An AI digital human device linkage control method based on fusion of internet of things

Abstract

This invention discloses a method for coordinated control of AI digital human devices integrated with the Internet of Things (IoT), relating to the fields of IoT and AI digital human technology. The method includes: deploying multiple types of IoT sensors to collect data, transmitting it via the MQTT protocol to an edge gateway for cleaning and conversion, and then uploading it to an AI digital human control platform; the AI ​​digital human interprets user commands through voice and visual interaction, extracting target devices and control parameters; constructing a reinforcement learning-based coordinated decision-making model to generate control strategies, and issuing commands to the target devices via the LoRaWAN protocol; after the devices execute the commands, they provide feedback on status data, the platform calculates the execution accuracy, accumulates deviation data to fine-tune the model, and continuously improves the OCR recognition accuracy. This invention improves the stability of data transmission and the accuracy of command parsing for IoT devices, resulting in efficient multi-device coordinated response; it optimizes energy consumption and fault handling, adapts to multiple scenarios, aligns with user habits, and enhances the intelligence of device control and user experience.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things (IoT) and AI digital human technology, and in particular to a method for the linkage control of AI digital human devices that integrates IoT. Background Technology

[0002] With the trend of integrating IoT and AI technologies, AI digital humans are gradually becoming the core carrier of human-computer interaction, widely used in device control in home, office, and commercial scenarios. Currently, most AI digital human device control solutions only achieve simple command responses for single devices, such as controlling a single light switch or air conditioner temperature adjustment via voice commands. They fail to fully integrate the sensing data from IoT devices and the multi-device linkage logic, resulting in limited control scenarios and low levels of intelligence. For example, when a user issues the command "turn on movie mode," traditional solutions can only control a single device, requiring the user to issue additional commands such as turning off the lights or drawing the curtains. This fails to achieve multi-device collaborative response, significantly reducing user interaction efficiency and experience.

[0003] Existing IoT-integrated device control technologies have significant shortcomings in data interaction and decision-making logic. On one hand, IoT devices are diverse, employing vastly different communication protocols and data formats. This leads to high transmission latency and packet loss rates during edge gateway data collection, especially in dynamically moving mobile scenarios where data interruptions are frequent, impacting the real-time perception of device status by AI-powered digital humans. On the other hand, linkage decision-making models often rely on fixed rules or single user commands, failing to consider environmental data and user habits, resulting in a lack of flexibility and adaptability in control strategies. For example, when a user commands "turn on the air conditioner," the system directly activates the air conditioner without considering whether anyone is indoors or the outdoor temperature, easily leading to energy waste. Furthermore, it cannot automatically generate linkage strategies based on long-term user habits, making it difficult to meet users' personalized and intelligent control needs.

[0004] Furthermore, existing solutions have shortcomings in safety, energy consumption optimization, and fault handling. For high-risk equipment such as gas water heaters and electric ovens, there is a lack of pre-emptive safety verification mechanisms, relying solely on the equipment's own protection functions, which can easily lead to safety risks due to accidental triggering of commands. When multiple devices are linked, energy consumption optimization is not considered, operating only at the rated power of individual devices, resulting in high overall energy consumption. When equipment communication is interrupted or hardware failures occur, manual troubleshooting and repair are required, leading to low fault handling efficiency and a lack of fault recording and iterative diagnostic logic mechanisms, hindering continuous improvement in fault handling capabilities. These problems collectively restrict the large-scale application of AI-powered digital human device linkage control technology integrated with the Internet of Things, necessitating technological innovation to build more efficient, intelligent, and safer control solutions. Summary of the Invention

[0005] This invention proposes an AI digital human device linkage control method integrating the Internet of Things to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for coordinated control of AI-powered digital human devices integrating the Internet of Things (IoT) includes the following steps: IoT device sensing data collection steps: Deploy temperature and humidity sensors, human infrared sensors, and device status sensors to collect data; transmit the data to the edge gateway via the MQTT protocol, and the gateway cleans and converts the data before uploading it to the AI ​​digital human control platform; AI digital human multimodal interaction understanding steps: The AI ​​digital human integrates a voice interaction unit and a visual interaction unit, receives user interaction commands, parses the command intent through a pre-trained BERT semantic model, extracts the target device and control parameters, and outputs structured command information; The steps for constructing and operating the linkage decision-making model are as follows: Construct a linkage decision-making model based on reinforcement learning, input features, train the model using the PPO algorithm, and set the objective function as "user demand satisfaction + equipment energy consumption optimization"; when a user command is received, the model combines real-time data to generate a linkage control strategy and output a sequence of control commands. Target device linkage execution control steps: The AI ​​digital human control platform sends control commands to the target device via the LoRaWAN protocol. The commands include the device ID, control parameters, and execution time limit. After receiving the commands, the device sends back confirmation information and executes the control operation. Execution feedback and model optimization steps: After the IoT device completes execution, it collects device status data in real time and feeds it back to the AI ​​digital human control platform through the edge gateway; the platform compares the execution result with the target parameters and calculates the execution accuracy: execution accuracy = (1 - |actual value - target value| / target value) × 100%; when the execution accuracy is < 95%, the deviation data is recorded, and the linkage decision model is fine-tuned and the control strategy is optimized for every 100 deviation data points accumulated.

[0007] Furthermore, it also includes a step for dynamically adjusting the priority of equipment linkage, using a formula. Calculate the device linkage priority, where P is the device linkage priority, W1 is the device status weight, S is the device real-time status, W2 is the device-user distance weight, D is the straight-line distance between the device and the user, W3 is the user usage frequency weight, and U is the device's user usage frequency in the past 7 days.

[0008] Furthermore, it also includes coordinated execution of energy consumption optimization steps, through formulas. Calculate the total energy consumption for coordinated execution, where E is the total energy consumption for multi-device coordinated execution, n is the number of coordinated devices, and P... i T is the rated power of the i-th device.i Let be the linkage operation time of the i-th device, and K be the energy consumption optimization coefficient.

[0009] Furthermore, in the AI ​​digital human multimodal interaction understanding process, the visual interaction unit also integrates human posture recognition function: it adopts the MediaPipe posture detection model to recognize user body movements and identify 25 key points; when the user issues a voice command accompanied by a gesture, the system adopts a fusion judgment mechanism of "voice command as the main method and visual posture as the auxiliary method".

