Active home device control scheduling method and system based on spatio-temporal behavior prediction

Abstract

The application provides a method and system for active home equipment control scheduling based on space-time behavior prediction, and relates to the technical field of smart home. The method comprises the following steps: acquiring multi-source space-time data of a home environment, wherein the multi-source space-time data comprises user position trajectory data, time series behavior data, home equipment state data, environment parameter data and historical behavior logs; and constructing at least one distributed space-time behavior data mart and receiving the multi-source space-time data. According to the method, the multi-source space-time data is acquired through Sp1, the distributed space-time behavior data mart is constructed through Sp2, the multi-source space-time data is converted into a weighted space-time behavior metadata chain containing trajectory metadata, intention metadata and demand metadata, and the weighted association relationship between various metadata is established, thereby breaking through the limitation of independent data processing in the prior art.

Description

Technical Field

[0001] This invention relates to the field of smart home technology, specifically to an active home device control and scheduling method and system based on spatiotemporal behavior prediction. Background Technology

[0002] With the rapid development of the smart home industry, users' control needs for home devices have gradually shifted from traditional manual control and passive response to proactive scheduling based on user behavior prediction. The core demand is to improve the comfort and convenience of home life while optimizing energy efficiency. In smart home scenarios, the effectiveness of home device control and scheduling directly affects the user experience and the level of intelligence of the home system. Therefore, how to achieve accurate behavior prediction and proactive scheduling has become a key research focus in the current smart home field.

[0003] Currently, common technologies for controlling and scheduling home appliances mainly include manual control, timed control, simple sensor-triggered control, and basic historical habit analysis control. Manual control is the most traditional method, where users directly operate the devices to turn them on or off or adjust parameters. Timed control triggers device start / stop or parameter adjustments through preset time nodes. Sensor-triggered control relies on simple sensors such as human body sensors and door / window sensors to detect changes in the environment or user status and trigger device actions. Basic historical habit analysis control achieves simple automated triggering by mining data such as the frequency and time periods of past device use. These technologies have been applied to some extent in home settings, reducing the user's operational burden to a certain degree.

[0004] However, existing home appliance control and scheduling technologies still have the following main problems: 1. Data processing methods are fragmented, lacking the ability to unify, integrate, and correlate multi-source data. Existing technologies mostly collect and process multi-source data such as user location trajectory, device status, and environmental parameters independently, without establishing a correlation mechanism between various types of data. This makes it impossible to achieve collaborative use of data, difficult to fully characterize the spatiotemporal behavioral features of users, and unable to provide strong support for accurate prediction.

[0005] 2. Lack of effective spatiotemporal behavior prediction capabilities, resulting in a passive scheduling mode. Existing technologies are mostly "post-event response" or "simple preset" modes, without building targeted spatiotemporal behavior prediction models. They cannot accurately predict users' future location movements, behavioral intentions, and equipment needs based on historical data and current status, causing equipment scheduling to always lag behind user behavior and failing to achieve truly proactive service.

[0006] Therefore, a proactive home appliance control and scheduling method and system based on spatiotemporal behavior prediction is needed to solve the above problems. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an active home appliance control and scheduling method and system based on spatiotemporal behavior prediction, which solves the problems mentioned in the background section above.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a proactive home appliance control and scheduling method based on spatiotemporal behavior prediction, comprising the following steps: Sp1. Acquire multi-source spatiotemporal data of the home environment, including user location trajectory data, time-series behavior data, home device status data, environmental parameter data, and historical behavior logs; Sp2. Construct at least one distributed spatiotemporal behavior data mart and receive the multi-source spatiotemporal data. In the distributed spatiotemporal behavior data mart, transform the multi-source spatiotemporal data into a weighted spatiotemporal behavior metadata chain. The weighted spatiotemporal behavior metadata chain includes trajectory metadata, intent metadata, and demand metadata. Sp3. Input the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation, and output the prediction results of the user's future spatiotemporal behavior through forward propagation; Sp4. Generate an active home device control command sequence based on the predicted results of the user's future spatiotemporal behavior, and send it to the corresponding home devices for execution; Sp5. Obtain the execution feedback data of the home devices and feed it back to the distributed spatiotemporal behavior data mart.

[0009] Preferably, the method for obtaining multi-source spatiotemporal data of the home environment in Sp1 is as follows: Sp1.1 uses IoT sensors and Wi-Fi / Bluetooth positioning modules deployed in each room to collect user location trajectory data and time-series behavior data in real time, and performs timestamp synchronization and spatial coordinate mapping processing. SP1.2: Obtain real-time status data of home devices through the smart home gateway interface; SP1.3, synchronously collect environmental parameter data; SP1.4 Extract user's past spatiotemporal behavior records from the historical behavior log database; Sp1.5 uses a multimodal data preprocessing module to perform denoising, normalization, and spatiotemporal sequence alignment on the acquired multi-source spatiotemporal data.

[0010] Preferably, the method for constructing a distributed spatiotemporal behavior data mart and transforming a weighted spatiotemporal behavior metadata chain in Sp2 is as follows: Sp2.1, Construct a distributed spatiotemporal behavior data mart; Sp2.2, convert user location trajectory data into trajectory metadata; Sp2.3, transform time-series behavioral data into intent metadata; Sp2.4. Transform home appliance status data and environmental parameter data into demand metadata; Sp2.5: Establish weighted associations between trajectory metadata, intent metadata, and demand metadata to form a weighted spatiotemporal behavior metadata chain.

[0011] Preferably, the method for inputting the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation in Sp3 is as follows: Sp3.1 Input the weighted spatiotemporal behavior metadata chain into the trajectory layer neurons, intention layer neurons, and demand layer neurons of the spatiotemporal behavior prediction model; Sp3.2, perform spatiotemporal feature extraction, fusion calculation and activation in neurons of each layer; Sp3.3 outputs predictions of users' future spatiotemporal behavior through forward propagation, including user location movement trajectory, behavioral intent chain, and device demand probability.

