Multi-device cooperative control method based on AI in Internet of Things environment

Abstract

The invention discloses an AI-based multi-device cooperative control method in an Internet of Things environment. By constructing a'perception-decision-control-feedback-self-adaption 'full-closed-loop architecture, terminal components such as an intelligent equipment group and a sensor group of an equipment layer, a multi-source data acquisition and enhancement preprocessing module of a perception layer, a dynamic equipment capability graph of an AI decision layer, an incremental AI decision model and other core units are integrated; and a cross-protocol instruction distribution and closed-loop feedback module of the control layer realizes deep processing of multi-source heterogeneous data, real-time perception of equipment capability, dynamic adjustment of task priority and autonomous optimization of a coordination strategy, thoroughly solves the technical problems of rigid coordination logic, response lag and unreasonable resource allocation of a traditional method, and improves the reliability of the system. And the intelligent level and the operation efficiency of the Internet of Things system are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, specifically to an AI-based multi-device collaborative control method in an IoT environment. Background Technology

[0002] With the widespread adoption of IoT technology, the number of smart devices in smart homes, industrial IoT, smart parks, and other scenarios is exploding. These devices encompass sensors, actuators, smart terminals, and industrial control equipment, forming a complex ecosystem of interconnected multi-device systems. Multi-device collaborative control, as a core function of IoT systems, directly determines the system's operating efficiency, resource utilization, user experience, and security stability, making it a crucial support for realizing intelligent applications in IoT scenarios.

[0003] However, existing IoT multi-device collaborative control methods still have many technical shortcomings and are difficult to meet the actual needs of complex scenarios: 1. Lack of closed-loop collaborative architecture: Traditional methods often adopt a unidirectional linear architecture of "collection-decision-execution", which lacks a real-time feedback and dynamic adjustment mechanism for the execution results of the equipment. This makes it impossible to form a complete closed loop, resulting in a disconnect between the control strategy and the actual needs of the scenario, leading to problems such as response lag and insufficient control accuracy. Furthermore, no adaptive handling mechanism is designed for abnormal scenarios such as equipment failure and command execution failure, resulting in poor system fault tolerance.

[0004] 2. Weak data processing capabilities: Existing technologies mostly use simple data cleaning and normalization methods, which cannot effectively handle random and systematic outliers in multi-source heterogeneous data, and lack in-depth mining of multi-dimensional features in the time domain, frequency domain and space, resulting in low credibility of decision data, which in turn affects the accuracy of collaborative control strategies.

[0005] 3. Rigid equipment capability perception: Relying on static equipment information tables to record equipment parameters can only reflect the basic attributes of the equipment when it leaves the factory. It cannot update the equipment's operating status (such as fault, standby), resource usage (such as computing resources, communication bandwidth), and collaborative adaptation history in real time. This leads to a mismatch between the equipment capability description and the actual status, resulting in resource allocation imbalance problems such as "excessive capacity" or "insufficient capacity".

[0006] 4. Poor adaptability of AI models: Traditional AI decision-making models are mostly deployed after static training, lacking incremental learning capabilities. When the equipment in the system is updated, the application scenario changes, or the user needs are upgraded, the entire model needs to be retrained, which is time-consuming, laborious, and seriously affects the continuity of system operation. At the same time, the models mostly use single feature fusion methods and simple optimization algorithms, which make it difficult to accurately capture the correlation between equipment, environment, and user needs, resulting in poor optimization effect of collaborative strategies.

[0007] 5. Protocol compatibility and distribution limitations: Some methods only support a single or a few IoT communication protocols. When there are multiple protocol devices such as MQTT, HTTP, CoAP, and Modbus in the system, an additional protocol conversion gateway needs to be deployed, which increases system complexity and communication latency. In addition, the instruction distribution adopts a unified transmission strategy and does not differentiate between transmission reliability and latency requirements based on task priority, resulting in untimely response to high-priority urgent tasks.

[0008] The aforementioned problems result in traditional multi-device collaborative control methods exhibiting low levels of intelligence, poor flexibility, and insufficient scalability, failing to adapt to the dynamic and changing needs of complex IoT scenarios. This severely restricts the application, promotion, and performance improvement of IoT systems. Therefore, there is an urgent need for a multi-device collaborative control method with a fully closed-loop architecture, deep processing of multi-source data, real-time perception of device capabilities, incremental optimization of AI models, cross-protocol compatibility, and adaptive anomaly handling capabilities to address the shortcomings of existing technologies. Summary of the Invention