[0010] Furthermore, in the construction and operation steps of the linkage decision model, user habit feature parameters are also introduced: user habit features are extracted by analyzing the device usage data of users in the past 30 days through the K-means clustering algorithm; the habit features are used as input parameters of the decision model, and when the user does not give a clear instruction, the model automatically generates linkage strategy based on the habit features; when user habits change, the model re-clusters and updates the habit features every 7 days.

[0011] Furthermore, in the target device linkage execution control steps, a safety verification step is added for high-risk equipment: after receiving the control command, the equipment first detects the safety status through built-in sensors; if the safety status is not met, the equipment reports "safety verification failed", and the AI ​​digital human alerts the user through voice and vision; after the safety status is met, the equipment executes the command in stages to reduce safety risks.

[0012] Furthermore, the execution feedback and model optimization steps also include user satisfaction feedback collection: the AI ​​digital human collects user satisfaction through voice inquiries and visual interactive interfaces; when the user satisfaction score is ≤3 points, the system triggers human interactive inquiries and records the user feedback content; user satisfaction and feedback content are used as supplementary features for model fine-tuning.

[0013] Furthermore, in the IoT device sensing data collection step, Bluetooth 5.0 + NB-IoT dual-mode communication is adopted for mobile devices: when the device is stationary, it transmits data through NB-IoT, and automatically switches to Bluetooth 5.0 when it moves; the edge gateway adopts a dynamic caching strategy for mobile device data to avoid data loss.

[0014] Furthermore, it includes a multi-scene mode customization process: the AI ​​digital human control platform provides a visual scene mode editing interface, supporting users to customize scene names, linked device combinations, control parameters, and trigger conditions; the system performs dual rationality checks on customized scenes, namely parameter logic checks and energy consumption conflict checks; customized scenes are stored in the platform's encrypted database, which users can call up with one click via voice commands or the control platform; it also supports scene sharing, where users can generate a sharing code for customized scenes, authorize family members to import scenes via the sharing code, and support secondary editing and parameter adjustment after import.

[0015] Furthermore, it also includes a self - repair step for device linkage faults: An edge gateway is built - in with a fault diagnosis engine to monitor the device communication status in real - time. When a device communication interruption is detected, the fault diagnosis engine executes a diagnostic process according to priorities. The first priority is to check the network connection, the second priority is to check the device power supply, and the third priority is to check the device hardware. After the fault is repaired, the system starts an instruction re - transmission mechanism. At the same time, a fault knowledge base is constructed to record the type, occurrence time, diagnostic process, repair measures, and repair time consumption of each fault. The diagnostic logic of the knowledge base is updated quarterly based on new fault records. [[ID=X]] [[ID=X]]

[0016] Compared with the existing technologies, the beneficial effects of the present invention are as follows: [[ID=X]] In terms of intelligence and convenience, through multi - type IoT sensors to collect environmental, human activity, and device operation data, combined with the multi - modal interaction understanding (voice + vision) of an AI digital human, accurate parsing of user instructions and multi - device collaborative response are achieved. Complex scenario linkages can be completed without the user issuing instructions multiple times, greatly improving the interaction efficiency and user experience. At the same time, the analysis of user habit characteristics is introduced, and the linkage decision - making model can automatically generate control strategies based on the user's long - term usage habits, reducing the user's active operations and further enhancing the personalization and intelligence of the control scheme. [[ID=X]] [[ID=X]]

[0017] In terms of security and energy consumption optimization, the present invention adds a pre - safety verification link for high - risk devices. The safety status is detected through sensors built into the devices, and when the safety conditions are not met, an alarm is given in a timely manner and the startup is prohibited, avoiding safety risks from the source. When multiple devices are linked, energy consumption optimization calculations are introduced, and the operating parameters are dynamically adjusted in combination with the device power, operating time, and energy - saving coefficient to reduce the overall energy consumption on the premise of meeting user needs, achieving a balance between environmental protection and economy. In addition, the dynamic data caching and dual - mode communication strategy of the edge gateway ensure the stability of mobile device data transmission, reduce data packet loss and interruption, and provide reliable data support for the AI digital human to perceive the device status in real - time. [[ID=X]] [[ID=X]]

[0018] In terms of fault handling and system sustainability, the device linkage fault self - repair mechanism quickly locates and repairs faults such as communication, power supply, and hardware through a hierarchical diagnostic process, reducing manual intervention and shortening the fault handling time. The construction of the fault knowledge base and the iteration of the diagnostic logic continuously improve the accuracy of system fault diagnosis and enhance the long - term operation stability. The multi - scenario mode customization function supports users to flexibly configure linked devices, parameters, and trigger conditions, adapting to the needs of multiple scenarios such as home and office, and supports scenario sharing and secondary editing, improving the versatility and scalability of the solution, and laying a foundation for the wide application of the AI digital human device linkage control technology integrating the Internet of Things. [[ID=X]] [[ID=X]] [[ID=X]]

[0019] [[ID=X]] Figure 1This is a schematic block diagram of an AI digital human device linkage control method integrating the Internet of Things proposed in this invention; Figure 2 A comparison chart showing the accuracy of AI digital human command parsing under different interaction methods; Figure 3 A comparison chart of data transmission performance of IoT devices; Figure 4 This is a comparison chart of equipment linkage energy consumption and fault handling. Detailed Implementation

[0020] 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.

[0021] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0023] Reference Figures 1 to 4 A method for coordinated control of AI digital human devices integrating the Internet of Things includes the following steps: IoT device sensing data acquisition steps: First, deploy multiple types of IoT sensors, including the SHT30 temperature and humidity sensor, the HC-SR501 human infrared sensor, and the RS485 device status sensor. These sensors collect environmental data, human activity data, and device operation data. Environmental data includes temperature and humidity, with a temperature range of -20℃ to 60℃ and a humidity range of 0% to 100%, with a sampling frequency of 10 seconds per sampling. Human activity data includes the detection distance and response time for human presence, with a detection distance of 3m to 7m and a response time ≤0.5 seconds. Device operation data includes voltage and current, with a voltage of 220V±10% and a current of 0 to 10A, with a sampling frequency of 5 seconds per sampling.