[0012] Preferably, the method for generating and issuing a sequence of active home device control commands based on the prediction results of the user's future spatiotemporal behavior in Sp4 is as follows: SP4.1: Perform chain matching of user location movement trajectory, behavioral intent chain and device demand probability to determine target home devices and optimal control timing; SP4.2 uses a multi-objective optimization algorithm to generate a sequence of proactive home appliance control instructions that includes device start / stop timing, operating parameter adjustment values, and advance execution amounts; Sp4.3. The active home device control command sequence is sent to the corresponding home devices for execution via the Matter protocol or Zigbee protocol to achieve predictive-driven active control scheduling.

[0013] Preferably, the method for obtaining and feeding back the home appliance execution feedback data in Sp5 is as follows: SP5.1: Collect actual operating status data from the corresponding home appliances after execution; SP5.2, Use the actual operating status data as the feedback data for home appliances; Sp5.3. Feedback data from the home devices is fed back to the distributed spatiotemporal behavior data mart to update the weighted spatiotemporal behavior metadata chain.

[0014] Preferably, the method includes a two-layer metadata-driven mechanism, which includes distributed spatiotemporal behavior data mart metadata-driven mechanism and spatiotemporal behavior prediction model metadata-driven mechanism. The two-layer metadata-driven mechanism is used to drive spatiotemporal behavior prediction and proactive home device control and scheduling.

[0015] Preferably, the method is divided into the following three stages based on the prediction time length of the user's future spatiotemporal behavior prediction results: During the short-term trajectory prediction period, metadata of the user's location movement sequence is generated in the first distributed spatiotemporal behavior data mart and trajectory layer neurons. During the mid-term intent prediction period, user behavior intent sequence metadata is generated in the second distributed spatiotemporal behavior data mart and intent layer neurons. During the long-term demand forecasting period, metadata of home appliance demand sequences is generated using the third distributed spatiotemporal behavioral data mart and demand layer neurons.

[0016] Preferably, the system includes: The multi-source spatiotemporal data acquisition module is used to execute the Sp1 step to acquire multi-source spatiotemporal data of the home environment; The distributed spatiotemporal behavior data mart module is connected to the multi-source spatiotemporal data acquisition module and is used to perform the Sp2 step, receive the multi-source spatiotemporal data and convert it into a weighted spatiotemporal behavior metadata chain; The spatiotemporal behavior prediction model module is connected to the distributed spatiotemporal behavior data mart module and is used to execute the Sp3 step, receive the weighted spatiotemporal behavior metadata chain for calculation and activation, and output the prediction result of the user's future spatiotemporal behavior. An active home device control command sequence generation module is connected to the spatiotemporal behavior prediction model module and is used to execute the Sp4 step, which generates an active home device control command sequence based on the user's future spatiotemporal behavior prediction results and sends it to the corresponding home devices for execution. The execution feedback data acquisition module is used to execute the Sp5 step, acquire the execution feedback data of the home device, and feed it back to the distributed spatiotemporal behavior data mart module.

[0017] Preferably, the system further includes a model optimization module, which is connected to the spatiotemporal behavior prediction model module and the execution feedback data acquisition module, and is used to perform backpropagation training and weight adjustment on the spatiotemporal behavior prediction model.

[0018] Beneficial effects This invention provides a proactive home appliance control and scheduling method and system based on spatiotemporal behavior prediction. It offers the following advantages: This invention achieves unified integration and correlation analysis of multi-source spatiotemporal data, solving the problems of fragmented data processing, lack of collaborative utilization of multi-source data, and difficulty in fully characterizing user spatiotemporal behavioral features in existing technologies. This method acquires multi-source spatiotemporal data through step Sp1, and constructs a distributed spatiotemporal behavioral data mart through step Sp2. It transforms multi-source spatiotemporal data into a weighted spatiotemporal behavioral metadata chain containing trajectory metadata, intent metadata, and demand metadata, and establishes weighted correlations between various types of metadata. This overcomes the limitations of independent data processing in existing technologies, achieving systematic integration and efficient utilization of multi-source data, and fully characterizing user spatiotemporal behavioral features, providing strong support for subsequent accurate spatiotemporal behavior prediction.

[0019] This invention achieves accurate prediction of user spatiotemporal behavior, solving the problems of existing technologies lacking effective spatiotemporal behavior prediction capabilities, passive scheduling modes, and inability to provide proactive services. This method inputs a weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model through the Sp3 step. After spatiotemporal feature extraction, fusion calculation, and activation of neurons at each layer, it outputs prediction results containing user location movement trajectories, behavioral intent chains, and device demand probabilities through forward propagation. Combined with a multi-stage prediction mechanism, it overcomes the limitations of existing technologies' "post-event response" or "simple preset" approaches, achieving accurate prediction of users' future spatiotemporal behavior. This drives home device scheduling from passive response to proactive service, improving the convenience and intelligence level of home use. Attached Figure Description

[0020] Figure 1 This is the main flowchart of the method of the present invention; Figure 2 This is a flowchart of the Sp1 step of the present invention; Figure 3 This is a flowchart of the Sp2 steps of the present invention; Figure 4 This is a flowchart of the Sp3 steps of the present invention; Figure 5 This is a flowchart of the Sp4 steps of the present invention; Figure 6 This is a flowchart of the Sp5 steps of the present invention; Figure 7 This is a system framework diagram of the present invention; Figure 8 This is a diagram showing the main interface of the system according to the present invention; Figure 9 This is a diagram showing the system data acquisition and monitoring interface of the present invention. Detailed Implementation

[0021] 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. Specific Implementation Example 1: Please see Figure 1 As shown, a proactive home appliance control and scheduling method based on spatiotemporal behavior prediction is provided, including the following steps: Sp1. Acquire multi-source spatiotemporal data of the home environment, including user location trajectory data, time series behavior data, home device status data, environmental parameter data, and historical behavior logs; Sp2. Construct at least one distributed spatiotemporal behavior data mart and receive multi-source spatiotemporal data. In the distributed spatiotemporal behavior data mart, transform the multi-source spatiotemporal data into a weighted spatiotemporal behavior metadata chain. The weighted spatiotemporal behavior metadata chain includes trajectory metadata, intent metadata, and demand metadata. Sp3. Input the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation, and output the prediction results of the user's future spatiotemporal behavior through forward propagation; Sp4. Generate a sequence of proactive home appliance control commands based on the prediction results of the user's future spatiotemporal behavior, and send them to the corresponding home appliances for execution; Sp5: Obtain feedback data from home appliances and feed it back to a distributed spatiotemporal behavior data mart.