[0009] To address the problems mentioned in the background art, the present invention aims to provide an AI-based multi-device collaborative control method in an Internet of Things (IoT) environment. This method addresses the issues of low intelligence level, poor flexibility, insufficient scalability, inability to adapt to the dynamic changes in complex IoT scenarios, and the serious constraints on the application promotion and efficiency improvement of IoT systems.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a multi-device collaborative control method based on AI in an Internet of Things (IoT) environment: Includes the following steps: S1. Multi-source heterogeneous data acquisition and enhanced preprocessing: Through the multi-type sensors and device interfaces of the IoT perception layer, collect full-state data of device operation, multi-dimensional environmental parameter data and user interaction intent data. Perform intelligent outlier removal, adaptive missing value filling, multi-dimensional feature enhancement and standardization processing on the raw data to obtain highly reliable structured data. S2. Dynamic Equipment Capability Map Construction and Update: During the initialization phase, the equipment's factory parameters, function list, communication protocol specifications, and operational constraint thresholds are collected to construct a basic capability map. During operation, based on real-time equipment status data, fault feedback information, and resource usage, the equipment capability description, available resource quantification value, and collaborative adaptation coefficient of the map are dynamically optimized through an incremental update algorithm. S3. Hierarchical Task Parsing and Dynamic Priority Determination: Natural language processing technology is used to parse the target parameters, execution constraints, and time requirements of user instructions and system-triggered tasks. By integrating subjective and objective weights through an improved analytic hierarchy process (AHP-entropy weight method), the task priority is dynamically determined. The priority is adjusted in real time according to the device status, the urgency of the environment, and the user's behavior habits. S4. Incremental AI Decision Model Training and Deployment: Construct a deep learning model that includes a feature fusion submodule, a task matching submodule, and a collaborative strategy optimization submodule. Use historical structured data to complete the initial training, continuously absorb new scenario data through incremental learning algorithms to optimize the model parameters, and deploy the trained model to the AI ​​decision layer to support real-time decision-making. S5. Dynamic collaborative control strategy generation: The AI ​​decision layer receives preprocessed real-time data, combines it with dynamic equipment capability map and real-time task priority, and analyzes equipment collaborative needs, resource constraints and optimization goals through trained AI models to generate the optimal collaborative control instruction set adapted to the current scenario. S6. Cross-protocol command distribution and closed-loop execution: The control layer uses an adaptive communication interface to be compatible with mainstream IoT protocols such as MQTT, HTTP, CoAP, and Modbus, accurately distributing control command sets to target devices. After the device executes the command, it provides real-time feedback on the execution status, result data, and resource consumption to the AI ​​decision layer. The decision layer compares the execution result with the preset target and dynamically adjusts the control strategy to form a closed-loop control.

[0011] As a preferred embodiment of the present invention, S1-1: Collect multi-source heterogeneous data, wherein the full-state data of equipment operation includes voltage, current, operating temperature, load rate, working mode, remaining power / energy consumption, fault codes, and response delay; the multi-dimensional environmental parameter data includes temperature, humidity, light intensity, and air quality (PM2.5, ... Concentration), noise level, vibration frequency, and spatial personnel density; user interaction intent data includes voice commands, text commands, button operations, gesture commands, and usage habit preference data; S1-2: Intelligent outlier removal. An improved Z-score algorithm combined with an isolated forest model is used. First, the Z-score algorithm is used to detect and remove random outliers that are more than 3 times the standard deviation. Then, the isolated forest model is used to identify systematic outliers caused by equipment failures and sensor malfunctions. S1-3: Adaptive imputation of missing values. For continuous data, a time-series prediction imputation method based on LSTM is used, and for discrete data, a probabilistic imputation method based on Bayesian network is used. The imputation process combines the historical operating patterns of the equipment with the environmental correlation characteristics. S1-4: Multi-dimensional feature enhancement, constructing a multi-dimensional feature set through time-domain feature extraction (sliding window mean, variance, peak value), frequency-domain feature extraction (spectral peak value after Fourier transform), and spatial feature extraction (correlation coefficient of multi-sensor data); S1-5: Standardization processing, using an improved Min-Max normalization algorithm, dynamically adjusting the normalization interval based on data distribution characteristics, eliminating dimensional differences while preserving the data distribution characteristics, resulting in structured data.

[0012] As a preferred embodiment of the present invention, S2-1: Construct a basic capability map, including a basic equipment information layer, storing static information such as equipment model, factory parameters, function type, communication protocol type, rated power, and operating environment threshold; S2-2: Real-time status layer update, based on the device's full operating status data, to update dynamic information such as the device's current working mode, operating status (normal / fault / standby), current load, and remaining power / energy consumption in real time; S2-3: Resource quantization layer construction, which uses a resource quantization model to convert the device's computing resources, storage resources, communication bandwidth, runtime, etc. into quantifiable values ​​in the range [0,1]. S2-4: Collaboration adaptation layer record, storing data such as the collaboration history of the storage device with other devices, adaptation success rate, collaboration latency, and energy consumption collaboration coefficient; S2-5: Dynamic update mechanism, which adopts a sliding time window mechanism, updates the information of each layer of the map every 1-5 seconds based on the latest collected data to ensure that the map is consistent with the actual state of the device.