[0024] The collected data is then transmitted to the edge gateway via the MQTT protocol, with a data transmission latency of ≤100ms and a packet loss rate of ≤10%. -5 The edge gateway first cleans the received data, filtering out outliers that exceed the measurement range, such as marking data with a temperature > 60℃ as invalid data. Then, it performs format conversion, uniformly converting the data into JSON format. Finally, it uploads the processed data stream to the AI ​​digital human control platform.

[0025] AI Digital Human Multimodal Interaction Understanding Steps: The AI ​​digital human integrates a voice interaction unit and a visual interaction unit. The voice interaction unit adopts the ASR algorithm with a recognition accuracy of ≥95% and supports Mandarin Chinese as well as dialects such as Cantonese and Sichuanese. The visual interaction unit is equipped with the YOLOv8 lightweight model, which can recognize user gestures such as waving and pointing, as well as facial expressions such as smiling and frowning, with a recognition response time of ≤300ms.

[0026] The AI ​​digital human receives user interaction commands, such as voice commands like "turn on the living room air conditioner and set it to 26℃" and visual commands like pointing to lighting devices. It then uses a pre-trained BERT semantic model to parse the command intent. This model's training dataset contains 100,000 home control commands. During parsing, it extracts the target device (e.g., "living room air conditioner") and control parameters (e.g., "26℃"), ultimately outputting structured command information in the format {"Device Type": "Air Conditioner", "Device ID": "AC-001", "Parameters": "26℃"}.

[0027] The steps for constructing and operating the linkage decision-making model are as follows: A linkage decision-making model based on reinforcement learning is constructed. Input features include user interaction commands, real-time status of IoT devices (such as current air conditioner temperature and light brightness), and environmental data (such as outdoor temperature and number of people indoors). The model training uses the PPO algorithm, with 1000 iterations and a learning rate of 1e-4. The objective function is "user demand satisfaction + device energy consumption optimization".

[0028] When a user command is received, the model combines real-time data to generate a linkage control strategy. For example, if the user command is "turn on the air conditioner", and no one is detected indoors, the strategy is adjusted to "first turn on the human body sensor, and then turn on the air conditioner after a person is detected", and outputs a sequence of control commands, including the execution order and parameter settings.

[0029] The target device executes control steps in a coordinated manner: The AI ​​digital human control platform sends control commands to the target device via the LoRaWAN protocol. This protocol has a communication range of 1km~3km, strong anti-interference capabilities, and the commands include the device ID, control parameters, and execution time limits, such as "complete temperature adjustment within 10 seconds." After receiving the command, the device sends back confirmation information, such as "AC-001 has received the temperature adjustment command," and executes the control operations, such as starting the air conditioner compressor, adjusting the fan speed to medium, and gradually bringing the temperature closer to 26℃. Simultaneously, it supports multi-device collaborative execution. For example, the "movie mode" command triggers "turn off the main light, turn on the ambient light, increase the audio volume, and draw the curtains," with a device synchronization deviation of ≤500ms.

[0030] Execution feedback and model optimization steps: After the IoT device completes execution, it collects device status data in real time (such as the current temperature of the air conditioner and the brightness of the lights), and feeds it back to the AI ​​digital human control platform through the edge gateway (feedback delay ≤200ms); the platform compares the execution result with the target parameter (such as the actual temperature of the air conditioner being 26.5℃ and the target temperature being 26℃, with a deviation of 0.5℃), and calculates the execution accuracy (execution accuracy = (1 - |actual value - target value| / target value) × 100%); when the execution accuracy is <95% (such as the lights not being adjusted to 50% brightness as instructed), the deviation data is recorded, and the linkage decision model is fine-tuned every 100 deviation data points (fine-tuning iterations are 50 rounds) to optimize the control strategy and improve the accuracy of subsequent linkage.

[0031] This invention also includes a step for dynamically adjusting the device linkage priority, using a formula. Calculate the device linkage priority, where P is the device linkage priority (value from 0 to 10, the larger the value, the higher the priority), W1 is the device status weight (fixed at 0.4, S=0 when the device is faulty, S=1 when it is normal), S is the real-time status of the device (1 indicates normal, 0 indicates fault), W2 is the device-user distance weight (fixed at 0.3), D is the straight-line distance between the device and the user (unit: m, value from 0.5 to 10), W3 is the user usage frequency weight (fixed at 0.3), and U is the device's user usage frequency in the past 7 days (unit: times / day, value from 0 to 20). For example, if the user commands "turn on the cooling device", the calculation for the living room air conditioner (S=1, D=3m, U=15) is P=0.4×1+0.3×(1 / 3)+0.3×15=0.4+0.1+4.5=5.0, and the calculation for the bedroom air conditioner (S=1, D=8m, U=5) is P=0.4×1+0.3×(1 / 8)+0.3×5≈0.4+0.0375+1.5=1.9375. The system prioritizes the activation of the living room air conditioner to ensure that the control response meets the user's immediate needs.

[0032] This invention also includes a linked execution energy consumption optimization step, using a formula. Calculate the total energy consumption for coordinated execution, where E is the total energy consumption for multi-device coordinated execution (unit: kWh), n is the number of coordinated devices (value from 1 to 10), and P... i Let T be the rated power of the i-th device (unit: kW, e.g., 1.5kW for an air conditioner, 0.06kW for a light), and T be the rated power of the i-th device. i Let be the linkage operation time of the i-th device (in hours), and K be the energy consumption optimization coefficient (valued from 0.8 to 1.0, with 0.8 for energy-saving mode and 1.0 for normal mode). For example, in "movie viewing mode," the following devices are linked: air conditioner (P=1.5kW, T=2h), ambient light (P=0.06kW, T=2h), and audio system (P=0.2kW, T=2h). With K=0.9, the calculated energy consumption is E=(1.5×2+0.06×2+0.2×2)×0.9=(3+0.12+0.4)×0.9=3.52×0.9=3.168kWh. Based on this, the system adjusts the device operating parameters (e.g., reducing the air conditioner fan speed to a low speed) to reduce total energy consumption while meeting user needs.

