Method and system for constructing linkage control scene of smart home

Abstract

The invention discloses a linkage control scene building method and system for smart home, and belongs to the technical field of smart home, and the method comprises the steps: S1, environment preparation and hardware deployment, S2, equipment configuration and protocol adaptation, S3, scene logic modeling and AI intention recognition, S4, edge side cooperative control, S5, AI model deployment and continuous learning, and S6, iterative optimization. On the basis of realizing intelligent home linkage control, network dependence can be reduced, and interaction can be simplified through AI intention recognition.

Description

Technical Field

[0001] This invention relates to the field of smart home technology, and more specifically, to a method and system for building a smart home linkage control scenario. Background Technology

[0002] Smart home technology uses the home as a platform to connect various devices within the home, creating an intelligent living environment. It integrates home-related facilities using comprehensive wiring, network communication, security, automatic control, and audio-visual technologies to build an efficient management system for residential facilities and daily household affairs, enhancing home security, convenience, comfort, and aesthetics.

[0003] Traditional systems rely on proprietary brand protocols, require multiple gateways for bridging across ecosystem devices, increasing deployment costs. The response latency for pure cloud-based decision-making is greater than 2 seconds, and the system is unavailable when the network is down. Furthermore, the rule engine requires manual configuration of IF-THEN logic and cannot parse ambiguous commands. In addition, centralized cloud processing requires uploading raw voice / video data, which increases the probability of privacy leaks due to hacker attacks. Summary of the Invention

[0004] 1. Technical problems to be solved To address the problems existing in the prior art, the purpose of this invention is to provide a method and system for building a smart home linkage control scenario. This invention can reduce network dependence and simplify interaction through AI intent recognition, while realizing smart home linkage control.

[0005] 2. Technical Solution

[0006] To solve the above problems, the present invention adopts the following technical solution: A method for building a smart home linkage control scenario includes: S1. Environment preparation and hardware deployment: Building the Matter protocol network, network layering architecture and deploying edge computing nodes, performing device configuration protocol adaptation and AI interaction hardware integration; S2. Equipment Configuration and Protocol Adaptation: Matter protocol device network access and authentication, edge computing node configuration and resource allocation, and multi-protocol device compatibility handling; S3, Scene Logic Modeling and AI Intent Recognition: Multimodal data collection and processing are performed, followed by fusion of user behavior data, scene state modeling based on the data, language understanding and dynamic generation of scene rules; S4. Edge-side collaborative control: Local decision-making and conflict resolution for the scheduling and allocation of edge computing resources; S5, AI Model Deployment and Continuous Learning: Lightweight the model, use federated learning and model updates to achieve scene adaptation; S6. Iterative optimization: Federated learning model updates and dynamic tuning of the rule engine.

[0007] As a preferred embodiment of the present invention, the equipment selected for protocol network construction in step S1 includes: Border Router: Deploy Matter border routers that support Thread / Wi-Fi dual-mode to bridge the Thread Mesh network and the IP network; Terminal devices: Select Matter-certified chips that support Wi-Fi 6 / Thread protocols to ensure low power consumption and high concurrency connections.

[0008] Security Authentication: The device must be pre-installed with a PKI certificate and can be configured for one-click network access via QR code or NFC. Users can complete the device's network access and security binding by scanning the QR code with a mobile app.

[0009] The network layered architecture in step S1 includes: Physical layer: It adopts a hybrid networking of 2.4GHz and 5GHz dual-band Wi-Fi and Thread Mesh network. 2.4GHz is used to ensure compatibility with old devices, and 5GHz is used to ensure high-speed transmission. Network layer: Unifies device addressing through the IPv6 protocol; Thread border routers are responsible for protocol conversion. Application layer: Deploy the Home Assistant open-source platform to uniformly manage device clusters and scene rules.

[0010] As a preferred embodiment of the present invention, the device configuration protocol adaptation step S1 involves mapping device capabilities to a Matter standard cluster, converting old device data to Matter TLV format via a Zigbee to Matter gateway bridge, selecting an ICS-43432 multi-microphone module to support beamforming and noise suppression, improving far-field speech recognition rate, deploying a MobileBERT lightweight speech recognition model on the edge gateway to support offline command parsing, and then performing sensor collaboration by inputting data from light sensors and human infrared sensors into the edge nodes, combining voice commands to generate composite intents, learning user habits through an LSTM model, preloading scene rules to reduce trigger latency, collecting fuzzy command samples, updating the edge node model through federated learning, and optimizing the recognition accuracy of temperature adjustment intents.

[0011] As a preferred embodiment of the present invention, the protocol device network access step S2 includes the following steps: Broadcast discovery: Devices seeking to join the network broadcast their own information via mDNS / DNS-SD. The Matter controller scans devices within the local area network and obtains the device's Discriminator identification code. Credential authentication: Users scan the device's QR code or enter a PIN code, and the controller establishes a secure temporary session with the device via the PAKE protocol to prevent man-in-the-middle attacks; The protocol device authentication in step S2 includes the following steps: Device certification: The controller extracts the DAC device certification certificate and PAI product certification intermediate certificate of the device to verify the legality of the device; Key negotiation: Generate NOC node operation certificate through CSR certificate signing request, install it on the device and configure network parameters; Fabric Join: Devices are assigned a unique node ID and joined to a specified Fabric security domain, enabling secure communication between devices within the domain; The edge computing node configuration in step S2 includes the following steps: Gateway selection: Use edge devices that support NPU acceleration and deploy a lightweight AI inference framework; Containerized deployment: Divide the system into containers using Docker / Kubernetes and run them separately; Matter protocol stack: handles device communication and state synchronization.

