collecting multi-source heterogeneous operation data, and preprocessing to form a standardized global linkage data set; establishing a cross-subsystem global collaborative decision engine, and formulating a multi-level linkage execution priority mechanism; carrying out feature matching and model reasoning through a reinforcement learning model, and dynamically generating an adaptive linkage rule; according to a self-adaptive linkage rule and a multi-level linkage execution priority, performing unified time sequence arrangement on the multi-device control instruction, and distributing the multi-device control instruction to a corresponding device terminal for execution; and running effect data is transmitted back to the reinforcement learning model, and iterative optimization of the adaptive linkage rule is completed. According to the method, heterogeneous data are processed in a standardized mode, a global collaborative decision engine is built, linkage rules are dynamically generated and iterated in combination with a reinforcement learning model, and the self-adaption and refinement level of linkage control is improved.

Description

Technical Field

[0001] This invention relates to the field of building intelligent control technology, and in particular to a method for multi-device linkage control of a building intelligent system. Background Technology

[0002] In the field of building intelligence, multi-device linkage control technology has been widely applied in the intelligent renovation and new construction projects of various buildings. Existing technologies generally collect operational data such as the building's internal environment, personnel, equipment, and energy consumption by deploying sensing nodes. Relying on a simple control logic engine combined with manually preset linkage rules, basic linkage control of various subsystem devices within the building is achieved. Some technical solutions also introduce machine learning models to optimize the linkage rules, forming a basic linkage process of "data acquisition - rule matching - instruction execution". This process has replaced manual control to a certain extent and improved the level of intelligence in building operation.

[0003] Existing multi-device linkage control methods in building intelligent systems still suffer from intractable technical defects. On the one hand, the linkage rules of existing technologies are mostly fixed logics preset by humans, which can only adapt to single or simple building operation scenarios. They cannot adaptively generate and optimize linkage rules based on dynamic changes in building environmental parameters, personnel status, and equipment operating conditions, lacking self-learning and self-iteration capabilities, resulting in extremely poor scenario adaptability. On the other hand, existing technologies lack a global collaborative decision-making architecture across subsystems. The linkage control of subsystems such as HVAC, lighting, and fire protection is independent of each other, and a unified linkage execution priority mechanism and standardized decision-making and execution process have not been established. This easily leads to problems such as conflicting control commands and disordered execution timing among multiple devices, making it impossible to achieve multi-objective collaborative optimization of building operation safety, energy saving, and comfort. At the same time, the lack of quantitative judgment standards for linkage decision-making and command arrangement further reduces the accuracy and reliability of multi-device linkage control. Summary of the Invention

[0004] To address the technical deficiencies in the background technology, this invention proposes a multi-device linkage control method for building intelligent systems, which solves the aforementioned technical problems and meets practical needs. The specific technical solution is as follows: A method for multi-device linkage control in a building intelligent system includes the following steps: Collect multi-source heterogeneous operational data from the entire building interior, perform unified format conversion and normalization on the collected data, and form a standardized global linkage data set; A cross-subsystem global collaborative decision-making engine is built based on a standardized global linkage data set, and a multi-level linkage execution priority mechanism is formulated to clarify the execution permissions of different linkage scenarios; By using a reinforcement learning model to perform feature matching and model inference on a standardized global linkage data set, adaptive linkage rules that fit the current operating scenario are dynamically generated. The global collaborative decision engine arranges the control commands of multiple devices in a unified time sequence based on adaptive linkage rules and multi-level linkage execution priorities, and distributes the arranged commands to the corresponding device terminals for execution. Collect the execution effect data after the command is executed and send it back to the reinforcement learning model to complete the iterative optimization of the adaptive linkage rules.

[0005] Furthermore, the specific steps for forming the standardized globally linked data set are as follows: Collect building interior environmental monitoring data, personnel status data, equipment operation status data, energy consumption data, security monitoring data, and fire early warning data point by point, and transmit the collected raw data to the local data preprocessing unit; The local data preprocessing unit performs outlier removal and noise filtering on the raw data to remove invalid data that exceeds the preset value range. The processed data is uniformly converted to a transmission format to eliminate data format differences caused by communication protocols of different subsystems. The data after the format is unified is mapped to numerical dimension normalization, which transforms data with different dimensions into the same numerical range. The processed data is integrated and packaged to form a standardized global linkage dataset, which is then merged and transmitted to the global collaborative decision-making engine.

[0006] Furthermore, the specific steps for building a cross-subsystem global collaborative decision-making engine are as follows: Based on the storage location and data type of the standardized global linkage data set, a global collaborative decision engine is built across HVAC, lighting, fire protection, security, elevator and access control subsystems; The global collaborative decision-making engine determines the priority of linkages based on the impact of linkage scenarios on building operation safety. Based on the judgment dimensions, the priority of the linkage execution is divided into multiple levels, and the linkage scenario type corresponding to each priority level is locked; Configure an independent instruction execution channel for each priority level and limit the resource consumption limit of linked requests of different priorities; The global collaborative decision engine establishes rules for determining the permission of linkage requests, determines the instruction execution permissions for different linkage scenarios based on priority levels, and receives standardized global linkage data sets and adaptive linkage rules to initiate the linkage decision execution process.

[0007] Furthermore, the specific steps for dynamically generating adaptive linkage rules that fit the current running scenario are as follows: Reinforcement learning models are used to extract environmental features, personnel features, equipment features, energy consumption features, and scenario features from a standardized global interconnected dataset. The extracted multidimensional features are input into a preset neural network structure for forward inference calculation; The system calculates and outputs the linkage trigger threshold, linkage device combination, and linkage operation parameters that are adapted to the current scenario. Based on the linkage trigger threshold, linkage device combination, and linkage operation parameters, control logic that does not match the current scenario is eliminated, and effective control logic is retained. Effective control logic is combined and encapsulated to generate adaptive linkage rules without human pre-setting and transmit them to the global collaborative decision engine.

[0008] Furthermore, the specific steps for timing the multi-device control commands and distributing them to the device terminals for execution are as follows: The global collaborative decision engine receives adaptive linkage rules and multi-level linkage execution priority information to determine the core constraints for instruction execution. Initiate a unified timing arrangement process for multi-device control commands, plan the execution order of multi-device control commands based on core constraints, and determine the trigger time node for each command; Perform device control conflict verification on the planned multi-device control commands to eliminate the conflict problem of multiple commands being triggered in parallel on the same device; The verified multi-device control commands are encapsulated in command format to generate execution commands that can be directly recognized by the terminal. The encapsulated instructions are distributed according to the target device address, driving the corresponding device terminal to perform control operations.

[0009] Furthermore, the specific steps for completing the iterative optimization of the adaptive linkage rule are as follows: Throughout the entire process of executing instructions on the device terminal, real-time data on the operational effects after instruction execution is collected and transmitted to the data aggregation unit. The data aggregation unit quantifies and transforms the operational performance data, converting non-numerical data into numerical data that can participate in model calculations. The data aggregation unit feeds back the quantified performance data to the input layer of the reinforcement learning model; The reinforcement learning model calculates the execution deviation value of the current adaptive linkage rule based on the returned execution effect data; The reinforcement learning model adjusts its internal network parameters based on the execution bias value, corrects the logic and parameters of the adaptive linkage rules, and completes the iterative optimization of the rules.

[0010] Furthermore, when initiating the coordinated decision-making execution process, the optimal coordinated strategy is determined through a global coordinated decision-making utility function, which is as follows: , in, Represents the total utility value of globally coordinated decision-making. This represents the weighting coefficient for the security dimension, with a value ranging from 0 to 1. The weighting coefficient for the security dimension is a fixed preset value. The score represents the safety dimension of the performance evaluation. The score value is determined based on the completeness of the performance evaluation in fire protection and security emergency scenarios. This represents the weighting coefficient for the energy-saving dimension, with a value ranging from 0 to 1. This weighting coefficient is dynamically adjusted based on the building's operating time. The score represents the energy-saving dimension, and the score value is determined based on the percentage reduction in equipment energy consumption. This represents the weighting coefficient for the comfort dimension, with a value ranging from 0 to 1. The weighting coefficient for the comfort dimension is dynamically adjusted based on the occupancy status of the personnel. The score represents the comfort dimension, and the score value is determined based on the degree of matching between environmental parameters and human comfort standards. The global collaborative decision engine selects the linkage strategy with the highest total utility value of global linkage decision as the final execution strategy.