[0033] In this invention, the visual interaction unit integrates human posture recognition functionality within the multimodal interaction understanding process of the AI ​​digital human. It employs the MediaPipe posture detection model to identify user body movements, such as "crossing arms" indicating a pause, and "waving" indicating confirmation. It can identify 25 key points, including the head, shoulders, and hands, with a posture recognition accuracy of ≥92%. When a user issues a voice command accompanied by a gesture, the system uses a fusion judgment mechanism that prioritizes voice commands and uses visual posture as a secondary factor, with a voice weight of 0.7 and a visual weight of 0.3. For example, if a user says "turn on the lights" while waving (confirmation gesture), the interaction understanding result is "execute the turn on the lights command immediately"; if the user says "turn on the lights" while crossing their arms (pause gesture), the interaction understanding result is "pause turning on the lights, wait for user confirmation," thus avoiding accidental control triggering.

[0034] In this invention, user habit feature parameters are also introduced in the construction and operation steps of the linkage decision model: the device usage data of users in the past 30 days are analyzed by K-means clustering algorithm. The number of clusters in K-means clustering algorithm is 5 to 8. The device usage data includes the time when the air conditioner is turned on, the temperature setting, and the brightness and duration of the lights on every day. User habit features are extracted by analysis, such as "the user turns on the living room air conditioner after 20:00 every night and sets the temperature to 25℃".

[0035] The extracted habitual features are used as input parameters for the decision-making model, with a weight of 0.2. When the user does not give a specific instruction, the model automatically generates a linkage strategy based on the habitual features. For example, if someone is detected in the living room at 8:00 PM, the model automatically triggers "turn on the air conditioner to 25°C". When user habits change, such as changing the temperature setting from 25°C to 27°C after the transition from summer to autumn, the model performs cluster analysis every 7 days to update the habitual features, ensuring that the linkage strategy matches the user's dynamic needs.

[0036] In this invention, a safety verification step is added for high-risk equipment (such as gas water heaters and electric ovens) during the target device linkage execution control steps: After receiving the control command, the device first detects the safety status through built-in sensors. For gas water heaters, the gas leakage concentration is detected, with a safety threshold of ≤0.1%VOL; for electric ovens, the door closure status is detected, with closure recorded as 1 and open as 0. If the safety status is not met, for example, if the gas concentration reaches 0.15%VOL or the oven door is not closed, the device will report "safety verification failed." At this time, the AI ​​digital human will alert the user through both voice (e.g., "Gas concentration is abnormal, do not start the water heater for now") and visual (e.g., display a safety warning icon on the screen). After the safety status is met, the device executes the command in stages. For example, the electric oven preheats for 5 minutes, and then starts the baking mode after the temperature reaches the set value, thereby reducing safety risks.

[0037] In this invention, the feedback and model optimization steps also include user satisfaction feedback collection: the AI ​​digital human collects user satisfaction (a score of 1 to 5, with 5 being the most satisfactory) through voice inquiries ("Are you satisfied with the current device linkage effect?") and a visual interactive interface (displaying "Satisfied / Average / Unsatisfied" options on the screen); when user satisfaction is ≤3, the system triggers a human interaction inquiry ("Please point out the unsatisfactory aspects, such as temperature deviation or response speed"), and records the user feedback (e.g., "The air conditioner cools down too slowly"); user satisfaction and feedback content are used as supplementary features for model fine-tuning (weight 0.15). For example, for the feedback "cooling down too slowly", the air conditioner linkage strategy is fine-tuned (e.g., the fan speed is set to high speed at startup, and then the speed is reduced after the temperature approaches the target), improving the user experience.

[0038] In this invention, during the IoT device sensing data collection step, Bluetooth 5.0 + NB-IoT dual-mode communication is used for mobile devices (such as smart bracelets and robot vacuums): when the device is stationary, data is transmitted via NB-IoT (low power consumption, transmission interval 30s / time), and when it moves, it automatically switches to Bluetooth 5.0 (low latency, transmission interval 5s / time); the edge gateway adopts a dynamic caching strategy for mobile device data (caching capacity 100MB, data timeout time 5min) to avoid data loss; for example, when the robot vacuum moves to clean, it transmits its location (accuracy ±1m) and cleaning status (cleaning / paused) in real time via Bluetooth 5.0, and the gateway caches the data and synchronizes it to the platform in real time to ensure that the AI ​​digital human can adjust its cleaning path in real time (such as the user command "avoid the living room carpet", the platform immediately issues a path adjustment command).

[0039] This invention also includes a multi-scene mode customization step: the AI ​​digital human control platform provides a visual scene mode editing interface, which supports drag-and-drop operation. Users can drag device icons to the scene list, and when setting parameters, a value adjustment slider or option box will pop up; the interface supports user-defined scene names (such as "parent-child game mode" and "work mode"), linkage device combinations (supporting up to 15 devices, covering types such as air conditioners, lights, curtains, and projectors), control parameters (such as light brightness adjustable from 0% to 100%, air conditioner temperature adjustable from 16℃ to 30℃, and curtain opening degree adjustable from 0% to 100%), and trigger conditions (timed triggering accurate to the minute, such as "starting at 19:30 every day"; event triggering such as "triggered when user returns home (sensed by smart door lock)" and "triggered when indoor humidity > 60%").

[0040] The system performs dual validity checks on custom scenes: first, parameter logic checks to avoid unreasonable parameters such as air conditioner temperature settings <16℃ or >30℃, and light brightness settings <0% or >100%; second, energy consumption conflict checks. If "turn on air conditioner cooling" and "open window" are set simultaneously, the system will calculate the energy consumption difference (energy consumption is 30%~50% higher when there is a conflict than when there is no conflict) and pop up a message "There is an energy consumption conflict in the current scene. Do you want to continue saving?".

[0041] Custom scenes are stored in the platform's encrypted database, which uses AES-256 encryption and supports the storage of 100 scenes, each occupying ≤50KB of storage space. Users can invoke scenes with a single click via voice commands (such as "Start Parent-Child Game Mode") or by controlling the platform, with a response time of ≤1 second. Scene sharing is also supported; users can generate sharing codes for custom scenes (shared codes are valid for 24 hours), authorizing family members to import scenes using the sharing codes. After import, secondary editing and parameter adjustments are supported.