[0012] OpenHab rule engine: executes local automated rules; AI model: MobileBERT compressed version used for intent recognition; Bandwidth management: Allocate high-priority bandwidth to security devices to ensure real-time response.

[0013] Computational resource scheduling: AI inference tasks are allocated to the NPU, while the rule engine uses CPU resources to avoid resource contention; The multi-protocol device compatibility processing in step S2 includes the following steps: Unified data model: Define common attributes for devices to ensure consistent control across different brands of devices; Standardized interfaces: JSON / XML data format is used, and data is transmitted via HTTP / MQTT, supporting device control interfaces and data query interfaces; Communication encryption: Matter sessions use AES-256 encryption, and communication between devices establishes a secure channel through CASE certificate authentication sessions; Privacy protection: Sensitive data is processed locally at the edge node, with only de-identified metadata uploaded to the cloud; Role definition: Administrators have full privileges, regular users have device control privileges, and visitors can only view. Access control auditing: The edge gateway records operation logs and periodically detects abnormal access.

[0014] As a preferred embodiment of the present invention, the multimodal data acquisition and processing in step S3 includes the following steps: Environmental data acquisition: Environmental data is acquired in real time through DHT22 temperature and humidity sensor, BH1750 light sensor and HC-SR501 human infrared sensor, and transmitted to edge gateway via Matter over Thread protocol; Device status monitoring: Obtain the real-time status of the device and map it to the Matter standard Cluster attribute; The user behavior data fusion in step S3 includes the following steps: Voice command parsing: The microphone array collects speech, which is then converted into text by edge ASR and keywords are extracted; Visual behavior recognition: Camera data is processed by a lightweight CNN model on a Jetson Nano to recognize human posture and facial expressions; Time-series data alignment: Synchronize multi-source data through timestamps to construct a unified time-series feature vector.

[0015] The scene state modeling in step S3 includes the following steps: Dynamic scene modeling: An indoor 3D structural model is constructed using SLAM, and combined with light intensity distribution data and regional temperature difference maps to generate a dynamic twin scene model; GNN (Graph Neural Network) Application: Constructing a relationship graph between devices, users, and the environment, where node attributes include device status and user habits, and edge weights are dynamically adjusted by an attention mechanism; Feature engineering optimization: Calculate device usage time and trigger interval, and add LSTM input sequence; Multi-scale feature fusion: Combining frequency domain features and spatial features to improve the robustness of scene classification.

[0016] As a preferred embodiment of the present invention, the language understanding in step S3 includes the following steps: Semantic understanding and slot filling: Parse user commands using an NLP module to extract intent slots; Context-aware correction: Combines dialogue history and environmental state to dynamically correct intent, and uses dynamic weight formula for sensor weighted scoring. Input data is normalized before weighted scoring. The dynamic generation of scene rules in step S3 includes the following steps: Rule engine configuration: Define basic scene logic, IF time∈[23:00-6:00] AND human sensor=no triggerTHEN start sleep mode; Reinforcement learning optimization: The PPO algorithm is adopted, with user satisfaction as the reward function, and the rule parameters are updated iteratively. AI intent translation scenario: Automatically mapping user fuzzy commands into device operation sequences; Habit learning prediction: The LSTM model learns users' historical behavior and preloads scene rules into the edge cache; The weighted scoring formula for the sensor is as follows:

[0017] Where m is the semantic score, n is the sensor score, and w b For sensor weights.

[0018] As a preferred embodiment of the present invention, the edge computing resource scheduling and allocation in step S4 includes the following steps: Priority allocation: The edge gateway dynamically allocates computing resources and bandwidth based on device type and scenario requirements. Security devices are allocated high-priority resources to ensure response latency <100ms, comfort devices are allocated medium-priority resources, and sensors and energy-saving devices are allocated low-priority resources. Load balancing: Computing tasks are divided using Docker / Kubernetes containerization technology, and the Matter protocol stack, MobileBERT AI model inference, and OpenHab rule engine are deployed in separate containers to avoid resource contention; Task offloading strategy: Simple instructions are processed directly by edge nodes, while complex tasks are generated by a large AI model in the cloud and then sent to edge nodes for execution, reducing reliance on the cloud. Data preprocessing: Edge nodes filter and compress sensor data, uploading only key features to the cloud to reduce bandwidth consumption; The local decision-making and conflict resolution in step S4 includes the following steps: Rule engine deployment: Deploy a lightweight rule engine on the edge gateway to define local automation rules and ensure that core scenarios can still be executed when the network is down; Reinforcement learning optimization: Using user satisfaction as the reward function, the PPO algorithm is used to dynamically adjust rule parameters to adapt to changes in habits.