[0011] Furthermore, when the global collaborative decision engine performs unified timing orchestration of control commands for multiple devices, it calculates the optimal execution timing through the linkage command timing scheduling cost function, which is as follows: , in, This represents the total cost of scheduling the linkage instructions. This represents the trigger execution time of the i-th control instruction. This represents the trigger execution time of the (i+1)th control instruction. This represents the instruction execution conflict penalty coefficient, a constant with a value greater than 0. The number of device control conflicts that occur during instruction execution; The global collaborative decision engine selects the timing scheme with the minimum total cost of linkage instruction timing scheduling as the final execution timing of multi-device control instructions.

[0012] Furthermore, when instructions are distributed to the corresponding device terminals for execution, a distributed execution architecture is adopted, and the specific execution process is as follows: The global collaborative decision engine determines the real-time level of instructions and marks control instructions with real-time requirements higher than a preset threshold as high real-time instructions. The global collaborative decision-making engine sends high real-time instructions to edge computing nodes, and the edge computing nodes directly establish communication connections with the corresponding device terminals. After receiving instructions, the edge computing node directly drives the device terminal to execute them, and the instruction execution response time is controlled within a preset range. The global collaborative decision-making engine uploads non-real-time global optimization instructions to the cloud server, where the cloud server performs global data statistics and long-term strategy optimization operations. Edge computing nodes maintain real-time data synchronization with cloud servers. When the network connection is interrupted, the edge computing node starts local independent operation mode and stores the execution data. After the network is restored, the stored data is automatically synchronized to the cloud server.

[0013] Furthermore, abnormal operating condition protection operations are performed throughout the entire process of adaptive linkage rule iterative optimization. The specific steps are as follows: The global collaborative decision-making engine monitors data acquisition status, network connection status, and device operation status in real time. When the global collaborative decision engine detects abnormal operating conditions such as missing data collection, network fluctuations and interruptions, or equipment failure and offline status, it terminates the execution process of the current adaptive linkage rule. The global collaborative decision engine invokes preset safety benchmark linkage strategies to prioritize driving fire protection, security, and emergency evacuation related equipment to perform control operations. The global collaborative decision-making engine suspends the execution of unnecessary energy-saving adjustments and comfort optimization commands to ensure the stable operation of the building's core safety functions; After the abnormal operating conditions are eliminated, the global collaborative decision engine automatically resumes the adaptive linkage rule execution mode. At the same time, it sends the abnormal operating condition data and the backup strategy execution data back to the reinforcement learning model to supplement the model training samples and improve the model's ability to adapt to abnormal operating conditions.

[0014] Compared with existing technologies, the multi-device linkage control method for building intelligent systems provided by this invention has the following beneficial effects: This invention first standardizes multi-source heterogeneous data across the entire domain to form a unified data set. Then, it builds a cross-subsystem global collaborative decision-making engine and establishes multi-level linkage priorities, effectively achieving global collaborative scheduling across multiple subsystems. This fundamentally avoids command conflicts and timing discrepancies, improving the orderliness and reliability of linkage control. Simultaneously, it dynamically generates adaptive linkage rules adapted to the current scenario through a reinforcement learning model. After the engine completes command orchestration and execution, the collected operational effect data is fed back to the model to complete rule iteration and optimization. This abandons the method of manually pre-setting fixed rules, realizing the dynamic generation and self-iteration of linkage rules, adapting to dynamic changes in the building's internal environment, personnel, and equipment operating conditions. The entire system forms a complete technical closed loop, achieving cross-subsystem global collaborative linkage, taking into account multiple objectives of building operation safety, energy conservation, and comfort, and significantly improving the adaptive capability and precision level of multi-device linkage control in building intelligent systems. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a multi-device linkage control method for an intelligent building system according to the present invention. Detailed Implementation

[0016] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0017] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0018] See Figure 1 This invention provides a method for multi-device linkage control in a building intelligent system, characterized by the following steps: Step S100: Collect multi-source heterogeneous operation data of the entire building interior, perform unified format conversion and normalization processing on the collected data, and form a standardized global linkage data set; The multi-source heterogeneous operational data within a building comprises unstructured or semi-structured raw data collected from various sensing nodes within the building, including environmental, personnel, equipment status, and energy consumption data. This data encompasses data sources from different subsystems, using different protocols and sampling frequencies, providing raw input information for subsequent data standardization and collaborative decision-making. The standardized global collaborative data set is a structured dataset formed after unifying the format, aligning semantics, and normalizing the numerical values ​​of the multi-source heterogeneous operational data within the building. It possesses consistent timestamps, unit systems, and field naming conventions, serving as a unified data foundation for cross-subsystem collaborative decision-making and model inference, supporting multi-dimensional data fusion analysis.

[0019] Step S200: Build a cross-subsystem global collaborative decision-making engine based on the standardized global linkage data set, formulate a multi-level linkage execution priority mechanism, and clarify the execution authorities for different linkage scenarios; The cross-subsystem global collaborative decision-making engine is a central decision-making module that integrates the control logics of multiple subsystems. It is responsible for receiving the adaptive linkage rules and priority mechanisms, executing instruction scheduling and distribution, and coordinating the control behaviors of subsystems such as HVAC, lighting, security, and fire protection. It is used to achieve the unified scheduling and orderly execution of multi-device control instructions, and ensure the coordination consistency of control actions in terms of logic and time. The multi-level linkage execution priority mechanism is a hierarchical strategy that defines the execution order of each control instruction under different linkage scenarios. It is used to solve the control conflict problem between multiple subsystems, determine the instruction execution order when multiple control requirements occur concurrently, and prevent instruction conflicts and timing disorders. The execution authorities for different linkage scenarios are the operation control scope and authorization levels set for specific operation scenarios, which restrict the executability of certain instructions under specific conditions. It is used to ensure that high-risk operations are only allowed to be executed when safety conditions are met, and improve the system security.

[0020] Step S300: Perform feature matching and model inference on the standardized global linkage data set through a reinforcement learning model, and dynamically generate adaptive linkage rules adapted to the current operation scenario; The reinforcement learning model is a machine learning model constructed based on the Markov decision process. It learns the optimal control strategy by interacting with the environment, and is used to dynamically generate adaptive linkage rules adapted to the current building operation state, replacing the manually preset rules. The adaptive linkage rule is a control logic expression dynamically output by the reinforcement learning model according to the current building operation state, which describes the linkage relationship and trigger conditions between devices, and is used to guide the cross-subsystem global collaborative decision-making engine to generate a control instruction sequence that meets the current scenario requirements.

[0021] Step S400: The global collaborative decision-making engine performs unified timing scheduling on multi-device control instructions according to the adaptive linkage rules and multi-level linkage execution priorities, and distributes the scheduled instructions to the corresponding device terminals for execution; The multi-device control instruction is a command signal generated by the decision-making engine for driving specific devices to execute actions, which includes elements such as target device identification, control parameters, and execution time. It is used to directly act on the end-execution devices to achieve physical operations such as temperature adjustment, light switch, and access control opening and closing. Performing unified timing scheduling on multi-device control instructions refers to the process in which the global collaborative decision-making engine uniformly plans and arranges the execution order and trigger time of multi-device control instructions according to the adaptive linkage rules and multi-level priorities, aiming to avoid instruction conflicts and ensure the orderly execution of multi-device linkages.

[0022] Step S500: Collect the running effect data after the command is executed and send it back to the reinforcement learning model to complete the iterative optimization of the adaptive linkage rules.

[0023] The operational effect data after the instruction is executed is the system response data fed back after the control instruction is executed. It reflects the deviation between the actual operating state and the expected goal. It is used as a reward signal source for the reinforcement learning model, to evaluate the effectiveness of the current rule and drive the model parameter update.

