Smart home non-inductive interaction system and method based on UWB positioning
By combining UWB positioning and multi-sensor data fusion with LSTM neural networks, high-precision seamless interaction and automated control of smart home devices have been achieved, solving the problems of cumbersome interaction and insufficient positioning accuracy in existing technologies, and improving user experience and security.
Patent Information
- Application Number
- CN202511125939.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-12
- Publication Date
- 2025-12-30
AI Technical Summary
Existing smart home interaction methods are cumbersome and inconvenient, lack positioning accuracy, infringe on user privacy, and are greatly affected by the environment.
By employing UWB positioning technology combined with multi-sensor data fusion and artificial intelligence algorithms, high-precision positioning is achieved through UWB tags and edge computing gateways. Combined with LSTM neural networks, behavioral intent recognition is performed to automatically control smart home devices.
It enables seamless interaction with smart home devices, automated scene control, and improves convenience, comfort, and security, while protecting user privacy.
Smart Images
Figure CN121239735A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart home technology, and in particular to a seamless smart home interaction system and method based on UWB positioning. Background Technology
[0002] Current smart home interaction methods have many limitations. Taking mobile APP control as an example, users need to find and open the corresponding APP on their mobile phones before operating it, which is cumbersome and extremely inconvenient when their hands are occupied. Although voice control is relatively convenient, its recognition accuracy drops significantly in noisy environments, and different dialects and accents also affect the recognition effect. Physical button control lacks intelligence and personalization and cannot automatically adjust according to user behavior and needs.
[0003] Existing indoor positioning technologies generally have positioning accuracy at the level of several meters, which is difficult to meet the needs of precise interaction in smart homes. Although camera visual positioning can achieve high-precision positioning, it poses the risk of infringing on user privacy and is greatly affected by lighting conditions, performing poorly in low-light or backlight environments. In view of the above, this application proposes a seamless smart home interaction system and method based on UWB positioning. Summary of the Invention
[0004] Based on the technical problems existing in the background technology, the present invention proposes a smart home seamless interaction system and method based on UWB positioning.
[0005] This invention proposes a seamless smart home interaction system based on UWB positioning, comprising a user terminal, a UWB positioning base station network, an edge computing gateway, a smart home device cluster, a cloud service platform, and a user mobile APP. The user terminal is connected to the UWB positioning base station network via UWB signals, the edge computing gateway is wirelessly connected to the UWB positioning base station network and the user mobile APP, and the edge computing gateway is connected to the smart home device cluster and the cloud service platform.
[0006] Preferably, the UWB positioning base station network is set up in various rooms of the home, including the living room, bedroom, kitchen, and bathroom. Depending on the room size and shape, 3-4 UWB positioning base stations are reasonably deployed in each room in a distributed layout. They are installed in the corners of the ceiling or high on the walls to ensure wide signal coverage without dead spots. Each UWB positioning base station communicates with each other through a wired or wireless network to form a three-dimensional coordinate positioning network for obtaining the location information of the user terminal in real time.
[0007] The positioning principle of a UWB positioning base station network is based on time-of-flight (TOF) ranging technology. UWB tags periodically send pulse signals to surrounding UWB positioning base stations. Upon receiving the signal, the UWB positioning base station records the timestamp and calculates the distance between the tag and the base station based on the speed of signal propagation in the air. The formula used is:
[0008] Where Δt is the time difference between the signal being sent from the tag to the base station and received by the base station, and C is the speed of light;
[0009] After obtaining the distances between the tag and at least three UWB positioning base stations, the trilateration algorithm is used to calculate the tag's coordinates (x, y, z) in three-dimensional space, achieving centimeter-level high-precision positioning. The principle of the trilateration algorithm is as follows: given the coordinates of the three sides (x, y, z)... i ,y i ,z i If ), then the system will satisfy The user's location can be obtained by solving this system of equations with an accuracy of 5-10 centimeters.
[0010] Preferably, the user terminal is designed as a lightweight and portable UWB tag with a small appearance, which can be made into various forms such as a bracelet, brooch, and keychain, making it convenient for users to wear or carry. The tag integrates a UWB communication module, a nine-axis sensor, a low-power Bluetooth module, and a micro lithium battery. The nine-axis sensor includes a 3-axis accelerometer, a 3-axis gyroscope, and a 3-axis magnetometer, which can collect the user's motion acceleration, rotation angle, and magnetic field information in real time.