[0019] As a preferred embodiment of the present invention, the model lightweighting step in step S5 includes: Pruning: Removing redundant neurons or connections in a neural network to reduce model complexity and size; Quantization: Converts 32-bit floating-point parameters into 8-bit integers, reducing memory usage and improving inference speed; Knowledge distillation: Using BERT pre-trained large models to guide the training of MobileBERT lightweight models; The federated learning and model update steps in step S5 include: Edge-cloud collaborative training: Each household edge node periodically uploads model gradients, which are then aggregated in the cloud to generate a global model; Differential privacy protection: Laplace noise is added before gradient transmission to prevent leakage of user behavior data; OTA differential update: only transmits the differences in the model, reducing bandwidth usage; A / B testing verification: Run the old and new models in parallel, and select the version with high user satisfaction for full deployment.

[0020] As a preferred embodiment of the present invention, the federated learning model update in step S6 includes the following steps: Edge-cloud collaboration: Each household node uploads model gradients, which are aggregated in the cloud to generate a global model, and are OTA differentially updated every 24 hours. RLHF reinforcement learning: using user satisfaction as the reward function, it optimizes scene rule parameters through the PPO algorithm; The dynamic optimization of the rule engine in step S6 includes the following steps: Conflict resolution strategy: Based on a hierarchical arbitration mechanism of security > comfort > energy saving, the rule weights are dynamically adjusted; Energy efficiency adaptive: Automatically switches the device's operating mode by combining electricity price data and user habits.

[0021] A smart home linkage control scene construction system includes: a device access and protocol adaptation module, an edge computing and local decision-making module, a scene logic and AI intent recognition module, a user interaction and control module, and a security and maintenance module; The device access and protocol adaptation module is used to achieve seamless access and interconnection of cross-brand devices through the Matter protocol's unified standard and multi-protocol compatibility layer, breaking down the barriers of the traditional smart home ecosystem. The edge computing and local decision-making module is used to utilize the local computing capabilities of edge nodes to achieve low-latency response and continuous operation of core functions in network outage scenarios, significantly reducing cloud dependence. The scene logic and AI intent recognition module is used to accurately parse users' ambiguous intents and dynamically generate scene rules through multimodal perception fusion and lightweight AI models. The user interaction and control module is used to provide multi-terminal interaction methods such as voice, APP, and central control screen, supporting one-click scene triggering and personalized rule configuration, simplifying user operation complexity. The security and maintenance module is used to ensure data security based on the Matter protocol's end-to-end encryption and device authentication mechanism, combined with OTA differential updates and redundancy design to ensure high system reliability.

[0022] 3. Beneficial effects

[0023] Compared with the prior art, the advantages of this invention are: (1) This invention unifies the access of multiple heterogeneous devices through the Matter protocol cluster, and achieves compatibility with old devices by cooperating with the Zigbee to Matter gateway, eliminating ecological barriers. Sensitive data is processed locally at the edge computing layer, and only desensitized features are uploaded to the cloud. Combined with differential privacy and model gradient encryption of federated learning, the risk of data leakage is reduced and complies with GDPR privacy specifications.

[0024] (2) The edge nodes of this invention reduce the scene triggering delay from 2s to within 200ms through three-level resource scheduling and containerized task isolation. The local rule engine and dual-machine hot standby mechanism ensure the availability of core scenes when the network is down, and reduce the dependence on the cloud. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the steps of a method for building a smart home linkage control scenario according to the present invention. Figure 2 This is a schematic diagram of a smart home linkage control scenario building system according to the present invention. Detailed Implementation

[0026] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] Example

[0028] Please see Figure 1 A method for building a smart home linkage control scenario, comprising: S1. Environment preparation and hardware deployment: Building the Matter protocol network, network layering architecture and deploying edge computing nodes, performing device configuration protocol adaptation and AI interaction hardware integration; S2. Equipment Configuration and Protocol Adaptation: Matter protocol device network access and authentication, edge computing node configuration and resource allocation, and multi-protocol device compatibility handling; S3, Scene Logic Modeling and AI Intent Recognition: Multimodal data collection and processing are performed, followed by fusion of user behavior data, scene state modeling based on the data, language understanding and dynamic generation of scene rules; S4. Edge-side collaborative control: Local decision-making and conflict resolution for the scheduling and allocation of edge computing resources; S5, AI Model Deployment and Continuous Learning: Lightweight the model, use federated learning and model updates to achieve scene adaptation; S6. Iterative optimization: Federated learning model updates and dynamic tuning of the rule engine.

[0029] In a specific embodiment of the present invention, multiple heterogeneous devices are accessed through the Matter protocol cluster, and compatibility with old devices is achieved with the Zigbee to Matter gateway, eliminating ecological barriers. Sensitive data is processed locally at the edge computing layer, and only de-identified features are uploaded to the cloud. Combined with differential privacy and model gradient encryption of federated learning, the risk of data leakage is reduced, which complies with GDPR privacy specifications. Edge nodes reduce the scene triggering latency from 2s to within 200ms through three-level resource scheduling and containerized task isolation. The local rule engine and dual-machine hot standby mechanism ensure the availability of core scenes when the network is down, reducing the dependence on the cloud.

[0030] Specifically, the devices selected for protocol network setup in step S1 include: Border Router: Deploy Matter border routers that support Thread / Wi-Fi dual-mode to bridge the Thread Mesh network and the IP network; Terminal devices: Select Matter-certified chips that support Wi-Fi 6 / Thread protocols to ensure low power consumption and high concurrency connections.