[0023] Please see Figure 2 As shown, a method for obtaining multi-source spatiotemporal data of the home environment in Sp1 is provided. The specific steps are as follows: Sp1.1 uses IoT sensors and Wi-Fi / Bluetooth positioning modules deployed in each room to collect user location trajectory data and time-series behavior data in real time, and performs timestamp synchronization and spatial coordinate mapping processing. SP1.2: Obtain real-time status data of home devices through the smart home gateway interface; SP1.3, synchronously collect environmental parameter data; SP1.4 Extract user's past spatiotemporal behavior records from the historical behavior log database; Sp1.5 uses a multimodal data preprocessing module to perform denoising, normalization, and spatiotemporal sequence alignment on the acquired multi-source spatiotemporal data.

[0024] Please see Figure 3As shown, the method for constructing a distributed spatiotemporal behavior data mart and transforming it into a weighted spatiotemporal behavior metadata chain in Sp2 is as follows: Sp2.1, Construct a distributed spatiotemporal behavior data mart; Sp2.2, convert user location trajectory data into trajectory metadata; Sp2.3, transform time-series behavioral data into intent metadata; Sp2.4. Transform home appliance status data and environmental parameter data into demand metadata; Sp2.5: Establish weighted associations between trajectory metadata, intent metadata, and demand metadata to form a weighted spatiotemporal behavior metadata chain.

[0025] Please see Figure 4 As shown, the method for inputting the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation in Sp3 is as follows: Sp3.1 Input the weighted spatiotemporal behavior metadata chain into the trajectory layer neurons, intention layer neurons, and demand layer neurons of the spatiotemporal behavior prediction model; Sp3.2, perform spatiotemporal feature extraction, fusion calculation and activation in neurons of each layer; Sp3.3 outputs predictions of users' future spatiotemporal behavior through forward propagation, including user location movement trajectory, behavioral intent chain, and device demand probability.

[0026] Please see Figure 5 As shown, the method for generating and executing a sequence of proactive home appliance control commands based on predictions of the user's future spatiotemporal behavior in SP4 is as follows: SP4.1: Perform chain matching of user location movement trajectory, behavioral intent chain and device demand probability to determine target home devices and optimal control timing; SP4.2 uses a multi-objective optimization algorithm to generate a sequence of proactive home appliance control instructions that includes device start / stop timing, operating parameter adjustment values, and advance execution amounts; SP4.3 uses the Matter or Zigbee protocol to send a sequence of active home device control commands to the corresponding home devices for execution, thereby achieving predictive-driven active control scheduling.

[0027] Please see Figure 6 As shown, the method for obtaining and feeding back home device execution feedback data in SP5 is as follows: SP5.1: Collect actual operating status data from the corresponding home appliances after execution; SP5.2, Use actual operating status data as feedback data for home appliances; SP5.3 feeds back the execution feedback data of home devices to the distributed spatiotemporal behavior data mart to update the weighted spatiotemporal behavior metadata chain.

[0028] The method includes a two-layer metadata-driven mechanism, which consists of distributed spatiotemporal behavior data mart metadata-driven mechanism and spatiotemporal behavior prediction model metadata-driven mechanism. The two-layer metadata-driven mechanism is used to drive spatiotemporal behavior prediction and proactive home device control and scheduling.

[0029] The method is divided into the following three stages based on the prediction time length of the user's future spatiotemporal behavior prediction results: During the short-term trajectory prediction period, metadata of the user's location movement sequence is generated in the first distributed spatiotemporal behavior data mart and trajectory layer neurons. During the mid-term intent prediction period, user behavior intent sequence metadata is generated in the second distributed spatiotemporal behavior data mart and intent layer neurons. During the long-term demand forecasting period, metadata of home appliance demand sequences is generated using the third distributed spatiotemporal behavioral data mart and demand layer neurons.

[0030] Please see Figure 7 As shown, the system includes: The multi-source spatiotemporal data acquisition module is used to execute the Sp1 step to acquire multi-source spatiotemporal data of the home environment; The distributed spatiotemporal behavior data mart module is connected to the multi-source spatiotemporal data acquisition module. It is used to perform the Sp2 step, receive multi-source spatiotemporal data and transform it into a weighted spatiotemporal behavior metadata chain. The spatiotemporal behavior prediction model module, connected to the distributed spatiotemporal behavior data mart module, is used to execute the Sp3 step, receive the weighted spatiotemporal behavior metadata chain for calculation and activation, and output the prediction results of the user's future spatiotemporal behavior. The active home appliance control command sequence generation module is connected to the spatiotemporal behavior prediction model module. It is used to execute the SP4 step, generate an active home appliance control command sequence based on the prediction results of the user's future spatiotemporal behavior, and send it to the corresponding home appliances for execution. The execution feedback data acquisition module is used to execute the SP5 steps, acquire the execution feedback data of the home devices, and feed it back to the distributed spatiotemporal behavior data mart module.

[0031] The system also includes a model optimization module, which is connected to the spatiotemporal behavior prediction model module and the execution feedback data acquisition module, and is used to perform backpropagation training and weight adjustment on the spatiotemporal behavior prediction model. Specific Implementation Example 2: Please see Figure 1 As shown, a main flowchart of the method is provided. The following is a detailed description of the method steps in Embodiment 1: This method can be applied to different smart home environments, including functional areas such as living room, master bedroom, secondary bedroom, kitchen or bathroom. First, it is necessary to deploy facilities such as IoT sensors, positioning modules, smart home gateways, and smart terminal devices. Then, this method can be used to accurately predict the daily spatial and temporal behavior of the elderly, actively dispatch various home devices, and improve the level of intelligence of elderly care services.

[0033] After deployment, perform the following steps: Sp1. Acquire multi-source spatiotemporal data of the home environment: This step utilizes multi-source acquisition facilities to comprehensively acquire multi-source spatiotemporal data from the home environment, as detailed below: Sp1.1, Collect user location trajectory data and time-series behavioral data: UWB positioning sensors, corresponding to IoT sensors, are deployed at key locations in each functional area. Bluetooth and Wi-Fi modules are integrated into the user's smart bracelet. The UWB positioning sensors collect the user's location coordinates (x, y, z) in real time at a sampling frequency of 1Hz, generating one location record every second. The Bluetooth and Wi-Fi modules assist in locating the user's position information through signal strength and access information. All positioning and data acquisition devices are time-stamped with the home edge computing gateway via the NTP protocol to ensure consistent time stamps for all data. A coordinate mapping relationship is pre-built within the home space, associating the physical boundaries of each functional area with area identifiers. The collected location coordinates can be directly mapped to the corresponding functional area and its relative position within that area. The collected raw location and behavioral data are transmitted to the edge computing gateway via a wireless communication link for initial data aggregation.