[0011] The working principle of the nine-axis sensor is as follows: Acceleration, angular velocity, and magnetic field data are collected and fused with UWB positioning data using a Kalman filter algorithm. Kalman filtering is an optimal estimation method based on a linear system state-space model. It continuously corrects the estimation of the system state through two steps: prediction and updating. The formula used in the prediction process is:
[0012] The formula used in its update process is:
[0013]
[0014] P k|k =(IK k H k )P k|k-1 ;
[0015] Where x^ k∣k-1 and x^ k∣k P represents the predicted and updated state estimates, respectively. k∣k-1 and Pk∣ k is the corresponding covariance matrix, F k Let B be the state transition matrix. k To control the input matrix, u k To control the input, H k Let z be the observation matrix. k For the observed value, Q k Let R be the process noise covariance matrix. k To observe the noise covariance matrix, K k For Kalman gain.
[0016] Preferably, the edge computing gateway adopts a high-performance hardware platform, equipped with a Rockchip RK3399 six-core processor, 2GB RAM and 16GB eMMC storage, runs the Ubuntu Core operating system, and deploys various applications based on Docker containerization technology. Its core functional modules include a UWB positioning data processing engine, an AI behavior intention recognition algorithm module, a scene automation rule engine, and a communication interface module. The AI behavior intention recognition algorithm module uses a deep learning-based Long Short-Term Memory (LSTM) network model to recognize user behavior based on the behavior intention recognition algorithm.
[0017] The smart home device cluster includes smart lights, smart air conditioners, smart TVs, and smart curtains. These devices establish connections with the edge computing gateway via Wi-Fi, Bluetooth, and ZigBee wireless communication technologies, based on the MQTT protocol, and receive and execute control commands sent by the gateway.
[0018] Preferably, the cloud service platform is built on a public or private cloud and adopts a distributed storage and computing architecture. Its main functions include storing users' historical behavior data, training and optimizing AI models, providing remote management interfaces, and protecting data security. The historical behavior data includes the user's time spent in different rooms and device usage habits. The training and optimization of AI models utilizes big data and machine learning algorithms to continuously improve the accuracy of behavioral intent recognition. By providing remote management interfaces, users can remotely view and manage home devices through a mobile APP and set personalized scenes. Data security protection adopts encrypted transmission and storage technologies to ensure user data security.
[0019] This invention also proposes a seamless smart home interaction method based on UWB positioning, comprising the following steps:
[0020] S1: System Startup and Initialization: After the system power is turned on, the edge computing gateway starts the Ubuntu Core system through the Rockchip RK3399 processor, loads the Docker containerized UWB positioning engine and AI algorithm module, and at the same time, the UWB positioning base station initializes the signal receiving channel. Smart home devices access the edge gateway through Wi-Fi / ZigBee and establish an MQTT communication link.
[0021] S2: User location awareness: The UWB tag worn by the user sends pulse signals to surrounding base stations at a frequency of 100-500 times / second, and the built-in nine-axis sensor synchronously collects motion data.
[0022] S3: User movement status determination: The edge gateway receives base station timestamp data, calculates three-dimensional coordinates through a trilateration algorithm, and integrates nine-axis sensor data to determine whether the user has moved. If the user has moved, the location coordinates will be updated; if the user has not moved, the current state will be maintained, and the system will immediately return to step two and continue monitoring.
[0023] S4: Location coordinate update and scene matching: When the user moves, the edge gateway updates the real-time coordinates and compares them with the preset scene area. The scene area calculates whether the coordinates fall into the area using a trilateration algorithm. If the coordinates match the preset scene, the edge gateway generates control commands through the scene rule engine and sends them to the corresponding device for execution via the MQTT protocol. If the coordinates do not match any scene, the device maintains its current state. At this time, it immediately returns to S2 to continue monitoring the location change.
[0024] S5: Behavioral Intent Analysis and Model Invocation: The edge gateway analyzes real-time user behavior through an LSTM neural network, combines it with time, and invokes a personalized interaction model generated in the cloud.
[0025] S6: Real-time push of customized services: Based on the model output results, the device performs personalized interaction. When the user approaches the TV, the gateway wakes up the TV through the HDMI-CEC protocol and pushes cloud-recommended movie and TV content. The device receives instructions through MQTT and executes hardware interface control.