[0024] Taking energy-saving control during off-hours in office areas as an example, the application method of this invention is as follows: When the system detects that there is no personnel activity in the office after 18:00 on a weekday (based on infrared sensors and access control records) and there is sufficient outdoor lighting, the standardized data set updates this status. The reinforcement learning model identifies this as a typical off-peak energy-saving scenario and generates an adaptive linkage rule of "turn off the air conditioner, dim the lighting, and keep the security system powered." The cross-subsystem global collaborative decision engine, combined with a multi-level linkage execution priority mechanism (security-related instructions take priority), confirms that the security system must not be powered off, and thus arranges an instruction sequence of "turn off the air conditioner first → turn off the lighting 5 minutes later → keep the security system powered." After the instructions are executed, the system collects the rate of increase in room temperature, the decrease in lighting power consumption, and nighttime intrusion alarm records as operational effect data, and sends them back to the model to evaluate the effectiveness of this energy-saving strategy, and optimizes the delayed light-off time in the next similar scenario to balance energy saving and backup lighting needs.

[0025] This invention lays a unified data foundation for cross-subsystem collaboration through data standardization processing. Relying on a global collaborative decision engine and a multi-level priority mechanism, it achieves full-domain collaborative scheduling of various intelligent subsystems in a building, fundamentally avoiding problems such as conflicting control commands and disordered execution timing, and improving the orderliness and reliability of linkage control. At the same time, it realizes the dynamic generation and self-iterative optimization of adaptive linkage rules through a reinforcement learning model, abandoning the traditional method of manually preset fixed rules. This allows linkage control to adapt to the dynamic changes in the building's internal environment, personnel, and equipment conditions in real time, solving the defects of traditional linkage rules being rigid and having poor scene adaptability. Overall, it realizes cross-subsystem full-domain collaborative linkage of multiple devices in the building's intelligent system, taking into account the multi-objective optimization of building operation safety, energy saving, and comfort, and significantly improving the intelligence, precision, and adaptability of multi-device linkage control in building intelligence.

[0026] In one embodiment of the present invention, the specific steps for forming a standardized globally linked data set are as follows: Step S101: Collect building interior environmental monitoring data, personnel status data, equipment operating status data, energy consumption data, security monitoring data, and fire early warning data point by point, and transmit the collected raw data to the local data preprocessing unit; Dedicated sensing and data acquisition nodes are deployed point by point in each functional area and subsystem of the building. Environmental monitoring data is collected through temperature, humidity, illuminance, CO2 concentration, and meteorological sensors; personnel status data is collected through binocular cameras, people detectors, and access control card readers; equipment operating status data is collected through equipment PLCs and status acquisition modules; energy consumption data is collected through smart meters, water meters, and energy consumption meters; security monitoring data is collected through video cameras and intrusion detectors; and fire early warning data is collected through smoke detectors, temperature detectors, and fire valve status acquisition modules. All collected raw data is transmitted in real time to the local data preprocessing unit deployed on the building site via building site industrial bus (KNX / BACnet / Modbus) or wireless communication (LoRa / WiFi) to complete the centralized reception of raw data.

[0027] Step S102: The local data preprocessing unit performs outlier removal and noise filtering on the raw data to remove invalid data that exceeds the preset value range. The local data preprocessing unit has a built-in data cleaning algorithm. First, it uses the 3σ principle and Grubbs' test to remove outliers from the raw data. It sets preset value ranges for each type of data (these ranges are based on building intelligent design specifications and equipment operating parameters) and automatically identifies and removes invalid and abnormal data that are outside the range. Then, it uses moving average filtering and median filtering to filter out noise from the data after removing outliers, eliminating random noise interference generated during sensor acquisition and data transmission, and ensuring the authenticity and validity of the data.

[0028] Step S103: Perform a unified conversion on the transmission format of the processed data to eliminate data format differences caused by communication protocols of different subsystems; To address the data format differences caused by heterogeneous communication protocols such as BACnet, KNX, Modbus, and proprietary protocols used in different subsystems such as HVAC, lighting, and fire protection, the local data preprocessing unit has a built-in multi-protocol conversion module and format parsing engine. This module converts heterogeneous data formats such as binary, hexadecimal, and character data under different protocols into the JSON / Protobuf standard data format. It also adds unified field identifiers, collection timestamps, and collection location tags to each type of data, eliminating data format barriers between different subsystems.

[0029] Step S104: Perform numerical dimension normalization mapping on the data after format unification to convert data with different dimensions to the same numerical range; To address the issue of data incompatibility due to dimensional differences in different data types (e.g., temperature in °C, illuminance in lx, energy consumption in kWh), a minimum-maximum normalization algorithm is used to map the numerical dimensions of the data after format unification. This transforms the values ​​of all data types to a unified numerical range of **[0,1]**. The transformation process strictly preserves the relative change characteristics of the data, ensuring that the normalized data can still accurately reflect the building's operational status and achieving the fusion of multiple data types.

[0030] Step S105: Integrate and encapsulate all processed data to form a standardized global linkage dataset, which is then merged and transmitted to the global collaborative decision engine.

[0031] The local data preprocessing unit classifies and integrates all data after it has been formatted and normalized into six categories: "environment / personnel / equipment / energy consumption / security / fire protection". It adds a check code to the integrated data using a data block encapsulation method and transmits it in real time to the dedicated database of the cross-subsystem global collaborative decision engine through an encrypted communication link. This forms a standardized global linkage data set that can be directly called by subsequent links. At the same time, it performs integrity verification of data transmission to ensure that the data is not lost or tampered with.

[0032] In one embodiment of the present invention, the specific steps for building a cross-subsystem global collaborative decision-making engine are as follows: Step S201: Based on the storage location and data type of the standardized global linkage data set, build a global collaborative decision engine that spans HVAC, lighting, fire protection, security, elevator and access control subsystems; Specifically, the storage location (building local edge database / cloud distributed database) and data types (structured equipment operating data, semi-structured environmental perception data, and time-series energy consumption data) of the standardized global linkage data set are analyzed. A cross-subsystem global collaborative decision-making engine is built based on a microservice distributed architecture. On the hardware side, an industrial-grade redundant server is used as the core computing node of the engine. On the software side, communication interfaces and protocol parsing modules compatible with HVAC, lighting, fire protection, security, elevator, and access control subsystems are developed to realize bidirectional data interaction and command transmission between the engine and the equipment of each subsystem. At the same time, a dedicated data call interface is configured for the engine to ensure that the standardized global linkage data set can be read in real time.

[0033] Step S202: Based on the impact of the linkage scenario on the building's operational safety, the global collaborative decision engine divides the linkage priority into judgment dimensions. The global collaborative decision-making engine has a built-in scenario impact analysis algorithm. Based on the degree of impact of linked scenarios on building operation safety, it divides the linkage priority judgment dimensions into three dimensions: impact on life safety, impact on equipment and asset safety, and impact on building operation efficiency. Among them, impact on life safety is the core judgment dimension, impact on equipment and asset safety is the secondary judgment dimension, and impact on building operation efficiency is the basic judgment dimension. Each dimension is set with quantitative scoring standards according to its impact weight, providing a quantitative basis for subsequent priority classification.

[0034] Step S203: Divide the linkage execution priority into multiple levels according to the judgment dimension, and lock the linkage scenario type corresponding to each priority level; Based on the defined judgment dimensions and quantitative scoring standards, the priority of linkage execution is divided into a multi-level gradient. The highest priority is determined first by the dimension of life safety impact, and then the lower priority is determined by combining the other dimensions in turn, which is conventionally divided into three levels of priority. At the same time, according to the actual scenarios of building intelligent operation, scenarios involving life safety, such as fire emergency and security intrusion, are locked as the highest priority, scenarios involving equipment asset safety, such as equipment failure protection and energy over-limit control, are locked as the medium priority, and scenarios involving operational efficiency, such as environmental comfort adjustment and daily energy saving optimization, are locked as the basic priority. A unique mapping relationship between priority and linkage scenario is established and solidified in the engine.