[0031] Security Authentication: The device must be pre-installed with a PKI certificate and can be configured for one-click network access via QR code or NFC. Users can complete the device's network access and security binding by scanning the QR code with a mobile app.

[0032] The network layered architecture in step S1 includes: Physical layer: It adopts a hybrid networking of 2.4GHz and 5GHz dual-band Wi-Fi and Thread Mesh network. 2.4GHz is used to ensure compatibility with old devices, and 5GHz is used to ensure high-speed transmission. Network layer: Unifies device addressing through the IPv6 protocol; Thread border routers are responsible for protocol conversion. Application layer: Deploy the Home Assistant open-source platform to uniformly manage device clusters and scene rules.

[0033] In a specific embodiment of the present invention, the border router connects to low-power devices via a Thread Mesh network and simultaneously connects to high-speed devices via Wi-Fi. IPv6 is used for unified addressing, dual-band separation at the physical layer reduces interference, and the application layer centrally manages rules through the Home Assistant open-source platform. Dual-mode networking reduces deployment costs, and IPv6 addressing improves device discovery efficiency.

[0034] Specifically, the device configuration protocol adaptation step S1 involves mapping device capabilities to the Matter standard cluster, converting old device data to Matter TLV format via a Zigbee to Matter gateway bridge, selecting the ICS-43432 multi-microphone module to support beamforming and noise suppression, improving far-field speech recognition rate, deploying the MobileBERT lightweight speech recognition model on the edge gateway to support offline command parsing, and then performing sensor collaboration by inputting data from light sensors and human infrared sensors into the edge nodes, combining them with voice commands to generate composite intents, learning user habits through an LSTM model, preloading scene rules to reduce trigger latency, collecting fuzzy command samples, updating the edge node model through federated learning, and optimizing the recognition accuracy of temperature adjustment intents.

[0035] In a specific embodiment of the present invention, the Zigbee to Matter gateway converts data from old devices into TLV format, the ICS-43432 microphone array suppresses noise through beamforming, the MobileBERT model parses commands such as temperature adjustment at the edge, sensor data and voice commands are fused and input into the LSTM model, the rule of automatically turning on the lights when you come home is preloaded, and the protocol conversion is compatible with old devices.

[0036] Specifically, step S2, which involves the protocol device joining the network, includes the following steps: Broadcast discovery: Devices seeking to join the network broadcast their own information via mDNS / DNS-SD. The Matter controller scans devices within the local area network and obtains the device's Discriminator identification code. Credential authentication: Users scan the device's QR code or enter a PIN code, and the controller establishes a secure temporary session with the device via the PAKE protocol to prevent man-in-the-middle attacks; The protocol device authentication in step S2 includes the following steps: Device certification: The controller extracts the DAC device certification certificate and PAI product certification intermediate certificate of the device to verify the legality of the device; Key negotiation: Generate NOC node operation certificate through CSR certificate signing request, install it on the device and configure network parameters; Fabric Join: Devices are assigned a unique node ID and joined to a specified Fabric security domain, enabling secure communication between devices within the domain; Step S2, edge computing node configuration, includes the following steps: Gateway selection: Use edge devices that support NPU acceleration and deploy a lightweight AI inference framework; Containerized deployment: Divide the system into containers using Docker / Kubernetes and run them separately; Matter protocol stack: handles device communication and state synchronization.

[0037] OpenHab rule engine: executes local automated rules; AI model: MobileBERT compressed version used for intent recognition; Bandwidth management: Allocate high-priority bandwidth to security devices to ensure real-time response.

[0038] Computational resource scheduling: AI inference tasks are allocated to the NPU, while the rule engine uses CPU resources to avoid resource contention; Step S2, multi-protocol device compatibility processing, includes the following steps: Unified data model: Define common attributes for devices to ensure consistent control across different brands of devices; Standardized interfaces: JSON / XML data format is used, and data is transmitted via HTTP / MQTT, supporting device control interfaces and data query interfaces; Communication encryption: Matter sessions use AES-256 encryption, and communication between devices establishes a secure channel through CASE certificate authentication sessions; Privacy protection: Sensitive data is processed locally at the edge node, with only de-identified metadata uploaded to the cloud; Role definition: Administrators have full privileges, regular users have device control privileges, and visitors can only view. Access control auditing: The edge gateway records operation logs and periodically detects abnormal access.

[0039] In a specific embodiment of the present invention, the device broadcasts the Discriminator identification code via mDNS, and after scanning the code by the mobile app, an encrypted session is established via the PAKE protocol; the DAC certificate verifies the legitimacy of the device, and the NOC certificate is assigned to join the Fabric security domain; the edge node accelerates AI inference through NPU, Docker containers isolate high-priority tasks, the PAKE protocol prevents man-in-the-middle attacks, and NPU acceleration makes the AI ​​inference latency <500ms.

[0040] Specifically, step S3, multimodal data acquisition and processing, includes the following steps: Environmental data acquisition: Environmental data is acquired in real time through DHT22 temperature and humidity sensor, BH1750 light sensor and HC-SR501 human infrared sensor, and transmitted to edge gateway via Matter over Thread protocol; Device status monitoring: Obtain the real-time status of the device and map it to the Matter standard Cluster attribute; Step S3, user behavior data fusion, includes the following steps: Voice command parsing: The microphone array collects speech, which is then converted into text by edge ASR and keywords are extracted; Visual behavior recognition: Camera data is processed by a lightweight CNN model on a Jetson Nano to recognize human posture and facial expressions; Time-series data alignment: Synchronize multi-source data through timestamps to construct a unified time-series feature vector.