[0013] As a preferred embodiment of the present invention, S3-1: hierarchical task parsing, extracting task target parameters, execution constraints and time requirements through the BERT pre-trained model, and converting natural language instructions into structured parameters, wherein the time requirements are divided into immediate execution tasks (response time ≤ 1 second), timed execution tasks and condition-triggered tasks. S3-2: Subjective weighting is determined using the Analytic Hierarchy Process (AHP) to determine subjective weights based on the importance of user needs and task type (safety / comfort / efficiency). S3-3: Objective weight calculation, using the entropy weight method to determine objective weights based on objective data such as equipment resource utilization rate and environmental urgency. S3-4: Dynamic priority determination, which combines subjective and objective weights to obtain a comprehensive priority. The priority value ranges from [0,10] and is adjusted in real time according to changes in equipment status and environment.

[0014] As a preferred embodiment of the present invention, S4-1: the feature fusion submodule is constructed by using a multi-head attention mechanism combined with a gated recurrent unit (GRU) to dynamically weight and fuse equipment operation features, environmental features and user demand features, and output a fused feature vector. S4-2: Task matching submodule design, based on fused feature vectors, uses deep neural networks (DNN) to calculate the matching degree between tasks and the capabilities of each device, and selects a set of candidate devices that meet the constraints; S4-3: Construction of the cooperative strategy optimization submodule. An improved deep reinforcement learning (DRL) algorithm is adopted. The multi-objective optimization function is to maximize system operating efficiency, minimize energy consumption, and maximize task completion rate. A reward mechanism is constructed to generate the optimal cooperative control strategy. S4-4: Incremental learning training, using the Elastic Weight Consolidation (EWC) algorithm, protects learned historical knowledge while optimizing the model by absorbing new data, avoiding catastrophic forgetting; S4-5: Model deployment, deploying the trained model to the AI ​​decision layer (edge ​​gateway or cloud server) for real-time decision-making.

[0015] As a preferred embodiment of the present invention, S5-1: Task-equipment matching, based on task constraints and equipment capability parameters in the dynamic equipment capability map, calculates the matching degree through the task matching submodule and filters out a set of candidate equipment. S5-2: Resource optimization and allocation. Based on the resource quantification values ​​of candidate devices and the resource requirements of the task, a resource allocation algorithm is used to decompose the task into several sub-tasks and allocate them to the corresponding devices. S5-3: Cooperative logic generation. Combines the historical cooperative data of the device cooperative adaptation layer to generate cooperative logic and control parameters between devices, forming a complete cooperative control instruction set.

[0016] As a preferred embodiment of the present invention, S6-1: adaptive protocol identification, the control layer has a built-in protocol identification module that automatically identifies the communication protocol type of the target device through handshake signals, protocol port numbers and data frame formats; S6-2: Command format conversion, based on a preset protocol command mapping library, converts standardized control commands into protocol formats supported by the target device; S6-3: Differentiated distribution strategy: High-priority tasks (priority ≥ 8) use UDP protocol to achieve low-latency transmission, while ordinary priority tasks use TCP protocol to ensure transmission reliability; S6-4: Closed-loop execution adjustment. The AI ​​decision-making layer calculates the deviation value between the execution result and the preset target through the deviation analysis model. When the deviation value > the preset threshold, the collaborative strategy generation process is re-triggered until the deviation value meets the requirements.

[0017] As a preferred embodiment of the present invention, the abnormal collaborative adaptive processing step is as follows: S7-1: When the equipment reports fault information or the control command fails to execute, the AI ​​decision-making layer quickly selects alternative equipment based on the dynamic equipment capability map; S7-2: Generate backup collaborative control strategies and distribute them for execution; S7-3: Record faulty equipment information, alternative solutions, and execution results to provide data support for subsequent equipment maintenance and collaborative logic optimization.

[0018] As a preferred embodiment of the present invention, the multi-objective optimization function is specifically: Where F is the optimization function value, η is the system operating efficiency (range [0,1]), E is the actual energy consumption, is the maximum allowable energy consumption, λ is the task completion rate (range [0,1]), α, β, γ is a weighting coefficient that satisfies α+β+γ=1 and can be dynamically adjusted according to the application scenario.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a fully closed-loop architecture of "perception-decision-control-feedback-adaptation," integrating terminal components such as intelligent device groups and sensor groups at the device layer, multi-source data acquisition and enhanced preprocessing modules at the perception layer, core units such as dynamic device capability maps and incremental AI decision models at the AI ​​decision layer, and cross-protocol command distribution and closed-loop feedback modules at the control layer. This enables in-depth processing of multi-source heterogeneous data, real-time perception of device capabilities, dynamic adjustment of task priorities, and autonomous optimization of collaborative strategies. It completely solves the technical pain points of rigid collaborative logic, slow response, and unreasonable resource allocation in traditional methods, and significantly improves the intelligence level and operating efficiency of IoT systems.