[0035] Step S204: Configure an independent instruction execution channel for each priority level and limit the resource consumption limit of linked requests of different priorities; Each priority level is configured with a dual-layer independent instruction execution channel consisting of physical isolation and logical isolation. The highest priority level is configured with an independent industrial bus physical communication channel, while the medium and basic priorities use logically isolated software communication channels to avoid interference between the transmission of instructions of different priorities. At the same time, based on the total amount of hardware computing resources (CPU, memory) and communication resources (bandwidth, transmission rate) of the building intelligent system, the resource consumption limit for each priority level is limited. The highest priority level has no upper limit on the resources it occupies and can preempt the resources of other priorities. The remaining resources are allocated proportionally to the medium and basic priorities to ensure the instruction execution efficiency of high-priority linkage scenarios.

[0036] Step S205: The global collaborative decision engine establishes rules for determining the permission of linkage requests, determines the instruction execution permissions for different linkage scenarios based on priority levels, and receives standardized global linkage data sets and adaptive linkage rules to start the linkage decision execution process.

[0037] The global collaborative decision-making engine establishes standardized rules for determining the permission of collaborative requests based on priority levels, types of collaborative scenarios, and resource usage rules. It clarifies the triggering conditions, execution permissions, interruption rules, and resource call permissions for collaborative requests of different priorities. The rules set the logic that high-priority requests can interrupt the execution of low-priority requests. After the rules are established, a data and rule receiving interface is developed for the engine to receive the standardized global collaborative data set and the subsequently generated adaptive collaborative rules in real time. This completes the engine's initial configuration, officially starts the collaborative decision-making execution process, and enters the state of waiting for decision and instruction scheduling.

[0038] In one embodiment of the present invention, the specific steps for dynamically generating adaptive linkage rules that fit the current running scenario are as follows: Step S301: Extract environmental features, personnel features, equipment features, energy consumption features, and scene features from the standardized global linkage data set using a reinforcement learning model; The reinforcement learning model utilizes a built-in PCA principal component analysis and lightweight convolutional neural network feature extraction module to extract targeted features from a standardized global interconnected dataset. This module accurately extracts five core feature categories: environmental features, personnel features, equipment features, energy consumption features, and scene features. The extracted raw features are then normalized and dimensionality reduced to remove redundant features and retain the core effective features. This results in a structured, multi-dimensional feature set that can be directly used for model inference. All data in the feature set are quantified values ​​to ensure the accuracy of subsequent inference calculations.

[0039] Step S302: Input the extracted multidimensional features into the preset neural network structure and perform forward inference calculation; The processed multidimensional feature set is input into a neural network structure pre-trained with historical building operation data according to a preset data format. In this scheme, the neural network structure adopts a deep Q network (DQN) (adapted to the decision-making reasoning requirements of reinforcement learning). The model performs forward propagation calculation according to a preset loss function, learning rate and number of iterations. The data is propagated from the network input layer through the hidden layer to the output layer. The mapping between multidimensional features and building linkage strategy is completed through neuron activation and weight matrix calculation, and the correlation weight value between the output features and linkage rules is output.

[0040] Step S303: Calculate and output the linkage trigger threshold, linkage device combination, and linkage operation parameters adapted to the current scene through reasoning; The reinforcement learning model, based on the association weight values ​​obtained by forward inference calculation, combined with the feature attributes of the current building's real-time operation scenario, infers and outputs three sets of core linkage parameters, namely linkage trigger threshold, linkage equipment combination, and linkage operation parameters. All output parameters are quantified values ​​or structured equipment sets, which can be directly converted into executable control logic. Moreover, the parameters are highly adapted to the current building's environment, personnel, and equipment operating conditions, without human intervention or preset.

[0041] Step S304: Based on the linkage trigger threshold, linkage device combination and linkage operation parameters, eliminate control logic that does not match the current scenario and retain effective control logic; The reinforcement learning model incorporates a scene matching quantification algorithm, which calculates the matching degree between the three sets of core linkage parameters output and the current real-time operation scene of the building. It presets a scene matching degree threshold (set according to the building intelligent operation specifications), automatically eliminates control logic with a matching degree lower than the threshold, and retains only the effective control logic that is highly adapted to the current scene, thereby avoiding linkage control errors caused by invalid logic and ensuring the effectiveness and accuracy of the rules.

[0042] Step S305: Combine and encapsulate the effective control logic to generate adaptive linkage rules without human preset and transmit them to the global collaborative decision engine.

[0043] The selected effective control logic is combined and encapsulated in a fixed structure of "trigger condition - linkage device combination - operating parameters - basic execution requirements". A unique identifier, scene tag and generation timestamp are added to the generated linkage rules to form adaptive linkage rules without manual preset. Then, the adaptive linkage rules are transmitted in real time to the dedicated rule storage module of the cross-subsystem global collaborative decision engine through an encrypted communication link to complete the delivery of the rules, which can be directly called by the engine when arranging instructions.

[0044] In one embodiment of the present invention, the specific steps of timing the multi-device control commands and distributing them to the device terminals for execution are as follows: Step S401: Receive adaptive linkage rules and multi-level linkage execution priority information through the global collaborative decision engine to determine the core constraints for instruction execution; The global collaborative decision engine receives adaptive linkage rules transmitted by the reinforcement learning model in real time through a built-in standardized data interface. At the same time, it retrieves its own pre-stored multi-level linkage execution priority information, and analyzes and merges the two types of information. Combining the physical operating characteristics of building intelligent equipment (such as the timing dependence of equipment start-up and shutdown, and mutual exclusion operations of the same equipment), the logical requirements of linkage scenarios (such as the sequential execution logic of smoke exhaust and fresh air in fire protection scenarios), and resource occupation rules, it extracts and determines the core constraints for the execution of multi-device control commands. These constraints provide a rigid rule basis for the subsequent command timing arrangement and are strongly bound to the linkage priority. Constraints in high-priority scenarios have execution priority.

[0045] Step S402: Start the unified timing arrangement process for multi-device control instructions, plan the execution order of multi-device control instructions based on core constraints, and determine the trigger time node of each instruction; Based on core constraints, the global collaborative decision engine initiates a unified timing orchestration process for multi-device control commands. It calls the built-in timing scheduling algorithm to plan the execution order of the multi-device control commands after the adaptive linkage rules are decomposed. Following the principles of "high priority first, strong dependency first, and conflict-free parallel execution," it determines the execution order of each command. At the same time, it combines the response latency characteristics and communication transmission time of each subsystem device in the building to mark a precise trigger time node for each control command, forming a preliminary command timing execution plan. This ensures that command execution matches device response and avoids linkage failure due to timing deviations.

[0046] Step S403: Perform device control conflict verification on the planned multi-device control commands to eliminate the conflict problem of multiple commands being triggered in parallel on the same device; The global collaborative decision engine performs device control conflict verification on the initially formed instruction timing execution scheme. Through a conflict detection algorithm based on the unique address of the device, it traverses all target devices of control instructions to check for conflicts such as multiple parallel control instructions being issued to the same device at the same time node, or mutual exclusion operation instructions being issued to the same device. If a conflict is detected, the engine will automatically adjust the triggering time node of the conflicting instructions or remove invalid instructions based on the core constraints and linkage priorities until all device control conflicts are eliminated, forming a conflict-free instruction timing execution scheme.

[0047] Step S404: Encapsulate the verified multi-device control commands into command formats to generate execution commands that can be directly recognized by the terminal; The global collaborative decision engine encapsulates each control instruction in a conflict-free instruction timing execution scheme into a device-specific instruction format. Based on the target device's communication protocol (such as BACnet, KNX, Modbus) and instruction parsing rules, it converts the instructions into a binary or character instruction format that the device can directly recognize. At the same time, it adds a unique target device address identifier, instruction execution timestamp, and data checksum to each instruction to ensure that there is no loss or tampering during instruction transmission, and that the device terminal can quickly recognize and parse it.

[0048] Step S405: Distribute the encapsulated instructions according to the target device address and drive the corresponding device terminal to perform control operations.