[0041] Step S3, scene state modeling, includes the following steps: Dynamic scene modeling: An indoor 3D structural model is constructed using SLAM, and combined with light intensity distribution data and regional temperature difference maps to generate a dynamic twin scene model; GNN (Graph Neural Network) Application: Constructing a relationship graph between devices, users, and the environment, where node attributes include device status and user habits, and edge weights are dynamically adjusted by an attention mechanism; Feature engineering optimization: Calculate device usage time and trigger interval, and add LSTM input sequence; Multi-scale feature fusion: Combining frequency domain features and spatial features to improve the robustness of scene classification.

[0042] In a specific embodiment of the present invention, the DHT22 sensor collects temperature and humidity, the BH1750 detects light intensity, and the data is transmitted via the Thread protocol; the camera runs a CNN on a Jetson Nano to recognize fall actions; SLAM constructs an indoor 3D model, GNN analyzes the "air conditioner-user" association weights, LSTM learns pre-loaded rules for user lighting habits, multi-sensor fusion improves scene recognition accuracy, and GNN reduces rule conflict rate.

[0043] Specifically, the language understanding step S3 includes the following steps: Semantic understanding and slot filling: Parse user commands using an NLP module to extract intent slots; Context-aware correction: Combines dialogue history and environmental state to dynamically correct intent, and uses dynamic weight formula for sensor weighted scoring. Input data is normalized before weighted scoring. Step S3, dynamic generation of scene rules, includes the following steps: Rule engine configuration: Define basic scene logic, IF time∈[23:00-6:00] AND human sensor=no triggerTHEN start sleep mode; Reinforcement learning optimization: The PPO algorithm is adopted, with user satisfaction as the reward function, and the rule parameters are updated iteratively. AI intent translation scenario: Automatically mapping user fuzzy commands into device operation sequences; Habit learning prediction: The LSTM model learns users' historical behavior and preloads scene rules into the edge cache; The weighted scoring formula for the sensor is as follows:

[0044] Where m is the semantic score, n is the sensor score, and w b Sensor weights; Where w b The dynamic adjustment steps for sensor weights are as follows: Initial weight allocation: Preset initial weights based on sensor type and scene requirements. The initial weight of the light sensor is 0.6 in cinema mode and 0.4 for the human infrared sensor. The initial weights are fine-tuned based on the range of light intensity changes over the past week and the fact that users often manually turn on the lights when the light intensity is less than 50 lux. Real-time data acquisition and feature extraction: Real-time data from associated sensors are collected via Wi-Fi IoT protocol. Sensor data of different dimensions are normalized to the [0,1] interval, with light intensity 50 lux → 0.2 and human infrared signal intensity 0.8 → 0.8, eliminating the influence of dimension differences on weight calculation.

[0045] Correlation analysis and reliability assessment: Based on historical data, a correlation model was trained to quantify the degree of correlation between sensor data and the target scene. The correlation coefficient between light intensity and the operation of turning on the lights was 0.85. Sensor reliability assessment: statistically analyze sensor data stability and introduce a fault detection mechanism; Dynamic weight updates: Weights are recalculated every 5–10 minutes to ensure the system responds quickly to environmental changes; Rule engine trigger: The sensor signal priority is reordered according to dynamic weights. When the light intensity weight is 0.7 > the human body infrared weight is 0.3, the light will be turned on first in response to changes in light intensity. Multi-sensor fusion decision: If the weights of multiple sensors are close, such as 0.55 for illumination and 0.45 for human infrared, then threshold superposition is used to trigger control commands to avoid misjudgment by a single sensor. User behavior learning and long-term optimization: Analyze users' historical operation data through machine learning using Hidden Markov Models and embed high-frequency behavior patterns into the weight calculation logic; Feedback closed-loop correction: Record user intervention records for automatic control, correct correlation scores in reverse, and continuously optimize weight allocation.

[0046] In a specific embodiment of the present invention, the NLP module interprets "too cold" as {intent: adjust temperature, value: +2℃}, the light sensor data is used to weight and correct the intent, the PPO algorithm optimizes rule parameters such as the temperature adjustment range with satisfaction as a reward, LSTM predicts user behavior to automatically generate rules for turning off the air conditioner when leaving home, context awareness improves the accuracy of understanding fuzzy instructions, and the PPO algorithm reduces the user operation cancellation rate.

[0047] Specifically, step S4, edge computing resource scheduling and allocation, includes the following steps: Priority allocation: The edge gateway dynamically allocates computing resources and bandwidth based on device type and scenario requirements. Security devices are allocated high-priority resources to ensure response latency <100ms, comfort devices are allocated medium-priority resources, and sensors and energy-saving devices are allocated low-priority resources. Load balancing: Computing tasks are divided using Docker / Kubernetes containerization technology, and the Matter protocol stack, MobileBERT AI model inference, and OpenHab rule engine are deployed in separate containers to avoid resource contention; Task offloading strategy: Simple instructions are processed directly by edge nodes, while complex tasks are generated by a large AI model in the cloud and then sent to edge nodes for execution, reducing reliance on the cloud. Data preprocessing: Edge nodes filter and compress sensor data, uploading only key features to the cloud to reduce bandwidth consumption; Step S4, local decision-making and conflict resolution, includes the following steps: Rule engine deployment: Deploy a lightweight rule engine on the edge gateway to define local automation rules and ensure that core scenarios can still be executed when the network is down; Reinforcement learning optimization: Using user satisfaction as the reward function, the PPO algorithm is used to dynamically adjust rule parameters to adapt to changes in habits.