[0042] This invention also includes a device linkage fault self-repair step: the edge gateway has a built-in fault diagnosis engine that monitors the device communication status in real time (detection frequency 1 second / time, confirming the device's online status via heartbeat packets, heartbeat packet interval 500ms, 3 no heartbeat packets are considered a communication interruption); when a device communication interruption is detected, the fault diagnosis engine executes a diagnostic process according to priority. The first priority is to check the network connection (first check the LoRaWAN signal strength, threshold ≥ -85dBm, if the signal is weak, switch to WiFi / Bluetooth communication; then check the network link between the gateway and the platform, test connectivity via ping command, if packet loss rate > 10%, restart the gateway network module), the second priority is to check the device power supply (using voltage data fed back by the device status sensor to determine if there is a power outage (voltage < 180V) or overvoltage (voltage > 240V), if there is a power outage, trigger the AI ​​digital human voice prompt "[Device Name] Power Abnormal, please check the socket / switch"), the third priority is to check the device hardware (by adjusting... The system retrieves historical fault records of the equipment. If the same type of hardware failure occurs ≥2 times within the past 3 months, it is judged as hardware degradation, and a maintenance reminder of "Recommended to replace the [equipment name] hardware module" is pushed to the user's mobile APP. After the fault is repaired, the system starts a command retransmission mechanism (retry 3 times, with an interval of 5 seconds between each retry. The complete command is retransmitted on the first retry. If it fails, the command parameters are simplified (e.g., only the core control parameters are retained, and unnecessary additional parameters are removed) and retransmitted a second time. If it still fails, a manual intervention prompt is triggered. At the same time, a fault knowledge base is built to record the type of each fault (communication fault / power fault / hardware fault), occurrence time, diagnosis process, repair measures, and repair time. The knowledge base is updated quarterly based on newly added fault records (≥50 records). For example, for the "LoRa signal interruption" problem that frequently occurs in a certain brand of air conditioner, a diagnostic branch of "prioritize switching to Bluetooth communication" is added. This improves the fault diagnosis accuracy from the initial 85% to over 95%, and reduces the fault repair time from an average of 5 minutes to less than 2 minutes.

[0043] The following two examples further illustrate the specific implementation of this system: Example 1: Smart Home Scene (Daily Interconnection and Control of Multiple Devices) This embodiment targets smart home scenarios and applies an AI-powered digital human device linkage control method that integrates the Internet of Things. It addresses the core issues of low accuracy, high energy consumption, and cumbersome fault handling in multi-device collaboration within home scenarios, and refines the technical parameters and practical procedures for each step.

[0044] I. Hardware and Software Deployment Hardware configuration: Multiple types of IoT sensors are deployed. In the living room / bedroom, a SHT30 temperature and humidity sensor (temperature measurement range -20℃~60℃, accuracy ±0.2℃; humidity 0%~100%, accuracy ±2%) and an HC-SR501 human infrared sensor (detection distance 3m~7m, response time 0.3s) are used. In the kitchen, an RS485 device status sensor (monitoring gas water heater voltage 220V±10%, current 0~5A) and an MQ-4 gas concentration sensor (detection range 0~1%VOL, accuracy ±0.05%VOL) are used. The edge gateway is a Huawei AG150 (supporting MQTT / LoRaWAN / WiFi / Bluetooth 5.0 protocols, transmission latency ≤80ms, data packet loss rate ≤10%). -6 AI digital human devices are customized terminals (equipped with an 8-inch touch screen, supporting microphone array (4 microphones, pickup distance 5m), camera (1080P resolution, frame rate 30fps)); robot vacuum cleaners (supporting Bluetooth 5.0 + NB-IoT dual mode), smart air conditioners (LoRaWAN communication module, address code AC-001~AC-003) and other devices access gateways.

[0045] Software parameters: The AI ​​digital human voice interaction unit adopts the iFlytek ASR algorithm; the visual interaction unit deploys the YOLOv8-nano model (gesture recognition response time 250ms, accuracy 93%) and the MediaPipe posture detection model; the linkage decision model is trained based on the PPO algorithm (1000 iterations, learning rate 1e-4, training dataset contains 50,000 home control samples); user habit analysis adopts K-means clustering (6 clusters, analyzing 100,000 device usage data in the past 30 days); the fault diagnosis engine has 100+ built-in fault rules, and the knowledge base is stored using AES-256 encryption.

[0046] II. Complete Implementation Process IoT device sensing data collection: SHT30 collects living room temperature (26℃) and humidity (50%) every 10 seconds; HC-SR501 detects someone in the living room (4m away); RS485 collects gas water heater voltage (220V) and current (1.2A) every 5 seconds. Data is transmitted to the gateway via MQTT protocol (70ms delay). The gateway filters out abnormal values ​​with temperature > 60℃ (currently no abnormalities are found), converts the data to JSON format (e.g., {"Device ID":"SHT30-001","Type":"Temperature and Humidity","Temperature":26,"Humidity":50,"Timestamp":20241001190000}), and uploads it to the AI ​​digital human control platform. When the robot vacuum is stationary, it transmits its location (living room (3,4)m) every 30 seconds via NB-IoT. When moving, it switches to Bluetooth 5.0 and transmits its cleaning status (cleaning in progress) every 5 seconds. The gateway dynamically caches the data (100MB cache capacity, 5min timeout) to prevent data loss.

[0047] AI Digital Human Multimodal Interaction Understanding: The user's voice command "Start movie viewing mode" (while waving to confirm) is recognized by ASR (accuracy 97%), and the visual unit recognizes the "waving" gesture (confidence 0.9). The user uses a fusion judgment of "voice weight 0.7 + visual weight 0.3" to interpret the intent as "link the living room air conditioner, lights, curtains, and speakers". The user extracts parameters (air conditioner 26℃, light brightness 20%, curtains closed, speaker volume 50%) and outputs a structured instruction ({"scene":"movie viewing","device":[{"ID":"AC-001","parameter":26},{"ID":"Light-001","parameter":20},{"ID":"Curtain-001","parameter":100},{"ID":"Speaker-001","parameter":50}]}).

[0048] Linked decision-making and prioritization / energy consumption calculation: The decision model is input with instructions, environmental data (occupants in the living room, temperature 26℃), and user habits (user prefers air conditioning at 26℃ when watching movies). The generated strategy is: first close the curtains, then adjust the lights / sound, and finally turn on the air conditioning. (Following the formula...) Calculate device priority: For the living room air conditioner (S=1, D=4m, U=18 times / day), the P = 0.4×1 + 0.3×(1 / 4) + 0.3×18 = 0.4 + 0.075 + 5.4 = 5.875. For the bedroom air conditioner (S=1, D=8m, U=5 times / day), the P = 0.4×1 + 0.3×(1 / 8) + 0.3×5 ≈ 0.4 + 0.0375 + 1.5 = 1.9375. The living room air conditioner takes priority. According to the formula... Calculate energy consumption: Air conditioner (1.5 kW, 2 h), lights (0.06 kW, 2 h), audio (0.2 kW, 2 h), K = 0.9, we get E = (the sum of the products of power and time) × 0.9 = (1.5×2 + 0.06×2 + 0.2×2)×0.9 = 3.52×0.9 = 3.168 kWh. Adjust the air conditioner fan speed to low.