[0020] 2. This invention accurately captures the static parameters and dynamic states of devices through a four-layer structure of dynamic device capability graphs (basic information layer, real-time state layer, resource quantification layer, and collaborative adaptation layer) and a sliding time window update mechanism. Leveraging the multi-head attention mechanism of an incremental AI decision-making model, combined with GRU feature fusion, DNN task matching, and improved DRL strategy optimization, along with the EWC incremental learning algorithm, the model achieves adaptive adaptation to device updates, scenario changes, and user demand upgrades. Simultaneously, through cross-protocol compatibility and closed-loop control mechanisms, it adapts to scenarios where multiple protocols coexist while ensuring control accuracy. Compared to existing technologies, this invention significantly enhances the flexibility, scalability, and practicality of the method, and can be widely adapted to complex collaborative needs in multiple fields such as smart homes and industrial IoT. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram of the four-layer structure of the dynamic device capability map of the present invention; Figure 4 This is a schematic diagram of the module structure of the incremental AI decision-making model of the present invention; Figure 5 This is a schematic diagram of the cross-protocol instruction distribution and closed-loop control process of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figures 1 to 5 As shown, the present invention provides an AI-based multi-device collaborative control method in an IoT environment, comprising the following steps: S1. Multi-source heterogeneous data acquisition and enhanced preprocessing: Through the multi-type sensors and device interfaces of the IoT perception layer, collect full-state data of device operation, multi-dimensional environmental parameter data and user interaction intent data. Perform intelligent outlier removal, adaptive missing value filling, multi-dimensional feature enhancement and standardization processing on the raw data to obtain highly reliable structured data. S2. Dynamic Equipment Capability Map Construction and Update: During the initialization phase, the equipment's factory parameters, function list, communication protocol specifications, and operational constraint thresholds are collected to construct a basic capability map. During operation, based on real-time equipment status data, fault feedback information, and resource usage, the equipment capability description, available resource quantification value, and collaborative adaptation coefficient of the map are dynamically optimized through an incremental update algorithm. S3. Hierarchical Task Parsing and Dynamic Priority Determination: Natural language processing technology is used to parse the target parameters, execution constraints, and time requirements of user instructions and system-triggered tasks. By integrating subjective and objective weights through an improved analytic hierarchy process (AHP-entropy weight method), the task priority is dynamically determined. The priority is adjusted in real time according to the device status, the urgency of the environment, and the user's behavior habits. S4. Incremental AI Decision Model Training and Deployment: Construct a deep learning model that includes a feature fusion submodule, a task matching submodule, and a collaborative strategy optimization submodule. Use historical structured data to complete the initial training, continuously absorb new scenario data through incremental learning algorithms to optimize the model parameters, and deploy the trained model to the AI ​​decision layer to support real-time decision-making. S5. Dynamic collaborative control strategy generation: The AI ​​decision layer receives preprocessed real-time data, combines it with dynamic equipment capability map and real-time task priority, and analyzes equipment collaborative needs, resource constraints and optimization goals through trained AI models to generate the optimal collaborative control instruction set adapted to the current scenario. S6. Cross-protocol command distribution and closed-loop execution: The control layer uses an adaptive communication interface to be compatible with mainstream IoT protocols such as MQTT, HTTP, CoAP, and Modbus, accurately distributing control command sets to target devices. After the device executes the command, it provides real-time feedback on the execution status, result data, and resource consumption to the AI ​​decision layer. The decision layer compares the execution result with the preset target and dynamically adjusts the control strategy to form a closed-loop control.

[0024] refer to Figure 1 S1-1: Collects multi-source heterogeneous data, including full-state equipment operating data such as voltage, current, operating temperature, load rate, operating mode, remaining power / energy consumption, fault codes, response delay, etc.; and multi-dimensional environmental parameter data including temperature, humidity, light intensity, and air quality (PM2.5, etc.). Data includes concentration, noise level, vibration frequency, and spatial personnel density; user interaction intent data includes voice commands, text commands, button operations, gesture commands, and usage habit preference data. S1-2: Intelligent outlier removal. An improved Z-score algorithm combined with an isolated forest model is used. First, the Z-score algorithm is used to detect and remove random outliers that are more than 3 times the standard deviation. Then, the isolated forest model is used to identify systematic outliers caused by equipment failures and sensor malfunctions. S1-3: Adaptive imputation of missing values. For continuous data, a time-series prediction imputation method based on LSTM is used, and for discrete data, a probabilistic imputation method based on Bayesian network is used. The imputation process combines the historical operating patterns of the equipment with the environmental correlation characteristics. S1-4: Multi-dimensional feature enhancement, constructing a multi-dimensional feature set through time-domain feature extraction (sliding window mean, variance, peak value), frequency-domain feature extraction (spectral peak value after Fourier transform), and spatial feature extraction (correlation coefficient of multi-sensor data); S1-5: Standardization processing, using an improved Min-Max normalization algorithm, dynamically adjusting the normalization interval based on data distribution characteristics, eliminating dimensional differences while preserving the data distribution characteristics, resulting in structured data.

[0025] As a technical optimization of the present invention, the comprehensive collection and enhanced preprocessing of multi-source heterogeneous data solves the problems of coarse and low reliability of traditional data processing, provides high-quality data support for subsequent AI decision-making, and improves the accuracy and reliability of decision results.