[0049] The global collaborative decision engine distributes the encapsulated control commands to the corresponding device terminals through the industrial communication bus (such as RS485, Ethernet) or wireless communication module (such as LoRa, WiFi) of the building intelligent system, based on the target device address identifier of each command. For high real-time commands, they are directly distributed to edge computing nodes, which directly drive the devices to execute them. For ordinary commands, they are distributed to the device terminals through conventional communication links. After receiving the command, the device terminal completes the corresponding control operation based on the command execution timestamp and its own control program, realizing the collaborative linkage of multiple devices.

[0050] In one embodiment of the present invention, the specific steps for completing the iterative optimization of the adaptive linkage rule are as follows: Step S501: During the entire process of executing instructions on the device terminal, collect the running effect data after the instructions are executed in real time and transmit it to the data aggregation unit; Throughout the entire lifecycle of the control commands executed by the equipment terminal, real-time data on the full-dimensional operational effects after command execution is collected through hardware such as sensing nodes, equipment status acquisition modules, and energy consumption meters deployed within the building. The collection scope covers all dimensions corresponding to the linkage control objectives such as safety execution, energy consumption control, and comfort adjustment. The collected raw effect data is transmitted in real-time to the data aggregation unit deployed locally in the building via the building industrial communication bus or encrypted wireless communication link. This unit maintains a dedicated communication connection with the reinforcement learning model to ensure the real-time performance and integrity of data transmission.

[0051] Step S502: Quantify and transform the running effect data through the data aggregation unit, converting non-numerical data into numerical data that can participate in model calculation; The data aggregation unit incorporates a data quantification and conversion algorithm. First, it categorizes and filters the received operational performance data, distinguishing between numerical and non-numerical data. For non-numerical data such as personnel comfort feedback, equipment operation status descriptions, and scenario adaptability evaluations, it uses a standardized coding method combined with Likert quantification scoring to complete the numerical conversion. This transforms qualitatively descriptive non-numerical data into standardized values ​​that can participate in model calculations. The conversion process strictly adheres to the principle that "the degree of description is positively correlated with the magnitude of the value," ensuring that the quantified data accurately reflects the actual execution results. At the same time, it performs precision calibration on the numerical data to guarantee the computational adaptability of all data.

[0052] Step S503: The data aggregation unit sends the quantized performance data back to the input layer of the reinforcement learning model; The data aggregation unit encapsulates all operational performance data after quantization and accuracy calibration according to the input data format requirements of the reinforcement learning model, adds data type identifiers, collection timestamps, and scene feature labels, and transmits it back to the input layer of the reinforcement learning model in real time through the model's dedicated data interface. During the transmission process, a data block transmission mechanism is adopted to ensure that large-scale data transmission is lossless and delay-free, and the model can directly read and parse the data without additional format conversion.

[0053] Step S504: The reinforcement learning model calculates the execution deviation value of the current adaptive linkage rule based on the returned running effect data; After receiving the feedback execution effect data, the reinforcement learning model retrieves the pre-stored target values ​​for each dimension of linkage control within the model and compares the quantitative data of the actual execution effect with the preset target values ​​dimension by dimension. The weighted difference ratio algorithm is used to calculate the execution deviation value of a single dimension, and then the overall execution deviation value of the current adaptive linkage rule is calculated based on the weight coefficient of each dimension in the linkage control. The magnitude of the deviation value directly reflects the degree of matching between the current linkage rule and the actual scenario requirements. The larger the deviation value, the worse the rule adaptability.

[0054] Step S505: The reinforcement learning model adjusts its internal network parameters based on the execution bias value, corrects the logic and parameters of the adaptive linkage rules, and completes the rule iterative optimization.

[0055] Based on the calculated overall execution deviation, the reinforcement learning model initiates its built-in gradient descent optimization algorithm to adaptively adjust the core parameters of the neural network, such as weights and biases. The adjustment aims to reduce the execution deviation and improve rule adaptability. After the model parameters are adjusted, the revised adaptive linkage rules are regenerated based on the new parameter system. The revision process retains the well-adapted control logic in the original rules and optimizes only the logic and parameters with large deviations. Finally, the iterative optimization of the adaptive linkage rules is completed, and the new rules are transmitted to the global collaborative decision engine in real time for subsequent linkage control decisions, forming a closed-loop process of "rule execution - effect feedback - parameter adjustment - rule optimization".

[0056] It should be noted that when initiating the coordinated decision-making execution process, the optimal coordinated strategy is determined through a global coordinated decision utility function, which is as follows: , in, Representing the overall utility value of global linkage decision-making, it is a quantitative evaluation index of the comprehensive execution effect of candidate linkage strategies. The larger the value, the better the comprehensive performance of the linkage strategy in terms of safety, energy saving and comfort. This represents the weighting coefficient for the safety dimension, ranging from 0 to 1. The safety dimension weighting coefficient is a fixed preset value, a quantitative coefficient that measures the importance of the safety dimension in global collaborative decision-making. This fixed preset value reflects the decision-making principle that safety is the primary goal of building operation. During the initial configuration of the global collaborative decision-making engine, it is set by technical personnel according to national standards such as the "Design Standard for Intelligent Building Systems" and the "Code for Fire Protection of Buildings." Once set, it is fixed in the engine's decision rule module and will not be modified unless there are special engineering requirements. The set value range is 0.6 to 0.9 (the commonly used engineering value is 0.7, the highest weight among the three dimensions). For example, office buildings and commercial buildings are preset to 0.7, while densely populated buildings such as hospitals and schools are preset to 0.8.

[0057] The score represents the execution score of the safety dimension. The score value is determined based on the completeness of execution in fire protection and security emergency scenarios. It is a quantitative score of the completeness and effectiveness of the candidate linkage strategy in fire protection and security emergency scenarios. The higher the score, the stronger the strategy's ability to protect the safety of building life and assets. The engine extracts real-time data from the fire protection and security subsystems and quantifies the score based on the completeness of execution in emergency scenarios. That is, the proportion of the number of items covered by the candidate linkage strategy in the fire protection / security emergency process to the total number of items, combined with the normal operation rate of emergency equipment. The quantitative formula is: S = (number of emergency process coverage items / total number of items) * 100 * normal operation rate of equipment. The result range is 0-100 points. If the candidate strategy covers all emergency processes and the normal operation rate of equipment is 100%, then S = 100 points.

[0058] This represents the weighting coefficient for the energy-saving dimension, ranging from 0 to 1. This weighting coefficient is dynamically adjusted based on the building's operating hours and is a quantitative coefficient measuring the importance of energy saving in global collaborative decision-making. It is a dynamically adjusted value to adapt to the energy consumption control needs of different building operating hours. The global collaborative decision engine extracts the building's real-time operating hours and energy consumption control targets from a standardized global collaborative data set, automatically calculating and obtaining them based on preset time-weight mapping rules, updating every minute. Energy consumption control targets are higher during peak electricity hours (e.g., industrial peak electricity consumption 8:00-12:00). Set the value to 0.2~0.3; the energy consumption control target during off-peak hours is low. Take a value of 0.05~0.1; when there is no energy consumption control target, Take 0.1.

[0059] The score represents the energy-saving dimension. The score value is determined based on the reduction ratio of equipment energy consumption. It is a quantitative score of the reduction ratio of equipment energy consumption of candidate linkage strategies. The higher the score, the more significant the energy-saving effect of the strategy. The engine extracts real-time energy consumption data of building equipment, calculates the reduction ratio of equipment energy consumption after the implementation of candidate linkage strategies, and quantifies the score according to the ratio. The higher the energy consumption reduction ratio, the higher the score. Energy consumption reduction ratio ≥ 30%, E=100 points; 20% ≤ ratio < 30%, E=80-99 points; 10% ≤ ratio < 20%, E=60-79 points; ratio < 10%, E=0-59 points. The comfort dimension weight coefficient, ranging from 0 to 1, is dynamically adjusted based on the occupancy status. It is a quantitative coefficient measuring the importance of the comfort dimension in global collaborative decision-making, and is a dynamically adjusted value to adapt to the comfort needs of different occupancy statuses within the building. The global collaborative decision-making engine extracts the occupancy density of each area of ​​the building from a standardized global collaborative data set, and automatically calculates it based on preset occupancy density-weight mapping rules, updating every 30 seconds. When the occupancy density is ≥80%, When the personnel density is 10% to 80%, take a value of 0.2 to 0.3. Take a value of 0.1~0.2; when the personnel density is ≤10%, Take a value of 0.05 to 0.1.