[0049] Target device linkage execution and safety verification: The platform issues instructions through LoRaWAN (signal strength -75 dBm). The curtain closes within 500 ms after receiving the instruction (feedback "executed"), and the lights are adjusted to 20% brightness within 10 s. The gas water heater is not triggered, so no safety verification is required. If the user's instruction is "start the gas water heater", the MQ-4 detects the gas concentration of 0.08%VOL (≤0.1%VOL), the door is closed (status 1), and the safety verification passes. The water heater pre-starts for 1 minute first and then heats to the set temperature. The deviation of multi-device synchronous execution ≤ 300 ms, meeting the collaborative requirements of the "movie viewing mode".

[0050] Execution feedback and fault self-repair: After the device executes, the sensor feedback shows that the air conditioner is at 26.2℃ and the lights are at 20% brightness. The platform calculates the execution accuracy rate (1 - |26.2 - 26| / 26) × 100% ≈ 99.2% (≥95%). The AI digital human voice asks about the satisfaction, and the user selects "satisfied" (5 points). If a communication interruption of the living room lights is detected, the gateway discovers it within 1 s (3 heartbeats not received). The diagnosis process: First, check the LoRa signal (-90 dBm < -85 dBm), switch to WiFi (signal strength -60 dBm), and restore communication within 10 s. Record the fault (communication fault, repair time 10 s) in the knowledge base and update the diagnosis logic (preferably switch to WiFi).

[0051] Scene customization and model optimization: The user drags "air conditioner + humidifier" to the "sleep mode" through the touch screen, sets parameters (air conditioner at 24℃, humidifier at 50% humidity), and sets the timing trigger (23:00). The system verifies the parameters (the air conditioner at 24℃ is within 16℃ - 30℃, no energy consumption conflict), and saves the scene (encrypted storage, occupying 40 KB). The user generates a sharing code, and after the family member imports it, adjusts the humidity to 55%. Every time 100 deviation data are accumulated (such as the air conditioner temperature deviation > 0.5℃), the PPO model is slightly adjusted (iterated 50 rounds, learning rate 5e-5), and the accuracy rate is increased by 5%.

[0052] III. Performance verification table Table 1 The data in Table 1 is based on 100 smart home scenario linkage tests (including modes such as watching movies, sleeping, and cooking). Due to incompatible communication protocols and lack of coordination strategies in the traditional solution, the linkage response takes 3 - 5 seconds, and the coordination deviation reaches 1 - 2 seconds, which cannot meet the immediate needs of users. Through dual - mode communication and priority scheduling, the response time of the present invention is reduced to within 1 second, and the deviation is ≤300 ms, improving the interaction fluency. In terms of energy consumption, there is no optimization in the traditional solution, and the energy consumption per single linkage is more than 4.2 kWh. Through energy consumption formula calculation and parameter adjustment, the energy consumption of the present invention is reduced to below 3.2 kWh, meeting the home energy - saving requirements. The fault repair time is reduced from the minute level to the second level, reducing user waiting. The user satisfaction is increased to above 4.6 points, due to functions such as multi - modal interaction and scene customization, solving the pain points of the traditional solution such as "complicated operation and poor experience".

[0053] Example 2: Smart office scenario (batch device and energy - saving linkage control) This example is for the smart office scenario: There are 10 offices, each containing an air conditioner, lights, a projector, and a human body sensor. It is required to achieve batch startup during work, meeting linkage, and energy - saving shutdown after work, with more than 200 daily device linkages. Applying the method of the present invention, it mainly solves the problems of low processing efficiency, high energy consumption, and insufficient fault diagnosis coverage of batch devices in the office scenario, and refines the batch control process and energy - saving strategy.

[0054] I. Hardware and software deployment Hardware configuration: In each office, a human body sensor DS18B20 (detection distance 2m - 5m, response time 0.4s) and a temperature - humidity sensor AHT21 (accuracy ±0.3℃ / ±2%) are deployed. An edge gateway (Huawei EG8145V5, supporting access of more than 100 devices, LoRaWAN communication distance 1.5 km) is deployed in the corridor; AI digital human terminals are deployed at the front desk (supporting voice wake - up and touch operation) and on the computer clients in each office; The batch devices include 10 air conditioners (ID: AC - 001~AC - 010), 30 lights (Light - 001~Light - 030), 10 projectors (Projector - 001~Projector - 010), and the high - risk devices are the electric water heaters on each floor (Power - 001~Power - 003).

[0055] Software parameters: The linkage decision model adapts to batch data (input features include environmental data of 10 offices, iterating 1200 rounds); The energy - consumption optimization coefficient K is adjusted according to time periods (taking 0.9 from ⑨:00 - 18:00 during work and 0.8 after work); Scene customization supports batch scenarios such as "meeting mode" and "after - work mode" (linking up to 50 devices at most); The fault diagnosis engine supports parallel diagnosis of batch devices (processing faults of 10 devices simultaneously), and the knowledge base is updated quarterly (adding more than 50 new office equipment fault rules).

[0056] II. Complete Implementation Process Batch IoT data collection: At 8:50 AM, the AHT21 collects the temperature (24℃~25℃) and humidity (45%~50%) of 10 offices every 10 seconds, and the DS18B20 detects that 8 offices are occupied; the data is transmitted in batches to the gateway via LoRaWAN (delay 90ms, packet loss rate ≤10^-6), the gateway cleans up invalid data (such as the sensor in 1 office being offline, marked as "to be supplemented"), converts it into JSON and uploads it to the platform in batches; the electric water heater status sensor transmits voltage 220V and current 8A every 5 seconds, with no abnormalities.