[0026] refer to Figure 3 S2-1: Construct a basic capability map, including a basic equipment information layer, storing static information such as equipment model, factory parameters, function type, communication protocol type, rated power, and operating environment thresholds; S2-2: Real-time status layer update, based on the device's full operating status data, to update dynamic information such as the device's current working mode, operating status (normal / fault / standby), current load, and remaining power / energy consumption in real time; S2-3: Resource quantization layer construction, which uses a resource quantization model to convert the device's computing resources, storage resources, communication bandwidth, runtime, etc. into quantifiable values ​​in the range [0,1]. S2-4: Collaboration adaptation layer record, storing data such as the collaboration history of the storage device with other devices, adaptation success rate, collaboration latency, and energy consumption collaboration coefficient; S2-5: Dynamic update mechanism, which adopts a sliding time window mechanism, updates the information of each layer of the map every 1-5 seconds based on the latest collected data to ensure that the map is consistent with the actual state of the device.

[0027] As a technical optimization solution of the present invention, the construction and real-time updating of a dynamic equipment capability map realizes a comprehensive and accurate description of equipment capabilities, avoiding the mismatch between capabilities and actual status caused by traditional static equipment information tables, and providing a reliable basis for resource optimization allocation and equipment collaborative combination.

[0028] refer to Figure 2 S3-1: Layered task parsing. The task target parameters, execution constraints and time requirements are extracted through the BERT pre-trained model. Natural language instructions are converted into structured parameters. The time requirements are divided into immediate execution tasks (response time ≤ 1 second), timed execution tasks and conditional triggering tasks. S3-2: Subjective weighting is determined using the Analytic Hierarchy Process (AHP) to determine subjective weights based on the importance of user needs and task type (safety / comfort / efficiency). S3-3: Objective weight calculation, using the entropy weight method to determine objective weights based on objective data such as equipment resource utilization rate and environmental urgency. S3-4: Dynamic priority determination, which combines subjective and objective weights to obtain a comprehensive priority. The priority value ranges from [0,10] and is adjusted in real time according to changes in equipment status and environment.

[0029] As a technical optimization solution of the present invention, the hierarchical task parsing and dynamic priority determination realize the accurate understanding and adaptive scheduling of tasks, ensuring that urgent tasks and high-importance tasks are executed first, avoiding the resource allocation imbalance caused by traditional fixed priorities, and improving the rationality and timeliness of system response.

[0030] refer to Figure 4 S4-1: Feature fusion submodule construction, which adopts a multi-head attention mechanism combined with gated recurrent units (GRU) to dynamically weight and fuse equipment operation features, environmental features and user demand features, and outputs a fused feature vector; S4-2: Task matching submodule design, based on fused feature vectors, uses deep neural networks (DNN) to calculate the matching degree between tasks and the capabilities of each device, and selects a set of candidate devices that meet the constraints; S4-3: Construction of the cooperative strategy optimization submodule. An improved deep reinforcement learning (DRL) algorithm is adopted. The multi-objective optimization function is to maximize system operating efficiency, minimize energy consumption, and maximize task completion rate. A reward mechanism is constructed to generate the optimal cooperative control strategy. S4-4: Incremental learning training, using the Elastic Weight Consolidation (EWC) algorithm, protects learned historical knowledge while optimizing the model by absorbing new data, avoiding catastrophic forgetting; S4-5: Model deployment, deploying the trained model to the AI ​​decision layer (edge ​​gateway or cloud server) for real-time decision-making.

[0031] As a technical optimization solution of the present invention, the incremental AI decision-making model with multi-module collaboration improves the feature representation capability, device matching accuracy and strategy optimization effect, while realizing the adaptive adjustment of the model to new scenarios and new requirements, avoiding the drawback of traditional static models that need to be retrained, and reducing system maintenance costs.

[0032] refer to Figure 2 S5-1: Task-Equipment Matching. Based on task constraints and equipment capability parameters in the dynamic equipment capability map, the matching degree is calculated through the task matching submodule to filter out the candidate equipment set. S5-2: Resource optimization and allocation. Based on the resource quantification values ​​of candidate devices and the resource requirements of the task, a resource allocation algorithm is used to decompose the task into several sub-tasks and allocate them to the corresponding devices. S5-3: Cooperative logic generation. Combines the historical cooperative data of the device cooperative adaptation layer to generate cooperative logic and control parameters between devices, forming a complete cooperative control instruction set.

[0033] As a technical optimization solution of the present invention, the adaptability and optimality of the collaborative strategy are ensured by precise matching of tasks and devices, optimized allocation of resources, and dynamic generation of collaborative logic. This improves the efficiency and stability of multi-device collaborative operation and avoids collaborative failure caused by resource waste or insufficient capabilities.