[0048] In a specific embodiment of the present invention, the edge gateway allocates high-priority resources to door locks and medium-priority resources to air conditioners. The Docker container isolates the OpenHab rule engine and the AI ​​model. When the network is down, the local rule engine executes core scenarios such as automatic window opening in case of fire. Resource hierarchical scheduling improves CPU utilization, and the local rule engine ensures availability in the scenario of network downtime.

[0049] Specifically, the model lightweighting step in step S5 includes: Pruning: Removing redundant neurons or connections in a neural network to reduce model complexity and size; Quantization: Converts 32-bit floating-point parameters into 8-bit integers, reducing memory usage and improving inference speed; Knowledge distillation: Using BERT pre-trained large models to guide the training of MobileBERT lightweight models; Step S5, the federated learning and model update steps, include: Edge-cloud collaborative training: Each household edge node periodically uploads model gradients, which are then aggregated in the cloud to generate a global model; Differential privacy protection: Laplace noise is added before gradient transmission to prevent leakage of user behavior data; OTA differential update: only transmits the differences in the model, reducing bandwidth usage; A / B testing verification: Run the old and new models in parallel, and select the version with high user satisfaction for full deployment.

[0050] In a specific embodiment of the present invention, redundant neurons in MobileBERT are removed by pruning, the model is compressed by 8-bit quantization, knowledge distillation preserves 90% of BERT's accuracy, gradients are uploaded from edge nodes, a global model is generated by cloud aggregation, OTA differential updates only transmit parameter increments ΔW, the model size is <10MB to adapt to edge NPUs, and differential privacy protection reduces the risk of data leakage.

[0051] Specifically, step S6, the federated learning model update, includes the following steps: Edge-cloud collaboration: Each household node uploads model gradients, which are aggregated in the cloud to generate a global model, and are OTA differentially updated every 24 hours. RLHF reinforcement learning: using user satisfaction as the reward function, it optimizes scene rule parameters through the PPO algorithm; Step S6, dynamic tuning of the rule engine, includes the following steps: Conflict resolution strategy: Based on a hierarchical arbitration mechanism of security > comfort > energy saving, the rule weights are dynamically adjusted; Energy efficiency adaptive: Automatically switches the device's operating mode by combining electricity price data and user habits.

[0052] In a specific embodiment of the present invention, the cloud aggregates the gradients of each household to generate a dialect recognition model; the PPO algorithm optimizes the air conditioner start-stop rules with electricity cost savings as a reward; security rules prioritize covering energy-saving rules; federated learning reduces the dialect recognition cycle from 7 days to 12 hours; and hierarchical arbitration reduces the conflict rate.

[0053] Please see Figure 2 A smart home linkage control scene construction system includes: a device access and protocol adaptation module for achieving seamless access and interconnection of cross-brand devices through the Matter protocol's unified standard and multi-protocol compatibility layer, breaking down traditional smart home ecosystem barriers; an edge computing and local decision-making module for utilizing the local computing capabilities of edge nodes to achieve low-latency response and continuous operation of core functions in network outage scenarios, significantly reducing cloud dependence; a scene logic and AI intent recognition module for accurately parsing users' ambiguous intents and dynamically generating scene rules through multimodal perception fusion and lightweight AI models; a user interaction and control module for providing multi-terminal interaction methods such as voice, APP, and central control screen, supporting one-click scene triggering and personalized rule configuration, simplifying user operation complexity; and a security and maintenance module for ensuring data security through end-to-end encryption and device authentication mechanisms based on the Matter protocol, combined with OTA differential updates and redundancy design to ensure high system reliability.

[0054] In a specific embodiment of the present invention, the device access module enables cross-brand control, the edge decision module executes local rules, the AI ​​module analyzes the viewing mode to trigger screen / sound linkage, the security module ensures reliability through AES-256 encryption and dual-machine hot standby, the modular design reduces maintenance costs, and the dual-machine hot standby ensures fault switching in less than 1 second.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A method for building a smart home linkage control scenario, characterized in that, include: S1. Environment preparation and hardware deployment: Building the Matter protocol network, network layering architecture and deploying edge computing nodes, performing device configuration protocol adaptation and AI interaction hardware integration; S2. Equipment Configuration and Protocol Adaptation: Matter protocol device network access and authentication, edge computing node configuration and resource allocation, and multi-protocol device compatibility handling; S3, Scene Logic Modeling and AI Intent Recognition: Multimodal data collection and processing are performed, followed by fusion of user behavior data, scene state modeling based on the data, language understanding and dynamic generation of scene rules; S4. Edge-side collaborative control: Local decision-making and conflict resolution for the scheduling and allocation of edge computing resources; S5, AI Model Deployment and Continuous Learning: Lightweight the model, use federated learning and model updates to achieve scene adaptation; S6. Iterative optimization: Federated learning model updates and dynamic tuning of the rule engine.