[0057] AI Digital Human Batch Interaction and Decision-Making: The front-end AI digital human receives the "Work Mode" command (voice "Start all office equipment"), interpreting the intent as "Batch start air conditioners (26℃) and lights (80%) in occupied offices, and turn off equipment in unoccupied offices"; the decision-making model inputs batch environmental data and user habits (preferring 26℃ for air conditioners during work hours), generating a batch strategy: linking according to office numbers 1-10, prioritizing the start of equipment in occupied offices. The priority of equipment is calculated using the priority formula: for office 1 air conditioner (S=1, D=3m, U=20 times / day), P=0.4×1+0.3×(1 / 3)+0.3×20=0.4+0.1+6=6.5; for office 10 air conditioner (S=1, D=5m, U=8 times / day), P=0.4×1+0.3×(1 / 5)+0.3×8≈0.4+0.06+2.4=2.86, prioritizing the linkage of equipment in office 1.

[0058] Batch execution and energy consumption optimization: The platform sends out instructions in batches through the gateway (5 devices per batch, 200ms interval). The air conditioners in offices 1-8 start within 10 seconds (with feedback "executed"), and the lights are adjusted to 80% within 5 seconds. Before the electric water heater starts, the built-in sensor detects the water temperature (currently 25℃) and power status (normal). After the safety check is passed, the water heater heats up in stages (first to 50℃, then to 100℃). Calculate the batch energy consumption using the energy consumption formula: 10 air conditioners (1.5kW, 8h), 30 lights (0.06kW, 8h), 3 electric water heaters (2kW, 8h), K=0.9, so E=(10×1.5×8+30×0.06×8+3×2×8)×0.9=(120+14.4+48)×0.9=182.4×0.9=164.16kWh. Adjust the air conditioner fan speed (medium speed when people are present, low speed when no one is present), and the energy consumption will decrease by another 8%.

[0059] Meeting Linkage and Fault Self-Repair: At 14:00, the user in Office 5 triggers "Meeting Mode" via the client, customizing parameters (air conditioning 25℃, lighting 50%, projector on, curtains closed). The system verifies no conflicts (no logical contradictions between projector and lighting parameters), saves the scene, and starts it with one click. Device synchronization execution deviation is ≤400ms. If the air conditioning communication in Office 3 is interrupted, the gateway performs parallel diagnosis: first checks the LoRa signal (-95dBm), then switches to WiFi (-65dBm), and restores it within 15 seconds; simultaneously, it detects a power outage in the lights of Office 7 (voltage 170V < 180V), and the AI ​​digital human voice prompts "Office 7 light power supply abnormality," which is repaired by maintenance personnel within 1 minute; the fault is recorded in the knowledge base, and a new rule "prioritize notification to maintenance personnel for office light power outages" is added.

[0060] After-get off work energy saving and model optimization: At 18:00, the DS18B20 detects that all offices are unoccupied, and the AI ​​digital human automatically triggers the "after-get off work mode," batch-turning off air conditioners, lights, and projectors (leaving only the corridor emergency lights on), and switching electric water heaters to keep-warm mode (K=0.8); the platform collects the execution results, calculating a batch execution accuracy of 98% (2 projectors were not turned off, requiring manual shutdown). For every 100 batch deviation data points accumulated (such as 5 offices where air conditioners were not turned off on time), the decision-making model is fine-tuned (iterated 50 times), improving the batch control accuracy by 6%; scene customization supports inter-department sharing, the marketing department shares the "product meeting mode" to the R&D department, imports it, and adjusts the projector resolution.

[0061] III. Performance Verification Form Table 2 Table 2 data is based on a 30-day smart office scenario test (an average of 10 batch linkages and 5 meeting linkages per day). Traditional solutions have low batch processing efficiency, at 30-40 units / minute, requiring 3-4 minutes to start up at the start of the workday, impacting office efficiency. This invention, through batch command issuance and priority scheduling, improves efficiency to 100 units / minute, requiring only 1 minute to start up at the start of the workday, aligning with the work rhythm. In terms of energy consumption, traditional solutions require manual shutdown of equipment at the end of the workday, consuming over 220kWh per day. This invention automatically triggers energy-saving mode, reducing energy consumption to below 165kWh, saving over 20,000 yuan in electricity costs annually. Meeting mode accuracy is improved from 80% to 98%, avoiding meeting delays due to equipment not being linked; fault diagnosis coverage is improved to over 95%, reducing "hidden faults"; scene call response time is reduced to within 1 second, improving operational convenience and solving the core pain points of traditional office control: "low efficiency, high energy consumption, and frequent faults."

[0062] Reference Figure 2This diagram clearly demonstrates the advantages of the multimodal interaction understanding of this invention, solving the problems of traditional solutions relying on single voice and low accuracy in parsing complex commands. Traditional solutions, with only single voice interaction, achieve an accuracy of only 72.3% for complex commands and lack the ability to integrate voice with gestures / postures. This invention, through the fusion of ASR voice recognition with YOLOv8 gestures and MediaPipe posture detection, improves the parsing accuracy of various commands to over 94%, with an improvement of over 20% for complex commands and fused interaction methods. For example, for complex commands like "start movie mode" that require multi-device interaction, this invention achieves an accuracy of 95.8%, ensuring that users can achieve multi-device collaboration without repeating commands, significantly improving interaction efficiency and experience in smart home and office scenarios, and avoiding device control deviations caused by mis-parsing of commands.

[0063] Reference Figure 3 This figure verifies the transmission optimization effect of the IoT data acquisition stage of this invention, solving the problems of high latency and high packet loss rate of mobile devices in traditional solutions. Traditional solutions use a single NB-IoT device, resulting in latency exceeding 300ms, and batch device transmission requires over 500ms, making it difficult to meet real-time control requirements. This invention optimizes LoRaWAN parameters (signal strength -75dBm) for fixed devices, uses Bluetooth 5.0 + NB-IoT dual-mode for mobile devices (switching to Bluetooth 5.0 when moving), shortens the transmission interval between high-risk devices and sensors, and optimizes gateway caching (dynamic cache 100MB), reducing latency to below 90ms, with mobile device latency even as low as 80ms, and packet loss rate reduced to below 0.15%. Batch device transmission latency is reduced to 200ms, ensuring rapid linkage of batch devices in scenarios such as commuting and leaving get off work, providing reliable data support for AI digital humans to perceive device status in real time and generate precise control strategies, avoiding delays in linkage strategies due to data latency.