[0034] refer to Figure 5 S6-1: Adaptive Protocol Identification. The control layer has a built-in protocol identification module that automatically identifies the communication protocol type of the target device through handshake signals, protocol port numbers, and data frame formats. S6-2: Command format conversion, based on a preset protocol command mapping library, converts standardized control commands into protocol formats supported by the target device; S6-3: Differentiated distribution strategy: High-priority tasks (priority ≥ 8) use UDP protocol to achieve low-latency transmission, while ordinary priority tasks use TCP protocol to ensure transmission reliability; S6-4: Closed-loop execution adjustment. The AI ​​decision-making layer calculates the deviation value between the execution result and the preset target through the deviation analysis model. When the deviation value > the preset threshold, the collaborative strategy generation process is re-triggered until the deviation value meets the requirements.

[0035] As a technical optimization solution of the present invention, the adaptive cross-protocol communication architecture and closed-loop control mechanism realize the compatibility and adaptation of multi-protocol devices, reduce system complexity and communication latency, and at the same time ensure the task execution effect and avoid the problems of instruction transmission loss or excessive execution deviation.

[0036] refer to Figure 2 S7-1: When the equipment reports a fault or the control command fails to execute, the AI ​​decision layer quickly selects alternative equipment based on the dynamic equipment capability map. S7-2: Generate backup collaborative control strategies and distribute them for execution; S7-3: Record faulty equipment information, alternative solutions, and execution results to provide data support for subsequent equipment maintenance and collaborative logic optimization.

[0037] As a technical optimization scheme of the present invention, the fault tolerance and stability of the system are improved through abnormal collaborative adaptive processing, ensuring that the task can continue to advance when the equipment fails or the instruction execution fails. At the same time, valuable data is accumulated for equipment maintenance and collaborative logic optimization, further improving the reliability of the system.

[0038] The multi-objective optimization function is as follows: Where F is the optimization function value, η is the system operating efficiency (range [0,1]), and E is the actual energy consumption. To determine the maximum allowable energy consumption, λ represents the task completion rate (range [0,1]), and α, β, γ is a weighting coefficient that satisfies α+β+γ=1 and can be dynamically adjusted according to the application scenario (e.g., in industrial scenarios, α is 0.4, β is 0.3, and γ is 0.3; in home scenarios, α is 0.2, β is 0.3, and γ is 0.5).

[0039] As a technical optimization scheme of the present invention, a balanced optimization of system operating efficiency, energy consumption control and task completion rate is achieved through a multi-objective optimization function. The weight coefficients can be dynamically adjusted according to the needs of different application scenarios, which improves the flexibility and adaptability of the method and meets the usage needs of diverse scenarios.

[0040] The working principle and usage process of this invention are as follows: First, the intelligent device group (such as air conditioners, machine tools, etc.), sensor group (such as temperature and humidity sensors, vibration sensors, etc.), and execution group (such as motors, valves, etc.) at the device layer collect data through communication modules (such as WiFi, Bluetooth, etc.), simultaneously receive subsequent control commands and execute corresponding operations, and provide real-time feedback on their own operating status; the multi-source data acquisition module at the perception layer collects the full-state data of device operation, multi-dimensional environmental parameter data, and user interaction intent data transmitted from the device layer. This data is then processed by an enhanced preprocessing module using an improved Z-score algorithm combined with an isolated forest model to remove outliers, and finally based on LSTM or Bayesian networks. The missing values ​​are filled in, and then the time domain, frequency domain, and spatial features are extracted by the feature extraction module. Subsequently, the normalization processing module uses an improved Min-Max normalization algorithm, dynamically adjusting the normalization interval based on the data distribution characteristics to eliminate dimensional differences while preserving the data's distribution features. Finally, the data transmission module uploads the highly reliable structured data to the AI ​​decision-making layer. The hierarchical task parsing module of the AI ​​decision-making layer analyzes the task objective parameters, execution constraints, and time requirements using a BERT pre-trained model. This is combined with a dynamic device capability map (including a basic information layer, a real-time status layer, a resource quantification layer, and a collaborative adaptation layer), using a sliding time window mechanism every 1- The system uses real-time device capability data (updated every 5 seconds) to dynamically determine task priorities using the AHP-entropy weighting method. The incremental AI decision-making model's feature fusion submodule fuses multi-dimensional features through a multi-head attention mechanism and GRU. The task matching submodule uses a DNN model to calculate the matching degree between tasks and devices and filter candidate devices. The collaborative strategy optimization submodule generates the optimal collaborative control instruction set based on an improved DRL algorithm and a multi-objective optimization function. If device failure or instruction execution failure occurs, the abnormal collaborative handling module quickly filters alternative devices and generates backup strategies. The control layer's protocol identification module identifies targets through handshake signals, port numbers, and data frame formats. The device's communication protocol type and instruction format conversion module convert standardized control instructions into protocol-compatible instructions based on the protocol instruction mapping library. The differentiated distribution module uses UDP protocol for low-latency transmission of tasks with priority ≥8 and TCP protocol for reliable transmission of tasks with priority <8, distributing instructions to the corresponding devices at the device layer. After the devices execute, they provide feedback on the execution status, result data, and resource consumption. The closed-loop feedback analysis module calculates the deviation value between the execution result and the preset target. If the deviation value is ≤ the threshold, the current strategy is maintained and continuously monitored by the perception layer. If the deviation value is > the threshold or the execution fails, the AI ​​decision layer is triggered to regenerate the strategy, forming a complete closed-loop usage process.