[0060] satisfy .

[0061] The score represents the comfort dimension, and the score value is determined based on the matching degree between environmental parameters and human comfort standards. It is a quantitative score of the matching degree between building environmental parameters and human comfort standards under candidate linkage strategies. The higher the score, the better the environmental comfort experience can be provided for people in the building. The engine extracts real-time data such as temperature, humidity, illuminance, and CO2 concentration from the environmental monitoring subsystem and calculates the matching degree between environmental parameters and human comfort standards. The matching degree is the proportion of environmental parameters that meet human comfort standards to the total number of parameters. The quantitative formula is: C = (number of parameters that meet comfort standards / total number of parameters) * 100. Human comfort standards follow the "Indoor Air Quality Standard" and the "Design Code for Heating, Ventilation and Air Conditioning of Civil Buildings". For example, an indoor temperature of 24~26℃ and an illuminance of 300~500lx in summer are considered to meet the standards. The value range is 0-100 points. If all environmental parameters meet human comfort standards, then C = 100 points.

[0062] The global collaborative decision engine selects the linkage strategy with the highest total utility value of global linkage decision as the final execution strategy.

[0063] The global linkage decision utility function is the core quantitative judgment model for multi-device linkage decision-making in building intelligence. It is applied to scenarios where, after the global collaborative decision engine starts the linkage decision execution process, it needs to select the optimal strategy from multiple candidate linkage strategies after receiving a standardized global linkage data set and adaptive linkage rules.

[0064] The core logic of the global linkage decision utility function is as follows: the three core evaluation dimensions of building linkage control—safety, energy saving, and comfort—are quantitatively scored and weighted to obtain the total utility value of global linkage decision. The global collaborative decision engine selects the candidate linkage strategy with the largest total utility value as the final execution strategy. Among them, the safety dimension is the core priority dimension, and a fixed weight is used to ensure the safety of the decision. The energy saving and comfort dimensions adopt dynamic weights to adapt to the real-time operation scenario of the building, realizing dynamic collaborative optimization of multiple objectives. The formula calculation process is completed by the numerical calculation module built into the global collaborative decision engine, with a calculation response time of ≤10ms, which meets the real-time requirements of linkage decision.

[0065] It should be noted that when the global collaborative decision engine performs unified timing orchestration of control commands for multiple devices, it calculates the optimal execution timing through the linkage command timing scheduling cost function. The linkage command timing scheduling cost function is as follows: , in, The total cost of the linkage instruction timing scheduling is a quantitative evaluation index of the comprehensive execution rationality of the candidate timing scheme. The smaller the value, the more reasonable the instruction execution interval of the timing scheme, the fewer the equipment control conflicts, and the higher the optimality of the timing arrangement. Represents the trigger execution time of the i-th control instruction, which is the planned start execution time of a single instruction in the timing scheme. It is the basic physical quantity for measuring the instruction execution timing interval. Represents the trigger execution time of the (i+1)th control instruction, which is the start execution time of the next instruction adjacent to the i-th control instruction; The penalty coefficient for instruction execution conflict is a constant with a value greater than 0. It is a fixed quantitative coefficient that weights the cost of a single backup control conflict, amplifying the impact of the conflict on the total timing cost and reflecting the principle of avoiding equipment control conflicts. During the initial configuration of the global collaborative decision engine, it is set by technical personnel based on the equipment control characteristics and conflict avoidance requirements of the building's intelligent system. Once set, it is fixed in the engine's timing orchestration rule module and is not modified unless there are special engineering requirements. Based on practical engineering experience in building equipment control, the set value range is 100~500ms. This range allows the penalty cost of a single conflict to be significantly higher than the conventional instruction timing interval, effectively avoiding control conflicts in the timing scheme. Specifically, a higher value (λ=400-500ms) is used for buildings with high control precision requirements (such as hospitals and data centers); a middle value (λ=200-300ms) is used for conventional office and commercial buildings; a lower value (λ=100-200ms) is used for small civil buildings; and the general engineering set value is λ=300ms.

[0066] It represents the number of device control conflicts that occur during instruction execution; it is the number of times the same device is planned to receive and execute multiple control instructions simultaneously in the candidate timing scheme, and is the core indicator for measuring the degree of conflict in the timing scheme. The global collaborative decision engine selects the timing scheme with the minimum total cost of linkage instruction timing scheduling as the final execution timing of multi-device control instructions.

[0067] The core logic of the linkage instruction timing scheduling cost function is as follows: the total timing interval of the multi-device control instructions is summed with the conflict penalty value to obtain the total cost T of the linkage instruction timing scheduling. The global collaborative decision engine selects the candidate timing scheme with the minimum total cost as the final execution timing of the multi-device control instructions; whereby... The basic cost of natural sequential execution of instructions. To address the penalty cost of device control conflicts, a fixed penalty coefficient is set to weight the cost of conflict scenarios, forcing timing schemes to avoid control conflicts. The formula calculation process is completed by the numerical calculation module built into the global collaborative decision engine, with a calculation response time ≤5ms, meeting the real-time requirements of multi-device linkage instruction timing arrangement. In the formula, the instruction number i is a positive integer, starting from 1 and ending at the total number of control instructions involved in the arrangement. If there are n control instructions, then i = 1, 2, 3...n-1.

[0068] In one embodiment of the present invention, a distributed execution architecture is adopted when the instruction is distributed to the corresponding device terminal for execution, and the specific execution process is as follows: Step S601: The global collaborative decision engine determines the real-time level of the instructions and marks control instructions with real-time requirements higher than the preset threshold as high real-time instructions. The global collaborative decision engine incorporates a real-time quantitative judgment algorithm for commands, pre-setting real-time thresholds for control commands of the building intelligent system. These thresholds are formulated based on the response requirements of different linkage scenarios in the building (e.g., the real-time threshold for fire and security emergency commands is 50 milliseconds, while daily energy-saving optimization commands do not have strict low-latency requirements). The engine performs scenario matching for each encapsulated control command, and combines the preset thresholds to determine the real-time level. Commands with real-time requirements higher than the threshold are marked as high real-time commands, while the rest are marked as non-real-time global optimization commands, and unique level identifiers are added to commands of different levels.

[0069] Step S602: The global collaborative decision engine sends high real-time instructions to the edge computing nodes, and the edge computing nodes directly establish communication connections with the corresponding device terminals. The global collaborative decision engine uses industrial-grade low-latency communication protocols (such as Modbus-RTU and Profinet) to direct control information marked as high real-time commands to edge computing nodes deployed on the building site. The edge computing nodes are distributed according to the building's functional areas (such as office areas and fire compartments). Each node establishes a direct communication link with the device terminals in its area. When the engine issues commands, it also carries the physical address and execution priority of the target device to ensure that the edge computing nodes accurately receive the commands.

[0070] Step S603: After receiving the instruction, the edge computing node directly drives the device terminal to execute it, and the instruction execution response time is controlled within a preset range; After receiving a high real-time instruction, the edge computing node sends an execution signal directly to the target device terminal through a direct communication link, bypassing the cloud forwarding link and strictly controlling the instruction execution response time within a preset low latency range. During execution, the edge computing node collects instruction execution status data from the device terminal in real time, completes local temporary storage of the execution results, and simultaneously sends instruction execution receipts back to the global collaborative decision engine to ensure that the engine has a grasp of the execution status of the high real-time instruction.

[0071] Step S604: The global collaborative decision engine uploads the non-real-time global optimization instructions to the cloud server, and the cloud server performs global data statistics and long-term strategy optimization operations. The global collaborative decision engine uploads control information marked as non-real-time global optimization instructions to the cloud server via wireless / wired wide area communication networks. The cloud server is equipped with a building intelligence global optimization analysis module, a big data statistics module, and a long-term strategy training module. After receiving the instructions, it combines the building's full life cycle operation data to complete non-real-time operations such as global energy consumption optimization, equipment operation efficiency analysis, and long-term linkage strategy generation. The generated global optimization results are then sent back to the global collaborative decision engine to provide data support for the subsequent generation of linkage rules.