[0064] Reference Figure 4 This diagram illustrates the advantages of this invention from multiple dimensions, including energy consumption, fault handling, safety, and user experience, addressing the problems of high energy consumption, inefficient fault handling, and insufficient safety verification in traditional solutions. Regarding energy consumption, this invention utilizes a formula... By calculating and adjusting device parameters, energy consumption in a 2-hour movie-watching mode is reduced by approximately 30%. Regarding fault handling, traditional solutions have a diagnostic accuracy rate of less than 70% and require over 5 minutes for repair. This invention, through layered diagnosis and iterative fault knowledge base updates, improves accuracy to over 95% and reduces repair time to within 30 seconds. Safety verification achieves 100% coverage, with dual system and device verification preventing accidental activation of high-risk devices. User satisfaction is improved to over 4.6 points, stemming from comprehensive optimization of energy consumption, fault handling, and safety, providing an efficient, safe, and economical solution for the coordinated control of AI-powered digital human devices integrated with the Internet of Things.

[0065] The above are merely preferred embodiments 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 inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for coordinated control of AI digital human devices integrating the Internet of Things, characterized in that, Includes the following steps: IoT device sensing data collection steps: Deploy temperature and humidity sensors, human infrared sensors, and device status sensors to collect data; transmit the data to the edge gateway via the MQTT protocol, and the gateway cleans and converts the data before uploading it to the AI ​​digital human control platform; AI digital human multimodal interaction understanding steps: The AI ​​digital human integrates a voice interaction unit and a visual interaction unit, receives user interaction commands, parses the command intent through a pre-trained BERT semantic model, extracts the target device and control parameters, and outputs structured command information; The steps for constructing and operating the linkage decision-making model are as follows: Construct a linkage decision-making model based on reinforcement learning, input features, train the model using the PPO algorithm, and set the objective function as "user demand satisfaction + equipment energy consumption optimization"; when a user command is received, the model combines real-time data to generate a linkage control strategy and output a sequence of control commands. Target device linkage execution control steps: The AI ​​digital human control platform sends control commands to the target device via the LoRaWAN protocol. The commands include the device ID, control parameters, and execution time limit. After receiving the commands, the device sends back confirmation information and executes the control operation. Execution feedback and model optimization steps: After the IoT device completes execution, it collects device status data in real time and feeds it back to the AI ​​digital human control platform through the edge gateway; the platform compares the execution result with the target parameters and calculates the execution accuracy: execution accuracy = (1 - |actual value - target value| / target value) × 100%; when the execution accuracy is < 95%, the deviation data is recorded, and the linkage decision model is fine-tuned and the control strategy is optimized for every 100 deviation data points accumulated.

2. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, It also includes a step for dynamically adjusting the priority of equipment linkage, using a formula. Calculate the device linkage priority, where P is the device linkage priority, W1 is the device status weight, S is the device real-time status, W2 is the device-user distance weight, D is the straight-line distance between the device and the user, W3 is the user usage frequency weight, and U is the device's user usage frequency in the past 7 days.

3. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, It also includes a coordinated energy consumption optimization step, implemented through formulas. Calculate the total energy consumption for coordinated execution, where E is the total energy consumption for multi-device coordinated execution, n is the number of coordinated devices, and P... i T is the rated power of the i-th device. i Let be the linkage operation time of the i-th device, and K be the energy consumption optimization coefficient.

4. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, In the AI ​​digital human multimodal interaction understanding process, the visual interaction unit also integrates human posture recognition function: it adopts the MediaPipe posture detection model to recognize user body movements and identify 25 key points; when the user issues a voice command accompanied by a gesture, the system adopts a fusion judgment mechanism of "voice command as the main and visual posture as the auxiliary".

5. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, In the construction and operation steps of the linkage decision model, user habit feature parameters are also introduced: user habit features are extracted by analyzing the user's device usage data for the past 30 days through the K-means clustering algorithm; the habit features are used as input parameters of the decision model, and when the user does not give a clear instruction, the model automatically generates linkage strategies based on the habit features; when user habits change, the model re-clusters and updates the habit features every 7 days.

6. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, In the target device linkage execution control steps, a safety verification step is added for high-risk equipment: after receiving the control command, the equipment first detects the safety status through built-in sensors; if the safety status is not met, the equipment reports "safety verification failed", and the AI ​​digital human alerts the user through voice and vision; after the safety status is met, the equipment executes the command in stages to reduce safety risks.

7. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, The execution feedback and model optimization steps also include user satisfaction feedback collection: the AI ​​digital human collects user satisfaction through voice inquiries and visual interactive interfaces; when the user satisfaction score is ≤3 points, the system triggers human interaction inquiries and records the user feedback content; user satisfaction and feedback content are used as supplementary features for model fine-tuning.

8. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, In the data collection process of IoT devices, Bluetooth 5.0 + NB-IoT dual-mode communication is used for mobile devices: data is transmitted via NB-IoT when the device is stationary, and automatically switches to Bluetooth 5.0 when the device moves. Edge gateways employ a dynamic caching strategy for mobile device data to prevent data loss.

9. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, It also includes a multi-scene mode customization process: the AI ​​digital human control platform provides a visual scene mode editing interface, supporting users to customize scene names, linked device combinations, control parameters, and trigger conditions; the system performs dual rationality checks on customized scenes, namely parameter logic checks and energy consumption conflict checks; customized scenes are stored in the platform's encrypted database, which users can call up with one click via voice commands or the control platform; it also supports scene sharing, where users can generate a sharing code for customized scenes, authorize family members to import scenes via the sharing code, and support secondary editing and parameter adjustment after import.

10. The method for coordinated control of AI digital human devices integrating the Internet of Things according to claim 1, characterized in that, It also includes a device linkage fault self-repair step: the edge gateway has a built-in fault diagnosis engine that monitors the device communication status in real time; when a device communication interruption is detected, the fault diagnosis engine executes the diagnosis process according to priority, with the first priority checking the network connection, the second priority checking the device power supply, and the third priority checking the device hardware; after the fault is repaired, the system starts the command retransmission mechanism; at the same time, a fault knowledge base is built to record the type of each fault, the time of occurrence, the diagnosis process, the repair measures, and the repair time. The knowledge base is updated quarterly based on the newly added fault records to update the diagnostic logic.

Images