[0041] In summary, this AI-based multi-device collaborative control method in the IoT environment, through a fully closed-loop intelligent architecture of "perception-decision-control-feedback-adaptation," innovatively integrates multi-source heterogeneous data enhanced preprocessing, dynamic device capability mapping, hierarchical task dynamic priority determination, incremental AI decision-making models, and cross-protocol closed-loop execution technology. This achieves precise collaboration, dynamic adaptation, and autonomous optimization of multiple IoT devices. Its core innovations lie in enhancing decision-making credibility through deep multi-dimensional data mining, ensuring rational resource allocation through real-time device capability updates, optimizing scheduling efficiency through dynamic task priority adjustment, enhancing scenario adaptability through incremental AI model learning, and improving system stability through cross-protocol communication and anomaly handling. This provides an intelligent, efficient, and flexible solution for multi-device collaborative control in the IoT, effectively contributing to the digital and intelligent development of IoT system equipment.

[0042] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-device collaborative control method based on AI in an Internet of Things (IoT) environment, characterized in that: Includes the following steps: S1. Multi-source heterogeneous data acquisition and enhanced preprocessing: Through the sensors and device interfaces of the IoT sensing layer, data on the full operating status of the device, multi-dimensional environmental parameters, and user interaction intent are collected. After outlier removal, missing value filling, feature enhancement, and standardization, highly reliable structured data is obtained. S2. Dynamic Equipment Capability Map Construction and Update: Initialize and collect static information such as equipment factory parameters and function list to build a basic map. During operation, based on real-time status, fault feedback and resource usage data, dynamically optimize the equipment capability description, available resource quantification value and collaborative adaptation coefficient through incremental update algorithm. S3. Hierarchical Task Analysis and Dynamic Priority Determination: Natural language processing technology is used to analyze task objectives, constraints and time requirements. By integrating subjective and objective weights through an improved analytic hierarchy process (AHP-entropy weight method), task priority is dynamically determined (adjusted in real time according to equipment status, environmental urgency and user habits). S4. Incremental AI Decision Model Training and Deployment: Construct a deep learning model containing sub-modules for feature fusion, task matching, and collaborative strategy optimization. After initial training with historical data, the model absorbs new scenario data through incremental learning to optimize parameters and is deployed to the AI ​​decision layer to support real-time decision-making. S5. Dynamic collaborative control strategy generation: The AI ​​decision layer combines preprocessed real-time data, dynamic equipment capability map and real-time task priority, and analyzes collaborative needs and resource constraints through trained models to generate the optimal collaborative control instruction set adapted to the current scenario. S6. Cross-protocol command distribution and closed-loop execution: The control layer is compatible with mainstream IoT protocols through an adaptive communication interface and accurately distributes control commands; after the device executes the commands, it provides feedback on the execution status, results, and resource consumption data. The AI ​​decision layer compares the results with preset targets and dynamically adjusts the strategy to form closed-loop control.

2. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that: S1-1: Collect multi-source heterogeneous data, including full-state data of equipment operation such as voltage, current, operating temperature, load rate, working mode, remaining power / energy consumption, fault codes, and response delay; Multidimensional environmental parameter data include temperature, humidity, light intensity, and air quality (PM2.5, ...). Concentration), noise level, vibration frequency, and spatial personnel density; user interaction intent data includes voice commands, text commands, button operations, gesture commands, and usage habit preference data; S1-2: Intelligent outlier removal. An improved Z-score algorithm combined with an isolated forest model is used. First, the Z-score algorithm is used to detect and remove random outliers that are more than 3 times the standard deviation. Then, the isolated forest model is used to identify systematic outliers caused by equipment failures and sensor malfunctions. S1-3: Adaptive imputation of missing values. For continuous data, a time-series prediction imputation method based on LSTM is used, and for discrete data, a probabilistic imputation method based on Bayesian network is used. The imputation process combines the historical operating patterns of the equipment with the environmental correlation characteristics. S1-4: Multi-dimensional feature enhancement, constructing a multi-dimensional feature set through time-domain feature extraction (sliding window mean, variance, peak value), frequency-domain feature extraction (spectral peak value after Fourier transform), and spatial feature extraction (correlation coefficient of multi-sensor data); S1-5: Standardization processing, using an improved Min-Max normalization algorithm, dynamically adjusting the normalization interval based on data distribution characteristics, eliminating dimensional differences while preserving the data distribution characteristics, resulting in structured data.

3. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that: S2-1: Construct a basic capability map, including a basic equipment information layer, storing static information such as equipment model, factory parameters, function type, communication protocol type, rated power, and operating environment thresholds; S2-2: Real-time status layer update, based on the device's full operating status data, to update dynamic information such as the device's current working mode, operating status (normal / fault / standby), current load, and remaining power / energy consumption in real time; S2-3: Resource quantization layer construction, which uses a resource quantization model to convert the device's computing resources, storage resources, communication bandwidth, runtime, etc. into quantifiable values ​​in the range [0,1]. S2-4: Collaboration adaptation layer record, storing data such as the collaboration history of the storage device with other devices, adaptation success rate, collaboration latency, and energy consumption collaboration coefficient; S2-5: Dynamic update mechanism, which adopts a sliding time window mechanism, updates the information of each layer of the map every 1-5 seconds based on the latest collected data to ensure that the map is consistent with the actual state of the device.

4. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that: S3-1: Layered task analysis, extracting task target parameters, execution constraints and time requirements through BERT pre-trained model, and converting natural language instructions into structured parameters. The time requirements are divided into immediate execution tasks (response time ≤ 1 second), timed execution tasks and conditional triggering tasks. S3-2: Subjective weighting is determined using the Analytic Hierarchy Process (AHP) to determine subjective weights based on the importance of user needs and task type (safety / comfort / efficiency). S3-3: Objective weight calculation, using the entropy weight method to determine objective weights based on objective data such as equipment resource utilization rate and environmental urgency. S3-4: Dynamic priority determination, which combines subjective and objective weights to obtain a comprehensive priority. The priority value ranges from [0,10] and is adjusted in real time according to changes in equipment status and environment.

5. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that: S4-1: Feature fusion submodule construction, which adopts a multi-head attention mechanism combined with a gated recurrent unit (GRU) to dynamically weight and fuse equipment operation features, environmental features and user demand features, and outputs a fused feature vector; S4-2: Task matching submodule design, based on fused feature vectors, uses deep neural networks (DNN) to calculate the matching degree between tasks and the capabilities of each device, and selects a set of candidate devices that meet the constraints; S4-3: Construction of the cooperative strategy optimization submodule. An improved deep reinforcement learning (DRL) algorithm is adopted. The multi-objective optimization function is to maximize system operating efficiency, minimize energy consumption, and maximize task completion rate. A reward mechanism is constructed to generate the optimal cooperative control strategy. S4-4: Incremental learning training, using the Elastic Weight Consolidation (EWC) algorithm, protects learned historical knowledge while optimizing the model by absorbing new data, avoiding catastrophic forgetting; S4-5: Model deployment, deploying the trained model to the AI ​​decision layer (edge ​​gateway or cloud server) for real-time decision-making.

6. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that: S5-1: Task-equipment matching. Based on task constraints and equipment capability parameters in the dynamic equipment capability map, the matching degree is calculated through the task matching submodule to filter out a set of candidate equipment. S5-2: Resource optimization and allocation. Based on the resource quantification values ​​of candidate devices and the resource requirements of the task, a resource allocation algorithm is used to decompose the task into several sub-tasks and allocate them to the corresponding devices. S5-3: Cooperative logic generation. Combines the historical cooperative data of the device cooperative adaptation layer to generate cooperative logic and control parameters between devices, forming a complete cooperative control instruction set.

7. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that: S6-1: Adaptive protocol identification. The control layer has a built-in protocol identification module that automatically identifies the communication protocol type of the target device through handshake signals, protocol port numbers, and data frame formats. S6-2: Command format conversion, based on a preset protocol command mapping library, converts standardized control commands into protocol formats supported by the target device; S6-3: Differentiated distribution strategy: High-priority tasks (priority ≥ 8) use UDP protocol to achieve low-latency transmission, while ordinary priority tasks use TCP protocol to ensure transmission reliability; S6-4: Closed-loop execution adjustment. The AI ​​decision-making layer calculates the deviation value between the execution result and the preset target through the deviation analysis model. When the deviation value > the preset threshold, the collaborative strategy generation process is re-triggered until the deviation value meets the requirements.

8. The multi-device collaborative control method based on AI in an IoT environment according to claim 1, characterized in that, It also includes anomaly cooperative adaptive processing steps: S7-1: When the equipment reports fault information or the control command fails to execute, the AI ​​decision-making layer quickly selects alternative equipment based on the dynamic equipment capability map; S7-2: Generate backup collaborative control strategies and distribute them for execution; S7-3: Record faulty equipment information, alternative solutions, and execution results to provide data support for subsequent equipment maintenance and collaborative logic optimization.

9. A multi-device collaborative control method based on AI in an IoT environment according to claim 5, characterized in that, The multi-objective optimization function is as follows: Where F is the optimization function value, η is the system operating efficiency (range [0,1]), and E is the actual energy consumption. To determine the maximum allowable energy consumption, λ represents the task completion rate (range [0,1]), and α, β, γ is a weighting coefficient that satisfies α+β+γ=1 and can be dynamically adjusted according to the application scenario.

Images