2. The method for building a smart home linkage control scene according to claim 1, characterized in that, The devices selected for protocol network setup in step S1 include: Border Router: Deploy Matter border routers that support Thread / Wi-Fi dual-mode to bridge the Thread Mesh network and the IP network; Terminal devices: Use Matter-certified chips that support Wi-Fi 6 / Thread protocols to ensure low power consumption and high concurrency connections; Security Authentication: The device must be pre-installed with a PKI certificate and can be configured for one-click network access via QR code or NFC. Users can complete the device's network access and security binding by scanning the QR code with a mobile app. The network layered architecture in step S1 includes: Physical layer: It adopts a hybrid networking of 2.4GHz and 5GHz dual-band Wi-Fi and Thread Mesh network. 2.4GHz is used to ensure compatibility with old devices, and 5GHz is used to ensure high-speed transmission. Network layer: Unifies device addressing through the IPv6 protocol; Thread border routers are responsible for protocol conversion. Application layer: Deploy the Home Assistant open-source platform to uniformly manage device clusters and scene rules.

3. The method for building a smart home linkage control scene according to claim 2, characterized in that, The device configuration protocol adaptation step in step S1 involves mapping device capabilities to a Matter standard cluster, converting old device data to Matter TLV format via a Zigbee to Matter gateway bridge, selecting an ICS-43432 multi-microphone module that supports beamforming and noise suppression to improve far-field speech recognition rate, deploying a MobileBERT lightweight speech recognition model on the edge gateway to support offline command parsing, and then performing sensor collaboration by inputting data from light sensors and human infrared sensors into the edge nodes, combining them with voice commands to generate composite intents, learning user habits through an LSTM model, preloading scene rules to reduce trigger latency, collecting fuzzy command samples, updating the edge node model through federated learning, and optimizing the recognition accuracy of temperature adjustment intents.

4. The method for building a smart home linkage control scene according to claim 3, characterized in that, The protocol device network access step S2 includes the following steps: Broadcast discovery: Devices seeking to join the network broadcast their own information via mDNS / DNS-SD. The Matter controller scans devices within the local area network and obtains the device's Discriminator identification code. Credential authentication: Users scan the device's QR code or enter a PIN code, and the controller establishes a secure temporary session with the device via the PAKE protocol to prevent man-in-the-middle attacks; The protocol device authentication in step S2 includes the following steps: Device certification: The controller extracts the DAC device certification certificate and PAI product certification intermediate certificate of the device to verify the legality of the device; Key negotiation: Generate NOC node operation certificate through CSR certificate signing request, install it on the device and configure network parameters; Fabric Join: Devices are assigned a unique node ID and joined to a specified Fabric security domain, enabling secure communication between devices within the domain; The edge computing node configuration in step S2 includes the following steps: Gateway selection: Use edge devices that support NPU acceleration and deploy a lightweight AI inference framework; Containerized deployment: Divide the system into containers using Docker / Kubernetes and run them separately; Matter protocol stack: handles device communication and state synchronization; OpenHab rule engine: executes local automated rules; AI model: MobileBERT compressed version used for intent recognition; Bandwidth management: Allocate high-priority bandwidth to security devices to ensure real-time response; Computational resource scheduling: AI inference tasks are allocated to the NPU, while the rule engine uses CPU resources to avoid resource contention; The multi-protocol device compatibility processing in step S2 includes the following steps: Unified data model: Define common attributes for devices to ensure consistent control across different brands of devices; Standardized interfaces: JSON / XML data format is used, and data is transmitted via HTTP / MQTT, supporting device control interfaces and data query interfaces; Communication encryption: Matter sessions use AES-256 encryption, and communication between devices establishes a secure channel through CASE certificate authentication sessions; Privacy protection: Sensitive data is processed locally at the edge node, with only de-identified metadata uploaded to the cloud; Role definition: Administrators have full privileges, regular users have device control privileges, and visitors can only view. Access control auditing: The edge gateway records operation logs and periodically detects abnormal access.

5. The method for building a smart home linkage control scene according to claim 4, characterized in that, The multimodal data acquisition and processing in step S3 includes the following steps: Environmental data acquisition: Environmental data is acquired in real time through DHT22 temperature and humidity sensor, BH1750 light sensor and HC-SR501 human infrared sensor, and transmitted to edge gateway via Matter over Thread protocol; Device status monitoring: Obtain the real-time status of the device and map it to the Matter standard Cluster attribute; The user behavior data fusion in step S3 includes the following steps: Voice command parsing: The microphone array collects speech, which is then converted into text by edge ASR and keywords are extracted; Visual behavior recognition: Camera data is processed by a lightweight CNN model on a Jetson Nano to recognize human posture and facial expressions; Time-series data alignment: Synchronize multi-source data using timestamps to construct a unified time-series feature vector; The scene state modeling in step S3 includes the following steps: Dynamic scene modeling: An indoor 3D structural model is constructed using SLAM, and combined with light intensity distribution data and regional temperature difference maps to generate a dynamic twin scene model; GNN (Graph Neural Network) Application: Constructing a relationship graph between devices, users, and the environment, where node attributes include device status and user habits, and edge weights are dynamically adjusted by an attention mechanism; Feature engineering optimization: Calculate device usage time and trigger interval, and add LSTM input sequence; Multi-scale feature fusion: Combining frequency domain features and spatial features to improve the robustness of scene classification.