[0072] Step S605: The edge computing node maintains real-time data synchronization with the cloud server. When the network connection is interrupted, the edge computing node starts local independent operation mode and stores the execution data. After the network is restored, the stored data is automatically synchronized to the cloud server.

[0073] Edge computing nodes and cloud servers maintain real-time data interaction through heartbeat detection and incremental data synchronization. Edge computing nodes upload locally stored instruction execution status and device operation data to the cloud server at preset time intervals. The cloud server distributes global optimization data and policy update data to the edge computing nodes to ensure data consistency between the edge and the cloud. When a network connection interruption is detected, the edge computing node immediately starts local independent operation mode, saving subsequent instruction execution data and device operation data through its built-in local storage module. After the network connection is restored, all data stored during the network outage is automatically synchronized to the cloud server, achieving no data loss and uninterrupted execution.

[0074] In one embodiment of the present invention, abnormal operating condition protection operations are performed throughout the entire process of adaptive linkage rule iterative optimization. The specific steps are as follows: Step S701: The global collaborative decision engine monitors the data acquisition status, network connection status, and device operation status in real time. Throughout the entire process of adaptive linkage rule iteration and optimization, the global collaborative decision engine uses a built-in multi-dimensional status monitoring module to perform high-frequency and high-precision real-time monitoring of three core states of the building intelligent system. The data acquisition status is determined by the heartbeat packet feedback of the sensing nodes, the data upload frequency, and the data missing rate. The network connection status is detected by the packet loss rate, transmission latency, and link connectivity of the communication link. The equipment operation status is obtained by the matching degree between the rated parameters and real-time operating parameters of the equipment obtained by the PLC and status acquisition module of each subsystem. The monitoring data is transmitted back to the status monitoring module in real time at preset time intervals. The module performs quantitative processing on all monitoring data to provide quantitative basis for subsequent abnormal condition judgment.

[0075] Step S702: When the global collaborative decision engine detects abnormal operating conditions such as missing data collection, network fluctuation interruption, or equipment failure and offline status, it terminates the execution process of the current adaptive linkage rule. The global collaborative decision-making engine's status monitoring module has a built-in abnormal condition quantitative judgment algorithm. It pre-sets abnormal judgment thresholds for three types of states: data acquisition, network connection, and equipment operation. When the monitored data exceeds the preset threshold (such as a data acquisition missing rate exceeding 10%, network transmission latency exceeding 200ms, or equipment operating parameters exceeding the rated value by 15%), the system is immediately judged to have entered an abnormal condition and an abnormal alarm signal is triggered. After receiving the abnormal alarm signal, the engine's logic control module immediately terminates the execution process of the current adaptive linkage rule through a hard interrupt and freezes all linkage control commands to be issued, avoiding problems such as equipment loss of control and safety risks caused by the execution of erroneous commands under abnormal conditions.

[0076] Step S703: The global collaborative decision engine calls the preset safety benchmark linkage strategy to prioritize driving fire protection, security and emergency evacuation related equipment to perform control operations; The global collaborative decision-making engine's local storage module pre-defines the security baseline linkage strategy. This strategy serves as the basic safety fallback for the building's intelligent system and is formulated only for core equipment related to fire protection, security, and emergency evacuation. When the system enters an abnormal operating condition, the engine immediately retrieves this strategy from the local storage module, skips the reasoning and generation of adaptive linkage rules, and directly issues control commands to core safety equipment such as fire exhaust equipment, emergency lighting equipment, evacuation passage access control, and security monitoring equipment according to the preset safety execution priority. This prioritizes driving the core safety equipment to perform basic safety operations according to standard specifications, ensuring the safety of building life and assets.

[0077] Step S704: The global collaborative decision engine suspends the execution process of unnecessary energy-saving adjustments and comfort optimization commands to ensure the stable operation of the building's core safety functions; While invoking the security baseline linkage strategy, the global collaborative decision engine, through its built-in resource scheduling module, reallocates the hardware computing resources (CPU, memory) and communication transmission resources (bandwidth, execution channels) of the building intelligent system. It immediately suspends the execution of non-essential control commands such as energy-saving adjustment and environmental comfort optimization, releases all resources occupied by such commands, and allocates the released resources and the remaining system resources to the execution of control commands of core security devices. This avoids resource preemption by non-essential commands, which could lead to delays or failures in the execution of core security commands, and ensures the stable operation of core security functions.

[0078] Step S705: After the abnormal operating conditions are eliminated, the global collaborative decision engine automatically resumes the adaptive linkage rule execution mode. At the same time, it sends the abnormal operating condition data and the fallback strategy execution data back to the reinforcement learning model to supplement the model training samples and improve the model's adaptability to abnormal operating conditions.

[0079] The global collaborative decision-making engine's status monitoring module continuously monitors the system status. When the monitored data recovers to within the preset normal threshold range, it determines that the abnormal condition has been eliminated and triggers a recovery signal. After receiving the recovery signal, the engine's logic control module automatically terminates the execution of the safety baseline linkage strategy and switches back to the original adaptive linkage rule execution mode to continue iterative optimization of the linkage rules. At the same time, the engine integrates and encapsulates abnormal condition data such as the type, duration, abnormal trigger threshold, and abnormal equipment parameters, as well as fallback strategy execution data such as the execution time, instruction execution effect, and resource allocation of the safety baseline linkage strategy. This data is then transmitted back to the reinforcement learning model's sample training library via incremental data transmission to supplement the model's abnormal condition training samples.

[0080] This invention addresses the technical shortcomings of existing building intelligent systems' multi-device linkage control, including fixed linkage rules lacking self-learning and self-iteration capabilities, poor scenario adaptability, lack of a cross-subsystem global collaborative decision-making architecture, absence of a unified linkage priority mechanism, susceptibility to command conflicts and timing errors, and lack of quantitative judgment standards for linkage decisions and command orchestration. The invention provides a comprehensive solution through the aforementioned end-to-end technical approach. First, it standardizes multi-source heterogeneous operational data across the entire building domain, laying a unified data foundation for global collaboration. Then, it builds a cross-subsystem global collaborative decision-making engine and establishes a multi-level linkage execution priority mechanism. Combined with a global linkage decision utility function, it achieves quantitative judgment of linkage decisions, fundamentally breaking down the linkage silos between subsystems and avoiding resource contention and command interference issues arising from linkage requests in different scenarios. Simultaneously, it extracts features from standardized data using a reinforcement learning model and dynamically generates adaptive linkage rules without manual pre-setting. Combined with the effect data after command execution, it iterative optimization of the rules is completed, thoroughly resolving the issue. By abandoning the method of manually pre-setting fixed rules, the linkage rules can learn and correct themselves according to the dynamic changes in the building environment, personnel, and equipment conditions, which greatly improves the scene adaptability. It also realizes the quantitative optimal judgment of instruction arrangement through the linkage instruction timing scheduling cost function, and ensures the orderly execution of instructions by conflict verification and targeted distribution. The distributed execution architecture ensures the efficient execution of instructions with different real-time requirements and the continuity of execution when the network is interrupted. The abnormal working condition protection operation prioritizes the protection of core security functions when the system experiences data, network, or equipment anomalies. It can also supplement model training samples to improve the robustness of the system. Each step is connected layer by layer to form a complete technical closed loop. Ultimately, it realizes cross-subsystem full-domain collaborative linkage of multiple devices in the building intelligent system, taking into account the multi-objective optimization of building operation safety, energy saving, and comfort. It solves the problem of the lack of quantitative judgment standards in existing technologies and greatly improves the adaptive capability, precision level, and operational reliability of multi-device linkage control in the building intelligent system.