[0026] S7: Real-time monitoring of user feedback: The system monitors feedback through changes in device status or tag sensor data. If there is feedback, it jumps to collect feedback data; if there is no feedback, it jumps to continue monitoring behavior and returns to S2.
[0027] S8: Cloud-based model reinforcement learning and update: The cloud uses reinforcement learning algorithms, with user satisfaction as the reward function, to optimize the LSTM model parameters and push the new model to the edge gateway, which then updates it in real time.
[0028] Preferably, in S5, the LSTM neural network includes a memory unit and three gate mechanisms. The LSTM neural network selectively retains and forgets historical information through the memory unit and gate mechanisms, making it suitable for processing time-series data such as user behavior. Its memory unit is used to store historical state information, serving as a "memory" carrier, allowing information to be transmitted over long time sequences without loss. It is also used to store the user's past position, speed, and acceleration features, and to determine the continuity of the current behavior.
[0029] The gating mechanism includes an input gate, a forget gate, and an output gate. The input gate determines whether new feature data at the current moment is stored in the memory unit. The forget gate controls the proportion of old information discarded in the memory unit. The output gate filters out features for behavior classification based on the state of the memory unit and the current input.
[0030] The logical steps of an LSTM neural network are as follows:
[0031] S501: Input layer data processing: Input features: fused user data, timestamp, room ID, and organized into a multi-dimensional vector according to time series;
[0032] S502: Memory Units and Gating Operations: Calculating the Forgetting Gating Value f using the sigmoid function (σ). t (Between 0 and 1) determines the historical state h in the memory unit. t-1 and the current input x t The forgetting rate is calculated using the formula: f t =(W f ·[h t-1 x t ]+b f );
[0033] And perform input gate calculation, using the formula: i t =(W i ·[h t-1 x t ]+b i ), Where i t Controlling new features The deposit ratio;
[0034] By combining the outputs of the forget gate and the input gate, the state C of the memory cell is updated. t It retains historical information related to the current behavior, using the following formula:
[0035] And perform output gate operations, using the following formula: o t =σ(W O ·[h t-1 xt ]+b o ), h t =o t ⊙tanh(C t ), where o t Controlling the output ratio of information in the memory unit generates the current hidden state h. t This is used for subsequent behavior classification;
[0036] S503: Output Layer and Behavior Classification: The output layer will include the hidden state h from the last time step. t The input is a fully connected layer, which is mapped to the behavior category dimension.
[0037] Preferably, in step S8, the cloud updates the weights of the LSTM neural network using gradient descent, and the edge gateway receives the model file via HTTP and overwrites the local version.
[0038] Compared with existing technologies, the beneficial effects of this invention are:
[0039] This invention utilizes UWB positioning technology, combined with multi-sensor data fusion and artificial intelligence algorithms, to achieve seamless interaction and automated scene control between smart home devices and users. By accurately locating the user's position and behavior, it enables automatic response and personalized interaction of smart home devices, improving the convenience, comfort, and security of users' home life while protecting user privacy. Attached Figure Description
[0040] Figure 1 This is a block diagram of a smart home seamless interaction system based on UWB positioning proposed in this invention;
[0041] Figure 2 This is a flowchart of a smart home seamless interaction method based on UWB positioning proposed in this invention. Detailed Implementation
[0042] The present invention will be further explained below with reference to specific embodiments.
[0043] Example
[0044] Reference Figure 1-2 This embodiment proposes a UWB-based smart home seamless interaction system, including a user terminal, a UWB positioning base station network, an edge computing gateway, a smart home device cluster, a cloud service platform, and a user mobile APP. The user terminal is connected to the UWB positioning base station network via UWB signals, the edge computing gateway is wirelessly connected to the UWB positioning base station network and the user mobile APP, and the edge computing gateway is connected to the smart home device cluster and the cloud service platform.
[0045] The UWB positioning base station network is set up in various rooms of the home, including the living room, bedroom, kitchen, and bathroom. Depending on the size and shape of the room, 3-4 UWB positioning base stations are reasonably deployed in each room in a distributed layout. They are installed in the corners of the ceiling or high on the walls to ensure wide signal coverage without dead spots. Each UWB positioning base station communicates with each other through a wired or wireless network to form a three-dimensional coordinate positioning network for obtaining the location information of the user terminal in real time.