[0034] SP1.2, Obtain Home Appliance Status Data: All home devices connected to the system communicate with the system through the smart home gateway interface. The gateway reads the operating status of each device in real time, including the device's start / stop status, operating mode, set parameters, current operating parameters, and operating power. The status data of each device is transmitted to the smart home gateway through the internal communication module. The gateway verifies the validity of the data, removes abnormal data that exceeds the reasonable range, and then transmits it to the edge computing gateway for aggregation.

[0035] SP1.3 Synchronously collect environmental parameter data: Environmental sensors, including temperature and humidity sensors, light intensity sensors, and air quality sensors, are deployed in each functional area, with a sampling frequency of 2Hz. The sensors collect environmental parameters in real time, such as temperature, relative humidity, light intensity, TVOC concentration, and equivalent carbon dioxide concentration. The environmental sensors transmit the collected parameter data to the edge computing gateway via an internal communication protocol. The gateway adds a unified timestamp to the data and completes data aggregation.

[0036] SP1.4 Extracting historical behavior logs: The edge computing gateway's local storage unit stores user home behavior logs from the past six months. Using the current time as a baseline, historical spatiotemporal behavior records for the same time period within the past seven days are extracted, including user location movement trajectories, behavior types, and device operation records. The extracted historical logs undergo deduplication and missing value imputation processes, removing duplicate time records and filling missing behavior data with interpolated records from adjacent time periods to ensure the integrity of the historical data.

[0037] Sp1.5, Multimodal Data Preprocessing: The edge computing gateway performs unified preprocessing on all aggregated multi-source data. The specific steps are as follows: Denoising: Wavelet denoising algorithm is used to remove noise points from the location trajectory data. When the collected location coordinates suddenly change beyond the user's normal movement speed range, the abnormal coordinate values ​​are corrected through algorithmic smoothing. Outliers in environmental parameter data are handled using a 3D algorithm. The criteria are used to identify and eliminate outliers, which are then replaced with the average value of the sensor over the same period.

[0038] Normalization: Multi-source data with different dimensions and value ranges are normalized to eliminate dimensional differences. The minimum-maximum normalization algorithm is used, and the formula is as follows: ; in, The original data value, such as the original temperature data collected for a certain area; the historical minimum value for this data dimension, such as the lowest temperature ever collected for this area. This represents the historical maximum value for this data dimension, such as the highest temperature ever recorded in this region. These are the normalized data values, ranging from 0 to 1. Normalization is performed on data from different dimensions, such as location coordinates, environmental parameters, and equipment operating power, to ensure that the value range of all data is consistent.

[0039] Spatiotemporal sequence alignment processing: Based on timestamps, location trajectory data, device status data, and environmental parameter data near the same time point are aligned to ensure that multi-source data from the same time form associated data blocks. For historical behavior logs, they are sorted chronologically to form a continuous historical behavior sequence, providing input data in a unified format for subsequent processing.

[0040] Sp2: Construct a distributed spatiotemporal behavior data mart and transform it into a weighted spatiotemporal behavior metadata chain: This step, based on the preprocessed multi-source data, constructs a distributed data mart and transforms the data into a weighted spatiotemporal behavioral metadata chain, as detailed below: SP2.1: Constructing a Distributed Spatiotemporal Behavioral Data Mart: A three-tiered distributed architecture for spatiotemporal behavioral data marts is constructed, comprising nodes at three levels: edge layer, fog layer, and cloud layer. Edge layer nodes are deployed in the local edge computing gateway of the home, responsible for receiving and storing local multi-source spatiotemporal data in real time, supporting high-frequency read and write operations, and meeting the needs of local real-time data processing.

[0041] The fog layer node is deployed on the server of the community smart elderly care service center. It is responsible for receiving aggregated data transmitted from multiple home edge layer nodes in the jurisdiction and supports regional spatiotemporal data query and analysis.

[0042] The cloud node is deployed on a cloud server and is responsible for receiving the full aggregated data transmitted by the fog node, storing massive amounts of historical behavior data, and supporting global distributed data analysis.

[0043] Edge layer nodes write preprocessed spatiotemporal data to local storage in real time, and push aggregated local data to fog layer nodes every hour; fog layer nodes push the full regional data of the day to cloud layer nodes once a day, completing the hierarchical storage and synchronization of data.

[0044] SP2.2: Convert user location trajectory data into trajectory metadata: Location trajectory data is extracted from the storage units of edge layer nodes, and core trajectory features are extracted, including the sequence of functional areas traversed by the user, the duration of stay in each functional area, and the user's movement speed. At the same time, the user's location trajectory over the past 7 days is clustered using a clustering algorithm to identify the user's core trajectory pattern.

[0045] Based on the extracted trajectory features, trajectory metadata is generated to characterize the user's location movement patterns and provide basic data for subsequent trajectory prediction.

[0046] SP2.3: Transforming time-series behavioral data into intent metadata: Time-series behavioral data is extracted, and a bidirectional long short-term memory network is used to identify user behavioral intent. A behavioral intent feature vector is constructed, with features including current location, dwell time, changes in adjacent locations, ambient temperature and humidity, light intensity, and historical behavioral sequence features.

[0047] A bidirectional long short-term memory (LSTM) network model is trained based on the user's historical behavior logs. The model consists of two LSM layers and one fully connected layer, with the output dimension corresponding to the user's core behavioral intent type. The behavioral feature vector within the current 5-minute time window is input into the trained model, which outputs a probability distribution of the user's behavioral intent. The intent corresponding to the highest probability is taken as the core behavioral intent.

[0048] Based on the identified behavioral intent and related features, intent metadata is generated to characterize the patterns of user behavioral intent and provide a basis for intent prediction.