[0081] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for multi-device linkage control in a building intelligent system, characterized in that, Includes the following steps: Collect multi-source heterogeneous operational data from the entire building interior, perform unified format conversion and normalization on the collected data, and form a standardized global linkage data set; A cross-subsystem global collaborative decision-making engine is built based on a standardized global linkage data set, and a multi-level linkage execution priority mechanism is formulated to clarify the execution permissions of different linkage scenarios; By using a reinforcement learning model to perform feature matching and model inference on a standardized global linkage data set, adaptive linkage rules that fit the current operating scenario are dynamically generated. The global collaborative decision engine arranges the control commands of multiple devices in a unified time sequence based on adaptive linkage rules and multi-level linkage execution priorities, and distributes the arranged commands to the corresponding device terminals for execution. Collect the execution effect data after the command is executed and send it back to the reinforcement learning model to complete the iterative optimization of the adaptive linkage rules.

2. The multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, The specific steps for forming a standardized, globally linked data set are as follows: Collect building interior environmental monitoring data, personnel status data, equipment operation status data, energy consumption data, security monitoring data, and fire early warning data point by point, and transmit the collected raw data to the local data preprocessing unit; The local data preprocessing unit performs outlier removal and noise filtering on the raw data to remove invalid data that exceeds the preset value range. The processed data is uniformly converted to a transmission format to eliminate data format differences caused by communication protocols of different subsystems. The data after the format is unified is mapped to numerical dimension normalization, which transforms data with different dimensions into the same numerical range. The processed data is integrated and packaged to form a standardized global linkage dataset, which is then merged and transmitted to the global collaborative decision-making engine.

3. The multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, The specific steps for building a cross-subsystem global collaborative decision-making engine are as follows: Based on the storage location and data type of the standardized global linkage data set, a global collaborative decision engine is built across HVAC, lighting, fire protection, security, elevator and access control subsystems; The global collaborative decision-making engine determines the priority of linkages based on the impact of linkage scenarios on building operation safety. Based on the judgment dimensions, the priority of the linkage execution is divided into multiple levels, and the linkage scenario type corresponding to each priority level is locked; Configure an independent instruction execution channel for each priority level and limit the resource consumption limit of linked requests of different priorities; The global collaborative decision engine establishes rules for determining the permission of linkage requests, determines the instruction execution permissions for different linkage scenarios based on priority levels, and receives standardized global linkage data sets and adaptive linkage rules to initiate the linkage decision execution process.

4. The multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, The specific steps for dynamically generating adaptive linkage rules that fit the current running scenario are as follows: Reinforcement learning models are used to extract environmental features, personnel features, equipment features, energy consumption features, and scenario features from a standardized global interconnected dataset. The extracted multidimensional features are input into a preset neural network structure for forward inference calculation; The system calculates and outputs the linkage trigger threshold, linkage device combination, and linkage operation parameters that are adapted to the current scenario. Based on the linkage trigger threshold, linkage device combination, and linkage operation parameters, control logic that does not match the current scenario is eliminated, and effective control logic is retained. Effective control logic is combined and encapsulated to generate adaptive linkage rules without human pre-setting and transmit them to the global collaborative decision engine.

5. The multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, The specific steps for timing the control commands for multiple devices and distributing them to the device terminals for execution are as follows: The global collaborative decision engine receives adaptive linkage rules and multi-level linkage execution priority information to determine the core constraints for instruction execution. Initiate a unified timing arrangement process for multi-device control commands, plan the execution order of multi-device control commands based on core constraints, and determine the trigger time node for each command; Perform device control conflict verification on the planned multi-device control commands to eliminate the conflict problem of multiple commands being triggered in parallel on the same device; The verified multi-device control commands are encapsulated in command format to generate execution commands that can be directly recognized by the terminal. The encapsulated instructions are distributed according to the target device address, driving the corresponding device terminal to perform control operations.

6. The multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, The specific steps for completing the iterative optimization of the adaptive linkage rule are as follows: Throughout the entire process of executing instructions on the device terminal, real-time data on the operational effects after instruction execution is collected and transmitted to the data aggregation unit. The data aggregation unit quantifies and transforms the operational performance data, converting non-numerical data into numerical data that can participate in model calculations. The data aggregation unit feeds back the quantified performance data to the input layer of the reinforcement learning model; The reinforcement learning model calculates the execution deviation value of the current adaptive linkage rule based on the returned execution effect data; The reinforcement learning model adjusts its internal network parameters based on the execution bias value, corrects the logic and parameters of the adaptive linkage rules, and completes the iterative optimization of the rules.

7. The multi-device linkage control method for a building intelligent system according to claim 3, characterized in that, When initiating the coordinated decision-making execution process, the optimal coordinated strategy is determined through the global coordinated decision-making utility function, which is as follows: ， in, Represents the total utility value of globally coordinated decision-making. This represents the weighting coefficient for the security dimension, with a value ranging from 0 to 1. The weighting coefficient for the security dimension is a fixed preset value. The score represents the safety dimension of the performance evaluation. The score value is determined based on the completeness of the performance evaluation in fire protection and security emergency scenarios. This represents the weighting coefficient for the energy-saving dimension, with a value ranging from 0 to 1. This weighting coefficient is dynamically adjusted based on the building's operating time. The score represents the energy-saving dimension, and the score value is determined based on the percentage reduction in equipment energy consumption. This represents the weighting coefficient for the comfort dimension, with a value ranging from 0 to 1. The weighting coefficient for the comfort dimension is dynamically adjusted based on the occupancy status of the personnel. The score represents the comfort dimension, and the score value is determined based on the degree of matching between environmental parameters and human comfort standards. The global collaborative decision engine selects the linkage strategy with the highest total utility value of global linkage decision as the final execution strategy.

8. A multi-device linkage control method for a building intelligent system according to claim 5, characterized in that, When the global collaborative decision engine performs unified timing orchestration of control commands for multiple devices, it calculates the optimal execution timing using the linkage command timing scheduling cost function. The linkage command timing scheduling cost function is as follows: ， in, This represents the total cost of scheduling the linkage instructions. This represents the trigger execution time of the i-th control instruction. This represents the trigger execution time of the (i+1)th control instruction. This represents the instruction execution conflict penalty coefficient, a constant with a value greater than 0. The number of device control conflicts that occur during instruction execution; The global collaborative decision engine selects the timing scheme with the minimum total cost of linkage instruction timing scheduling as the final execution timing of multi-device control instructions.

9. A multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, When instructions are distributed to the corresponding device terminals for execution, a distributed execution architecture is adopted. The specific execution process is as follows: The global collaborative decision engine determines the real-time level of instructions and marks control instructions with real-time requirements higher than a preset threshold as high real-time instructions. The global collaborative decision-making engine sends high real-time instructions to edge computing nodes, and the edge computing nodes directly establish communication connections with the corresponding device terminals. After receiving instructions, the edge computing node directly drives the device terminal to execute them, and the instruction execution response time is controlled within a preset range. The global collaborative decision-making engine uploads non-real-time global optimization instructions to the cloud server, where the cloud server performs global data statistics and long-term strategy optimization operations. Edge computing nodes maintain real-time data synchronization with cloud servers. When the network connection is interrupted, the edge computing node starts local independent operation mode and stores the execution data. After the network is restored, the stored data is automatically synchronized to the cloud server.

10. A multi-device linkage control method for a building intelligent system according to claim 1, characterized in that, Throughout the entire process of adaptive linkage rule iterative optimization, abnormal operating condition protection operations are performed. The specific steps are as follows: The global collaborative decision-making engine monitors data acquisition status, network connection status, and device operation status in real time. When the global collaborative decision engine detects abnormal operating conditions such as missing data collection, network fluctuations and interruptions, or equipment failure and offline status, it terminates the execution process of the current adaptive linkage rule. The global collaborative decision engine invokes preset safety benchmark linkage strategies to prioritize driving fire protection, security, and emergency evacuation related equipment to perform control operations. The global collaborative decision-making engine suspends the execution of unnecessary energy-saving adjustments and comfort optimization commands to ensure the stable operation of the building's core safety functions; After the abnormal operating conditions are eliminated, the global collaborative decision engine automatically resumes the adaptive linkage rule execution mode. At the same time, it sends the abnormal operating condition data and the backup strategy execution data back to the reinforcement learning model to supplement the model training samples and improve the model's ability to adapt to abnormal operating conditions.

Images