[0046] The positioning principle of a UWB positioning base station network is based on time-of-flight (TOF) ranging technology. UWB tags periodically send pulse signals to surrounding UWB positioning base stations. Upon receiving the signal, the UWB positioning base station records the timestamp and calculates the distance between the tag and the base station based on the speed of signal propagation in the air. The formula used is:
[0047] Where Δt is the time difference between the signal being sent from the tag to the base station and received by the base station, and C is the speed of light;
[0048] After obtaining the distances between the tag and at least three UWB positioning base stations, the trilateration algorithm is used to calculate the tag's coordinates (x, y, z) in three-dimensional space, achieving centimeter-level high-precision positioning. The principle of the trilateration algorithm is as follows: given the coordinates of the three sides (x, y, z)... i ,y i ,z i If ), then the system will satisfy The user's location can be obtained by solving this system of equations with an accuracy of 5-10 centimeters.
[0049] The user terminal is designed as a lightweight, portable UWB tag. It is small in size and can be made into various forms such as a bracelet, brooch, and keychain for easy wear or carrying. The tag integrates a UWB communication module, a nine-axis sensor, a low-power Bluetooth module, and a micro lithium battery. The nine-axis sensor includes a 3-axis accelerometer, a 3-axis gyroscope, and a 3-axis magnetometer, which can collect the user's motion acceleration, rotation angle, and magnetic field information in real time.
[0050] The working principle of the nine-axis sensor is as follows: Acceleration, angular velocity, and magnetic field data are collected and fused with UWB positioning data using a Kalman filter algorithm. Kalman filtering is an optimal estimation method based on a linear system state-space model. It continuously corrects the estimation of the system state through two steps: prediction and updating. The formula used in the prediction process is:
[0051] The formula used in its update process is:
[0052]
[0053] P k|k =(IK k H k )P k|k-1 ;
[0054] Where x^ k∣k-1 and x^ k∣k P represents the predicted and updated state estimates, respectively. k∣k-1 and P k∣ k is the corresponding covariance matrix, F k Let B be the state transition matrix. k To control the input matrix, u k To control the input, H k Let z be the observation matrix. k For the observed value, Q k Let R be the process noise covariance matrix. k To observe the noise covariance matrix, K k Kalman gain;
[0055] The edge computing gateway adopts a high-performance hardware platform, equipped with a Rockchip RK3399 six-core processor, 2GB RAM and 16GB eMMC storage, and runs the Ubuntu Core operating system. It deploys various applications based on Docker containerization technology. Its core functional modules include a UWB positioning data processing engine, an AI behavior intention recognition algorithm module, a scene automation rule engine, and a communication interface module. Among them, the AI behavior intention recognition algorithm module uses a deep learning-based Long Short-Term Memory (LSTM) network model to identify user behavior based on the behavior intention recognition algorithm.
[0056] The smart home device cluster includes smart lights, smart air conditioners, smart TVs, and smart curtains. These devices establish connections with the edge computing gateway through Wi-Fi, Bluetooth, and ZigBee wireless communication technologies, based on the MQTT protocol, and receive and execute control commands sent by the gateway.
[0057] The cloud service platform is built on a public or private cloud and adopts a distributed storage and computing architecture. Its main functions include storing users' historical behavior data, training and optimizing AI models, providing remote management interfaces, and protecting data security. Its historical behavior data includes the time users spend in different rooms and their device usage habits. The training and optimization of AI models utilizes big data and machine learning algorithms to continuously improve the accuracy of behavioral intent recognition. By providing remote management interfaces, users can remotely view and manage home devices through a mobile APP and set personalized scenes. Data security protection adopts encrypted transmission and storage technologies to ensure the security of user data.
[0058] This embodiment also proposes a seamless smart home interaction method based on UWB positioning, including the following steps:
[0059] S1: System Startup and Initialization: After the system power is turned on, the edge computing gateway starts the Ubuntu Core system through the Rockchip RK3399 processor, loads the Docker containerized UWB positioning engine and AI algorithm module, and at the same time, the UWB positioning base station initializes the signal receiving channel. Smart home devices access the edge gateway through Wi-Fi / ZigBee and establish an MQTT communication link.
[0060] S2: User location awareness: The UWB tag worn by the user sends pulse signals to surrounding base stations at a frequency of 100-500 times / second, and the built-in nine-axis sensor synchronously collects motion data.
[0061] S3: User movement status determination: The edge gateway receives base station timestamp data, calculates three-dimensional coordinates through a trilateration algorithm, and integrates nine-axis sensor data to determine whether the user has moved. If the user has moved, the location coordinates will be updated; if the user has not moved, the current state will be maintained, and the system will immediately return to step two and continue monitoring.