[0049] SP2.4: Transform home appliance status data and environmental parameter data into demand metadata: Extract home appliance status data and environmental parameter data, and combine them with the user's historical device usage records to extract the user's device demand characteristics, including the current operating status of the device, environmental parameters, and the user's historical device parameter setting habits in the corresponding scenarios.

[0050] Based on the extracted demand features, demand metadata is generated to characterize the patterns of user demand for home appliances, providing a basis for predicting appliance demand.

[0051] SP2.5: Establish weighted associations to form a weighted spatiotemporal behavioral metadata chain: Establish a weighted association relationship between trajectory metadata, intent metadata, and demand metadata. The association weight is calculated based on the co-occurrence frequency of historical data, as shown in the following formula: ; in, For metadata With metadata The correlation weight between them; For metadata With metadata The number of times they appear together in historical data; For metadata Total number of times it appears in historical data.

[0052] This weight represents the degree of correlation between different metadata. The three types of metadata are linked together with the correlation weight in chronological order to form a weighted spatiotemporal behavioral metadata chain, which provides input for subsequent prediction models.

[0053] Sp3: Input the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation: This step inputs the metadata chain into the prediction model to complete feature extraction, fusion, and prediction, as detailed below: Sp3.1: Input hierarchical neurons: The weighted spatiotemporal behavior metadata chain is input into the trajectory layer neurons, intent layer neurons, and demand layer neurons of the spatiotemporal behavior prediction model, respectively. Trajectory metadata is input into the trajectory layer, intent metadata into the intent layer, and demand metadata into the demand layer, thereby achieving hierarchical feature processing.

[0054] SP3.2: Hierarchical Feature Extraction and Fusion Activation: Each layer of neurons extracts features from the input metadata: the trajectory layer neurons extract the spatiotemporal features of the trajectory to capture the temporal patterns of the user's position movement; the intent layer neurons extract the temporal features of the behavior to capture the changing patterns of the user's behavioral intent; and the demand layer neurons extract the features of the device demand to capture the changing patterns of the user's demand for the device.

[0055] After completing the hierarchical feature extraction, an attention-weighted fusion algorithm is used to fuse the features from each layer, as shown in the following formula: ; in, This is the total feature vector after fusion; For the first The attention weights of the layer features are calculated through the attention mechanism and are used to characterize the degree of contribution of each layer feature to the prediction result. For the first Feature vectors extracted by layer neurons These correspond to the trajectory layer, intent layer, and requirement layer, respectively.

[0056] After fusion, the ReLU activation function is used to activate the fused features, introducing nonlinear transformation to improve the model's expressive power.

[0057] SP3.3: Outputs future spatiotemporal behavior predictions for users through forward propagation, including user location movement trajectories, behavioral intent chains, and device demand probabilities. The forward propagation output includes predictions of the user's future spatiotemporal behavior, which include the user's location movement trajectory, behavioral intent chain, and device demand probability.

[0058] The core calculation of the spatiotemporal behavior prediction model uses the forward propagation formula: ; in This represents the input vector, i.e., the weighted spatiotemporal behavior metadata chain. Represents the trajectory layer weight matrix. This represents the intention layer weight matrix. This represents the weight matrix of the demand layer. This represents the trajectory layer bias vector. This represents the intention layer bias vector. This represents the demand layer bias vector. This represents the activation function. The output vector represents the user's location movement trajectory, behavioral intent chain, and device demand probability. The specific calculation logic steps are as follows: First, the weighted spatiotemporal behavioral metadata chain is used as the input vector. The data is fed into the trajectory layer neurons, and the trajectory layer output is calculated. The first step is to complete the trajectory feature extraction; the second step is to... The input is fed into the intent layer neurons, and the intent layer output is calculated as follows: The third step is to complete the fusion of intent features; The vector is fed into the demand layer neurons as input to calculate the final output vector. This allows us to obtain the user's location movement trajectory, behavioral intent chain, and device demand probability.

[0059] Sp4. Generate a sequence of active home appliance control commands based on the prediction results and issue them for execution: This step strictly corresponds to all sub-steps of Sp4. Based on the prediction results, it clarifies the generation logic of control commands, the specific command content, and the execution operation of the device, as follows: SP4.1, Chain matching to determine targets and timing: The predicted user location movement trajectory, behavioral intent chain, and device demand probability are chain-matched. Based on the prediction results, the target home devices are matched as a smart air conditioner in the living room, a main light in the living room, a dining table warming device, and a small water heater in the kitchen. At the same time, the optimal control timing is determined by considering the start-up and preheating time of each device: the user is expected to arrive in the living room 300 seconds after the current time, the air conditioner needs 120 seconds to preheat, the light does not need to preheat but only needs to be turned on 30 seconds in advance to avoid darkness, the warming device needs to be started 300 seconds in advance to keep the food warm, and the small water heater needs 100 seconds to heat.

[0060] SP4.2, Multi-objective optimization generation control instructions: A multi-objective optimization algorithm is used to generate a sequence of active control commands. The optimization objectives are to maximize user comfort, minimize device energy consumption, and minimize response latency. The objective function is as follows: ; in The target vector to be optimized contains three optimization objectives; User comfort is rated, with a value ranging from 0 to 1. A higher value indicates a better user experience. The total energy consumption of the equipment is expressed in kilowatt-hours. This represents the device's response latency, measured in seconds.

[0061] At the same time, constraints are set, including range constraints on equipment operating parameters, such as air conditioning temperature needing to be between 16℃ and 30℃, light brightness needing to be between 0 and 1000 lux, and constraints on the start and stop status of the equipment.

[0062] The multi-objective optimization problem is solved using a non-dominated sorting genetic algorithm to obtain the optimal control parameters. In this scenario, the final generated sequence of active home appliance control commands is as follows, where all parameters strictly correspond to the device start / stop timing, operating parameter adjustment values, and advance execution amounts: The living room smart air conditioner control command is as follows: the start / stop time is 180 seconds after the current time, the operating parameter adjustment value is heating mode, the set temperature is 26℃, and the advance execution amount is 120 seconds. That is, the air conditioner will start 120 seconds before the user expects to arrive at the living room, allowing time for the air conditioner to preheat and warm up.