6. The method for building a smart home linkage control scene according to claim 5, characterized in that, The language understanding in step S3 includes the following steps: Semantic understanding and slot filling: Parse user commands using an NLP module to extract intent slots; Context-aware correction: Combines dialogue history and environmental state to dynamically correct intent, and uses dynamic weight formula for sensor weighted scoring. Input data is normalized before weighted scoring. The dynamic generation of scene rules in step S3 includes the following steps: Rule engine configuration: Define basic scene logic, IF time∈[23:00-6:00] AND human sensor=no triggerTHEN start sleep mode; Reinforcement learning optimization: The PPO algorithm is adopted, with user satisfaction as the reward function, and the rule parameters are updated iteratively. AI intent translation scenario: Automatically mapping user fuzzy commands into device operation sequences; Habit learning prediction: The LSTM model learns users' historical behavior and preloads scene rules into the edge cache; The weighted scoring formula for the sensor is as follows: ； Where m is the semantic score, n is the sensor score, and w b For sensor weights.

7. The method for building a smart home linkage control scene according to claim 6, characterized in that, The edge computing resource scheduling and allocation in step S4 includes the following steps: Priority allocation: The edge gateway dynamically allocates computing resources and bandwidth based on device type and scenario requirements. Security devices are allocated high-priority resources to ensure response latency <100ms, comfort devices are allocated medium-priority resources, and sensors and energy-saving devices are allocated low-priority resources. Load balancing: Computing tasks are divided using Docker / Kubernetes containerization technology, and the Matter protocol stack, MobileBERT AI model inference, and OpenHab rule engine are deployed in separate containers to avoid resource contention; Task offloading strategy: Simple instructions are processed directly by edge nodes, while complex tasks are generated by a large AI model in the cloud and then sent to edge nodes for execution, reducing reliance on the cloud. Data preprocessing: Edge nodes filter and compress sensor data, uploading only key features to the cloud to reduce bandwidth consumption; The local decision-making and conflict resolution in step S4 includes the following steps: Rule engine deployment: Deploy a lightweight rule engine on the edge gateway to define local automation rules and ensure that core scenarios can still be executed when the network is down; Reinforcement learning optimization: Using user satisfaction as the reward function, the PPO algorithm is used to dynamically adjust rule parameters to adapt to changes in habits.

8. The method for building a smart home linkage control scene according to claim 7, characterized in that, The model lightweighting step in step S5 includes: Pruning: Removing redundant neurons or connections in a neural network to reduce model complexity and size; Quantization: Converts 32-bit floating-point parameters into 8-bit integers, reducing memory usage and improving inference speed; Knowledge distillation: Using BERT pre-trained large models to guide the training of MobileBERT lightweight models; The federated learning and model update steps in step S5 include: Edge-cloud collaborative training: Each household edge node periodically uploads model gradients, which are then aggregated in the cloud to generate a global model; Differential privacy protection: Laplace noise is added before gradient transmission to prevent leakage of user behavior data; OTA differential update: only transmits the differences in the model, reducing bandwidth usage; A / B testing verification: Run the old and new models in parallel, and select the version with high user satisfaction for full deployment.

9. The method for building a smart home linkage control scene according to claim 8, characterized in that, The federated learning model update in step S6 includes the following steps: Edge-cloud collaboration: Each household node uploads model gradients, which are aggregated in the cloud to generate a global model, and are OTA differentially updated every 24 hours. RLHF reinforcement learning: using user satisfaction as the reward function, it optimizes scene rule parameters through the PPO algorithm; The dynamic optimization of the rule engine in step S6 includes the following steps: Conflict resolution strategy: Based on a hierarchical arbitration mechanism of security > comfort > energy saving, the rule weights are dynamically adjusted; Energy efficiency adaptive: Automatically switches the device's operating mode by combining electricity price data and user habits.

10. A smart home linkage control scene construction system, applied to the smart home linkage control scene construction method described in claims 1-9, characterized in that, include: The module includes: device access and protocol adaptation module, edge computing and local decision-making module, scene logic and AI intent recognition module, user interaction and control module, and security and maintenance module. The device access and protocol adaptation module is used to achieve seamless access and interconnection of cross-brand devices through the Matter protocol's unified standard and multi-protocol compatibility layer, breaking down the barriers of the traditional smart home ecosystem. The edge computing and local decision-making module is used to utilize the local computing capabilities of edge nodes to achieve low-latency response and continuous operation of core functions in network outage scenarios, significantly reducing cloud dependence. The scene logic and AI intent recognition module is used to accurately parse users' ambiguous intents and dynamically generate scene rules through multimodal perception fusion and lightweight AI models. The user interaction and control module is used to provide multi-terminal interaction methods such as voice, APP, and central control screen, supporting one-click scene triggering and personalized rule configuration, simplifying user operation complexity. The security and maintenance module is used to ensure data security based on the Matter protocol's end-to-end encryption and device authentication mechanism, combined with OTA differential updates and redundancy design to ensure high system reliability.

Images