[0062] S4: Location Coordinate Update and Scene Matching: When the user moves, the edge gateway updates the real-time coordinates and compares them with the preset scene area. The scene area uses a trilateration algorithm to calculate whether the coordinates fall within the area. If the coordinates match the preset scene, the edge gateway generates control commands (dim the lights, adjust the air conditioner temperature) through the scene rule engine and sends them to the corresponding device for execution via the MQTT protocol (for example, entering the bedroom triggers the "sleep preparation" scene, the gateway will send a PWM dimming command to the smart lights and an infrared encoded signal to the air conditioner). If the coordinates do not match any scene, the device maintains the current state (such as the light brightness remains unchanged). At this time, it immediately returns to S2 to continue monitoring the position change.
[0063] S5: Behavioral Intent Analysis and Model Invocation: The edge gateway analyzes real-time user behavior through an LSTM neural network, combines it with time, and invokes a personalized interaction model generated in the cloud.
[0064] LSTM neural networks include memory units and three gate mechanisms. LSTM neural networks selectively retain and forget historical information through memory units and gate mechanisms, making them suitable for processing time series data such as user behavior. Their memory units are used to store historical state information, acting as "memory" carriers that allow information to be transmitted over long time series without being lost. They are also used to store the user's past position, velocity, and acceleration features, and to determine the continuity of the current behavior.
[0065] The gating mechanism includes an input gate, a forget gate, and an output gate. The input gate determines whether new feature data at the current moment is stored in the memory unit. The forget gate controls the proportion of old information discarded in the memory unit. The output gate filters out features for behavior classification based on the state of the memory unit and the current input.
[0066] The logical steps of an LSTM neural network are as follows:
[0067] S501: Input layer data processing: Input features: fused user data, timestamp, room ID, and organized into a multi-dimensional vector according to time series;
[0068] S502: Memory Units and Gating Operations: Calculating the Forgetting Gating Value f using the sigmoid function (σ). t (Between 0 and 1) determines the historical state h in the memory unit. t-1 and the current input x t The forgetting rate is calculated using the formula: f t =(W f ·[h t-1 x t ]+b f );
[0069] And perform input gate calculation, using the formula: i t =(W i ·[h t-1 x t ]+b i ), Where i t Controlling new features The deposit ratio;
[0070] By combining the outputs of the forget gate and the input gate, the state C of the memory cell is updated. t It retains historical information related to the current behavior, using the following formula:
[0071] And perform output gate operations, using the following formula: o t =σ(W O ·[h t-1 x t ]+b o ), h t =o t ⊙tanh(C t ), where o t Controlling the output ratio of information in the memory unit generates the current hidden state h. t This is used for subsequent behavior classification;
[0072] S503: Output Layer and Behavior Classification: The output layer will include the hidden state h from the last time step. t The input is a fully connected layer, mapped to the behavior category dimension;
[0073] S6: Real-time push of customized services: Based on the model output results, the device performs personalized interaction. When the user approaches the TV, the gateway wakes up the TV through the HDMI-CEC protocol and pushes cloud-recommended movie and TV content. The device receives instructions through MQTT and executes hardware interface control.
[0074] S7: Real-time monitoring of user feedback: The system monitors feedback through changes in device status or tag sensor data. If there is feedback, it jumps to collect feedback data; if there is no feedback, it jumps to continue monitoring behavior and returns to S2.
[0075] S8: Cloud-based model reinforcement learning and update: The cloud uses reinforcement learning algorithms, with user satisfaction as the reward function, to optimize the LSTM model parameters and push the new model to the edge gateway. The edge gateway performs real-time updates, where the cloud updates the LSTM neural network weights using gradient descent, and the edge gateway receives the model file via HTTP and overwrites the local version.