[0063] Living room main light control command: Start and stop timing is 270 seconds after the current time, the operating parameter adjustment values ​​are brightness 300 lux and color temperature 4000K, and the advance execution amount is 30 seconds, that is, the light will be turned on 30 seconds before the user's arrival time.

[0064] Control commands for the table warming equipment: Start and stop time is the current time, the operating parameter adjustment value is the warming temperature of 45℃, and the advance execution amount is 300 seconds, that is, the warming equipment will start immediately to complete the preparation of food warming in advance.

[0065] Kitchen water heater control commands: Start and stop timing is 200 seconds after the current time, the operating parameter adjustment value is water outlet temperature 42℃, and the advance execution amount is 100 seconds, that is, start the hot water equipment in advance to prepare warm water for subsequent dishwashing operations.

[0066] The above command parameters are all optimized results: Regarding user comfort, the parameters most preferred by the user in historical dining scenarios were selected, with a comfort score of 0.95; Regarding energy consumption, the advance execution amount was optimized, and to avoid energy waste caused by premature device startup, the total energy consumption corresponding to this command was only 0.12 kWh, which is a 15% reduction compared to scenarios without advance control; Regarding response latency, it was ensured that the execution timing of all devices perfectly matches the user's behavior time, with no delays or premature execution.

[0067] SP4.3, Command Issuance and Device Execution: The generated control command sequence is converted into a device-recognizable format using the Matter protocol and sent to the corresponding home devices. After receiving the command, each device first parses the command and verifies its legality. Then, it caches the command in its local task queue and waits to execute it according to the start and stop times specified in the command.

[0068] Once the execution time specified in the instruction is reached, each device will automatically perform the corresponding operation: The table warming equipment starts immediately at the current time, heating the internal temperature to 45℃ and maintaining it to ensure that the food is always at a suitable temperature.

[0069] The mini water heater automatically starts 200 seconds after the current time, heating the water to 42℃ and maintaining it to prepare warm water for the user's subsequent dishwashing operations.

[0070] The smart air conditioner in the living room will automatically start 180 seconds after the current time, switch to heating mode, adjust the set temperature to 26℃, and start heating to ensure that the room temperature has reached a comfortable 26℃ when the user arrives at the living room.

[0071] The living room main light will automatically turn on 270 seconds after the current time, adjusting the brightness to 300 lux and the color temperature to 4000K, just before the user enters the living room, thus preventing the user from entering a dark environment.

[0072] After all devices have completed their execution, they update their operating status in real time and await subsequent feedback collection, achieving predictive-driven proactive control and scheduling. This allows for pre-emptive device adjustments without manual user intervention. SP5: Acquires and feeds back execution feedback data from home appliances. This step strictly corresponds to all sub-steps in SP5, collecting execution feedback from the device to complete closed-loop optimization of the system, as detailed below: SP5.1: Collect actual operating status: Data on the actual operating status of each home appliance after the control command is executed is collected, including the actual start / stop status, actual operating parameters, and actual operating power of the appliances. In this scenario, the actual status collected is as follows: the air conditioner is on, the set temperature is 26℃, the current room temperature is 25.8℃, the lights are on, the brightness is 300 lux, the heat preservation device is on, the temperature is 45℃, and the water heater is on, the water temperature is 42℃.

[0073] SP5.2: Generate execution feedback data: The collected actual operating status data is organized into home appliance execution feedback data, and combined with the user's actual behavior data to form complete feedback information.

[0074] SP5.3: Feedback to Data Mart: The execution feedback data is sent back to the distributed spatiotemporal behavior data mart to update the association weights of the weighted spatiotemporal behavior metadata chain, adjust the association relationships between metadata, and improve the accuracy of subsequent metadata.

[0075] This method employs the aforementioned dual-layer metadata-driven mechanism, which includes metadata-driven distributed spatiotemporal behavior data marts and metadata-driven spatiotemporal behavior prediction models. The metadata of the data marts is used to organize and manage multi-source data, providing a data foundation for prediction. The metadata of the models is used to drive the computation and activation of the models. The two work together to drive the entire spatiotemporal behavior prediction and active control scheduling process, achieving efficient system operation. Specific Implementation Example 3: Please see Figure 7 As shown, a system framework diagram is provided. The following is a detailed description of the system content of this solution: The system adopts an edge-cloud collaborative architecture. The control flow is designed as follows: the multi-source spatiotemporal data acquisition module first collects raw multi-source spatiotemporal data and completes preliminary preprocessing. Then, the data is transmitted to the distributed spatiotemporal behavior data mart module for metadata transformation. The weighted spatiotemporal behavior metadata chain is then sent to the spatiotemporal behavior prediction model module to complete prediction calculation. The prediction results are sent to the active home device control command sequence generation module to generate control command sequences and send them to the home devices for execution through the protocol. Finally, the execution feedback data acquisition module collects actual operating status data and feeds it back to the distributed spatiotemporal behavior data mart module to form a closed loop. At the same time, the model optimization module performs backpropagation training and weight adjustment on the spatiotemporal behavior prediction model module.

[0077] The hardware components of the multi-source spatiotemporal data acquisition module include IoT sensors deployed in each room, Wi-Fi Bluetooth positioning modules, and smart home gateways. This module is interconnected via Zigbee or Wi-Fi networks and is responsible for collecting user location trajectory data, time-series behavioral data, home device status data, and environmental parameter data in real time. After completing timestamp synchronization, spatial coordinate mapping, and multimodal preprocessing, it outputs standardized data to the next module.

[0078] The hardware components of the distributed spatiotemporal behavior data mart module include an edge server cluster, with each server equipped with a CPU and GPU computing card and a large-capacity SSD storage. This module is connected to the multi-source spatiotemporal data acquisition module through a local area network. It is responsible for receiving standardized multi-source spatiotemporal data, converting it into a weighted spatiotemporal behavior metadata chain and storing it. At the same time, it receives execution feedback data to update the metadata chain in real time.

[0079] The hardware components of the spatiotemporal behavior prediction model module include an edge computing device with an integrated GPU accelerator card and sufficient memory. This module is connected to the distributed spatiotemporal behavior data mart module via PCIe or network. It is responsible for receiving the weighted spatiotemporal behavior metadata chain, extracting, fusing, calculating and activating spatiotemporal features in trajectory layer neurons, intent layer neurons and demand layer neurons, and outputting the prediction results of the user's future spatiotemporal behavior through forward propagation.