[0076] 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 inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A smart home non-inductive interaction system based on UWB positioning, characterized in that, The application relates to a UWB positioning system for smart home, which comprises a user terminal, a UWB positioning base station network, an edge computing gateway, a smart home device cluster, a cloud service platform and a user mobile phone APP, the user terminal is connected with the UWB positioning base station network through a UWB signal, the edge computing gateway is wirelessly connected with the UWB positioning base station network and the user mobile phone APP, and the edge computing gateway is connected with the smart home device cluster and the cloud service platform. 2.The smart home non-inductive interaction system based on UWB positioning of claim 1, wherein, The UWB positioning base station network is arranged in each room of a family, including a living room, a bedroom, a kitchen and a bathroom, 3-4 UWB positioning base stations are reasonably arranged in each room according to the area and shape of the room, and the UWB positioning base stations are distributed and arranged on the corners of a ceiling or high positions of walls to ensure that the signal coverage range is wide and there is no dead angle, each UWB positioning base station communicates through a wired network or a wireless network to form a three-dimensional coordinate positioning network for acquiring the position information of the user terminal in real time. The positioning principle of the UWB positioning base station network is based on time-of-flight ranging technology. The UWB tag periodically sends a pulse signal to the surrounding UWB positioning base station, and the UWB positioning base station records the time stamp after receiving the signal. According to the propagation speed of the signal in the air, the distance between the tag and the base station is calculated, and the formula used is: Wherein, Delta t is the time difference between the signal sending time of the tag and the receiving time of the base station, and C is the speed of light. After obtaining the distances between the tag and the at least three UWB positioning base stations, the coordinates (x, y, z) of the tag in the three-dimensional space are calculated by using a trilateration algorithm to achieve high-precision positioning at the centimeter level. The principle of the trilateration algorithm is that the coordinates of the three edges are (x i ,y i ,z i ), and the system will satisfy The user position is obtained by solving the equation set, and the accuracy is 5-10 centimeters. 3.The smart home non-inductive interaction system based on UWB positioning of claim 1, wherein, The user terminal is designed as a light-weight and portable UWB tag, has a small appearance and can be made into a bracelet, a brooch or a key ring, so that the user can conveniently wear or carry the tag, the tag is internally integrated with a UWB communication module, a nine-axis sensor, a low-power Bluetooth module and a micro lithium battery, the nine-axis sensor comprises a 3-axis accelerometer, a 3-axis gyroscope and a 3-axis magnetometer, and can collect the motion acceleration, rotation angle and magnetic field information of the user in real time. The working principle of the nine-axis sensor is that the collected acceleration, angular velocity and magnetic field data and the UWB positioning data are fused through a Kalman filtering algorithm. The Kalman filtering is an optimal estimation method based on a linear system state space model. It continuously corrects the estimation of the system state through two steps of prediction and update. The formula used in the prediction process is: The formula used in its update process is: P k|k = (I - K k H k )P k|k-1 ; where x^ k∣k-1 and x^ k∣k are the predicted and updated state estimates, respectively, P k∣k-1 and P k∣ are the corresponding covariance matrices, F k is the state transition matrix, B k is the control input matrix, u k is the control input, H k is the observation matrix, z k is the observation, Q k is the process noise covariance matrix, R k is the observation noise covariance matrix, and K k is the Kalman gain.
4. The intelligent home non-inductive interaction system based on UWB positioning according to claim 1, characterized in that, The edge computing gateway adopts a high-performance hardware platform, is equipped with a Rockchip RK3399 six-core processor, is provided with 2GB RAM and 16GB eMMC storage, runs an Ubuntu Core operating system, deploys various application programs based on a Docker container technology, and the core function modules of the edge computing gateway include a UWB positioning data processing engine, an AI behavior intention recognition algorithm module, a scene automatic rule engine and a communication interface module, the AI behavior intention recognition algorithm module adopts a long short-term memory network model based on deep learning to recognize the user behavior based on a behavior intention recognition algorithm. The smart home device cluster comprises smart lights, a smart air conditioner, a smart television and a smart curtain, the devices are connected with the edge computing gateway based on a MQTT protocol through Wi-Fi, Bluetooth and ZigBee wireless communication technologies, and receive and execute the control instructions sent by the gateway. 5.The smart home non-inductive interaction system based on UWB positioning of claim 1, wherein, The cloud service platform is constructed on a public cloud or a private cloud, adopts a distributed storage and computing architecture, and mainly has the functions of storing historical behavior data of the user, training and optimizing an AI model, providing a remote management interface and protecting data security, the historical behavior data of the user includes the staying time of the user in different rooms and the device use habit, the AI model is trained and optimized by using big data and a machine learning algorithm to continuously improve the accuracy of behavior intention recognition, the user can remotely check and manage the home devices, set personalized scenes and protect data security through the remote management interface and the mobile phone APP, and the data security protection adopts an encryption transmission and storage technology to ensure the safety of the user data.