[0080] The hardware components of the proactive home device control command sequence generation module include an edge controller, which integrates an embedded processor and a protocol conversion chip. This module is connected to the spatiotemporal behavior prediction model module through an internal bus. It is responsible for receiving the user's future spatiotemporal behavior prediction results, performing chain matching of location movement trajectory, behavioral intention chain, and device demand probability, and using a multi-objective optimization algorithm to generate a proactive home device control command sequence that includes device start / stop timing, operating parameter adjustment values, and advance execution amount. The sequence is then sent to the corresponding home devices for execution via the Matter protocol or Zigbee protocol.

[0081] The hardware components of the execution feedback data acquisition module include built-in sensors of each home device and a smart home gateway. This module is connected to the active home device control command sequence generation module through the same protocol network. It is responsible for collecting actual operating status data after execution from the home devices as home device execution feedback data, and transmitting the feedback data to the distributed spatiotemporal behavior data mart module.

[0082] The hardware components of the model optimization module include a cloud server equipped with a high-performance CPU and GPU cluster. This module is connected to the spatiotemporal behavior prediction model module and the execution feedback data acquisition module through a secure encrypted channel. It is responsible for receiving execution feedback data, performing backpropagation training and weight adjustment on the spatiotemporal behavior prediction model, and realizing continuous system optimization. Specific Implementation Example 4: The following provides further supplementary explanations to the above embodiments: The spatiotemporal behavior prediction model is a three-layer forward propagation model based on trajectory layer neurons, intention layer neurons, and demand layer neurons. The model is initialized using the Xavier initialization method to randomly initialize the trajectory layer weight matrix, intention layer weight matrix, and demand layer weight matrix, and the bias vector is initialized to a zero vector. Before actual operation, offline pre-training is performed using sample data from the historical behavior log database. During training, the Adam optimizer is used to update the weights and biases to ensure that the model can accurately capture spatiotemporal features.

[0084] The Matter protocol is an IoT standard protocol that supports interoperability between IP and non-IP devices, enabling seamless communication between devices from different vendors. The Zigbee protocol is a low-power wireless communication protocol used to achieve highly reliable, low-latency data transmission and command issuance between sensors, gateways, and home devices.

[0085] The multi-objective optimization algorithm adopts a weighted sum optimization method, specifically using particle swarm optimization to dynamically optimize the weight coefficients, ensuring that comfort, energy consumption, and safety reach the global optimum.

[0086] After receiving standardized multi-source spatiotemporal data, the distributed spatiotemporal behavior data mart module uses a distributed consensus protocol (Raft algorithm) to synchronize data between three edge servers, ensuring that the weighted spatiotemporal behavior metadata chain is consistent in real time within the cluster.

[0087] The control command sequence generated by the proactive home device control command sequence generation module includes the target device identifier, start / stop command type, specific values ​​of operating parameters, execution lead time, and execution priority. After being sent through the Matter or Zigbee protocol, the home device receives the command, immediately parses it, and executes the corresponding action, realizing predictive-driven proactive control scheduling.

[0088] After receiving the execution feedback data, the model optimization module uses the backpropagation algorithm to update the weight matrix and bias vector of the spatiotemporal behavior prediction model, forming a complete closed-loop optimization process.

[0089] Detailed Design and Training of the Spatiotemporal Behavior Prediction Model: The spatiotemporal behavior prediction model is a three-layer forward propagation model based on trajectory layer neurons, intent layer neurons, and demand layer neurons, with 128, 256, and 128 nodes in each layer, respectively. Model initialization uses the Xavier initialization method to randomly initialize the weight matrices of the trajectory layer, intent layer, and demand layer, and the bias vector is uniformly initialized to zero. Before actual model operation, offline pre-training is performed using no less than 10,000 sample data entries from a historical behavior log database. The training epochs are set to 200, the batch size to 32, and the learning rate to 0.001. During training, the Adam optimizer is used to update the weights and biases, and the cross-entropy loss function is selected. Training stops when the validation set loss no longer decreases for 10 consecutive epochs to ensure that the model can accurately capture the user's spatiotemporal behavior characteristics.

[0090] Specific application specifications of the communication protocols: The Matter protocol uses version 1.2, supports interoperability between IP and non-IP devices, transmits data via UDP / IP at a baud rate of 115200, and enables seamless communication between devices from different vendors. The Zigbee protocol uses the 3.0 standard, is a low-power wireless communication protocol, operates in the 2.4GHz frequency band, has a transmission distance of 50 meters, and uses a star topology. It is suitable for highly reliable, low-latency data transmission and command issuance between sensors, gateways, and home devices, with a command transmission packet loss rate controllable below 0.1%.

[0091] The specific implementation of the multi-objective optimization algorithm: The multi-objective optimization algorithm adopts a weighted sum optimization method. The initial weight coefficients are allocated according to user comfort 0.5, equipment energy consumption 0.3, and response latency 0.2. The weight coefficients are dynamically optimized by the particle swarm optimization algorithm. The particle swarm size is 50, the number of iterations is 100, the inertia weight is 0.729, and the individual learning factor and global learning factor are both 1.494. Through the parameter settings, the user comfort, equipment energy consumption and system security are ensured to reach the global optimum.

[0092] Synchronization and Storage Mechanism of Distributed Data Mart: After receiving standardized multi-source spatiotemporal data, the distributed spatiotemporal behavior data mart module uses the Raft algorithm, a distributed consensus protocol, to synchronize data between three edge servers. The synchronization cycle is 10 seconds, ensuring that the weighted spatiotemporal behavior metadata chain is consistent in real time within the cluster. Data storage uses the time-series database InfluxDB, sharded and partitioned by data type. Location trajectory data is retained for 90 days, device status data for 30 days, environmental parameter data for 15 days, and historical behavior logs are permanently retained, supporting second-level data query requirements.

[0093] Control command format and execution rules: The control command sequence generated by the active home appliance control command sequence generation module is in JSON format, containing six core fields: target device identifier, start / stop command type, specific values ​​of operating parameters, execution lead time, execution priority, and command validity period. After the command is sent via Matter or Zigbee protocol, the home appliance receives the command and immediately performs a validity check. If the check passes, the command is cached in local Flash; if the check fails, an error message is returned to the gateway. The device executes operations according to command priority. High-priority safety protection commands can interrupt low-priority comfort adjustment commands. Commands that have not been executed within their validity period automatically expire, ensuring the accuracy and security of command execution.