6. A smart home non-inductive interaction method based on UWB positioning, characterized in that, The application further discloses a UWB positioning system for smart home, which comprises the following steps: S1: System startup and initialization: after the system power is turned on, the edge computing gateway starts the Ubuntu Core system through the Rockchip RK3399 processor, loads the UWB positioning engine and AI algorithm module, and initializes the signal receiving channel of the UWB positioning base station. The smart home device accesses the edge gateway through Wi-Fi / ZigBee and establishes an MQTT communication link; S2: User position sensing: the user wears a UWB tag that sends pulse signals to the surrounding base station at a frequency of 100-500 times per second, and a built-in nine-axis sensor synchronously collects motion data; S3: User movement state judgment: the edge gateway receives the base station timestamp data, calculates the three-dimensional coordinates through the trilateration algorithm, and judges whether the user is moving by fusing the nine-axis sensor data. If the user is moving, the position coordinates will be updated; if the user is not moving, the current state will be maintained, and the second step will be returned immediately and continue to be monitored; S4: Position coordinate update and scene matching: when the user moves, the edge gateway updates the real-time coordinates and compares them with the preset scene area. The scene area calculates whether the coordinates fall into the area through the trilateration algorithm. If the coordinates match the preset scene, the edge gateway generates control instructions through the scene rule engine and sends them to the corresponding device for execution. If the coordinates do not match any scene, the device maintains the current state, and the position change is monitored immediately by returning to S2; S5: Behavior intention analysis and model calling: the edge gateway analyzes the user's real-time behavior through the LSTM neural network, combines the time, and calls the personalized interaction model generated by the cloud; S6: Customized service real-time push: according to the model output result, the device executes personalized interaction. When the user approaches the TV, the gateway wakes up the TV through the HDMI-CEC protocol and pushes the cloud-recommended video content. The device receives the instructions through MQTT and executes hardware interface control; S7: Real-time monitoring of user feedback: the system monitors feedback through device state changes or tag sensor data. If there is feedback, it jumps to collect feedback data; if there is no feedback, it continues to monitor behavior and returns to S2; S8: Cloud model reinforcement learning and update: the cloud uses reinforcement learning algorithm to optimize the LSTM model parameters with user satisfaction as the reward function, and pushes the new model to the edge gateway for real-time update.
7. The intelligent home non-inductive interaction method based on UWB positioning of claim 6, characterized in that, In S5, the LSTM neural network includes memory cells and three gate mechanisms. The LSTM neural network realizes selective retention and forgetting of historical information through memory cells and gate mechanisms, which is suitable for processing time series data such as user behavior. The memory cells are used to store historical state information as "memory" carriers, allowing information to be transmitted over a long time sequence without loss, and are used to store the user's past position, speed, and acceleration characteristics to judge the continuity of the current behavior. The gating mechanism includes an input gate, a forget gate and an output gate, the input gate decides whether the new feature data at the current moment is stored in the memory unit, the forget gate is used for controlling the discard proportion of old information in the memory unit, and the output gate screens out the features used for behavior classification according to the memory unit state and the current input; The logical steps of the LSTM neural network operation are as follows: S501: input layer data processing: input features: fused user data, timestamp, room ID, and organized as a multi-dimensional vector in time sequence; S502: Memory cell and gate operation: Calculate the forget gate value f by the sigmoid function (σ) t (0-1 between), determine the proportion of forgetting of the history state h in the memory cell t-1 and the current input x t , the formula used is: f t = (W f · [h t-1 , x t ]+b f ); and input gate computation is performed using the formula: t = (W i · [h t-1 , x t ]+b i ), where i t controls the proportion of new features to store. Update the memory cell state C in conjunction with the output of the forget gate and the input gate t , preserving history information related to the current behavior, which uses the formula: And the output gate operation is carried out, and the formula used is: t = σ(W O · [h t-1 , x t ]+b o ), h t = o t ⊙ tanh(C t ), wherein o t controls the output proportion of information in the memory unit, and generates the current hidden state h t for subsequent behavior classification; S503: Output layer with behavior classification: The output layer will output the hidden state h t An input fully connected layer, mapping to the behavior class dimension.
8. The intelligent home non-inductive interaction method based on UWB positioning of claim 6, characterized in that, In the S8, the cloud updates the LSTM neural network weight through the gradient descent method, and the edge gateway receives the model file through the HTTP protocol and covers the local version.