[0094] The closed-loop optimization process of the model optimization module is as follows: After receiving execution feedback data, the model optimization module uses the backpropagation algorithm to update the weight matrix and bias vector of the spatiotemporal behavior prediction model. The optimization cycle is 24 hours, and each optimization uses the execution feedback data from the previous 24 hours as new training samples, with a sample size of no less than 500. The optimization process uses mini-batch incremental training with a batch size of 16 and a learning rate decaying to 0.0001. After training is completed, the updated model parameters are sent to the spatiotemporal behavior prediction model module on the edge side to achieve continuous iterative optimization of the model, forming a complete closed-loop optimization process.

[0095] Update rules for the weighted spatiotemporal behavior metadata chain: After receiving the execution feedback data, the distributed spatiotemporal behavior data mart module dynamically updates the association weights of trajectory metadata, intent metadata, and demand metadata. The normal update cycle is 1 hour. When the feedback data shows that the metadata association deviates from the user's actual behavior by more than 10%, an emergency update is immediately triggered to recalculate the association weights and ensure that the metadata chain can accurately match the dynamic changes in user behavior habits. Specific Implementation Example 5: The following describes the operating environment of the technical solution to demonstrate its real-time performance and feasibility: I. Hardware operating environment: The edge hardware uses an NVIDIA Jetson Nano 4GB or RK3588 8GB embedded development board as the core computing platform, integrating an NVIDIA T4 GPU accelerator card with a computing power of 130 TOPS. It is equipped with 32GB SSD local storage, a 1GB Ethernet port, and Zigbee 3.0 / Matter 1.2 protocol modules, supporting CUDA 11.4 and TensorRT 8.5 accelerated inference, ensuring a total latency of less than 800 milliseconds for a single prediction and instruction delivery. The edge devices are powered by DC12V / 2A and operate at temperatures from 0-40℃, suitable for typical home environments. The cloud hardware uses an Alibaba Cloud ECSg7 instance as the cloud server, equipped with two Intel Xeon Platinum 8369B CPUs, 128GB of memory, and a cluster of four NVIDIA A10 GPUs, with a single GPU computing power of 23 TOPS. It also features a 1TB SSD cloud disk and 100Mbps bandwidth, primarily used for offline training, parameter synchronization, and massive historical data storage for the model optimization module, supporting simultaneous access for over 1000 edge devices. The data acquisition layer hardware all uses low-power IoT sensors with an operating current of ≤10mA and a standby current of ≤1μA. Power supply options include battery power and PoE power, with battery-powered devices offering over one year of battery life. The Wi-Fi / Bluetooth positioning module uses Bluetooth 5.0 + Wi-Fi 2.4G dual-mode with a positioning accuracy of ±1 meter; the UWB positioning sensor has a positioning accuracy of ±0.3 meters, and the communication distance between the base station and the tag is ≥20 meters. All data acquisition devices have passed the national 3C certification, meeting home security standards.

[0097] II. Software Development and Deployment Environment: The development framework utilizes Python 3.10 and PyTorch 2.1 deep learning frameworks to build spatiotemporal behavior prediction models. Cross-platform deployment of the model is achieved through the ONNX 1.14 format, improving model inference speed by 30%. The front-end uses Vue 3.0 to develop the system's visual interface, while the back-end uses Django 4.2 to build service interfaces, supporting access from multiple devices including PCs, mobile devices, and home control screens. The data transmission protocol adopts a layered design. The underlying data acquisition uses the MQTT 3.1.1 protocol with a 30-second heartbeat interval to ensure online device monitoring. HTTPS is used for data transmission between the edge and the cloud, with AES-256 encryption before transmission to protect user privacy. Device-to-device command transmission uses Matter 1.2 or Zigbee 3.0 protocols to meet low-latency and low-power transmission requirements. System deployment must adhere to unified specifications. The edge computing platform is deployed within the home's low-voltage distribution box and directly connected to the smart home gateway via Ethernet. Data acquisition devices are wall-mounted or ceiling-mounted at a height of 2.2–2.5 meters, and the installation location must be unobstructed. The cloud server is deployed using containers based on Docker and Kubernetes, supporting elastic scaling. As the number of connected devices increases, computing nodes are automatically added. After system deployment, a 72-hour continuous stability test is required before the system can be officially put into use.

[0098] III. Equipment Compatibility and Expandability: The system is compatible with mainstream smart home devices from major brands and supports devices using mainstream communication protocols such as Matter 1.0 and above, Zigbee 3.0, Wi-Fi 2.4G, and Bluetooth 5.0. Users do not need to replace existing home devices, effectively reducing usage costs. When connecting new devices, they can be added with a single click through the system settings interface. The system can automatically identify the device type and controllable parameters, eliminating the need for manual configuration by the user. The system adopts a modular design, supporting layered expansion at the data acquisition layer, edge layer, and cloud layer. The data acquisition layer can be expanded with the number of sensors and positioning modules according to user needs, supporting up to 100 data acquisition devices simultaneously. The edge layer can add edge computing nodes to adapt to multi-family and multi-floor home environments. The cloud layer can add storage and computing nodes to support large-scale deployment at the community and neighborhood levels, demonstrating strong overall scalability.

[0099] IV. Fault Handling and Anomaly Warning: The system incorporates a device status monitoring module to monitor the operational status of the data acquisition devices, edge computing platform, and home appliances in real time. When a device goes offline, malfunctions, or experiences data anomalies, a red alert is immediately issued on the system interface, and users are notified via SMS and app push notifications. For minor faults such as sensor data drift, the system can automatically perform calibration; for serious faults such as edge computing platform offline, the system automatically switches to local emergency mode, using preset rules to control the devices and ensure normal operation of home appliances. When the acquired multi-source spatiotemporal data exhibits sudden changes in location coordinates or abnormal device status, the system automatically identifies and processes these issues using algorithms such as the 3σ criterion and wavelet denoising, removing abnormal data and supplementing missing data using interpolation to ensure data integrity and accuracy. When the data anomaly rate exceeds 20%, the system automatically triggers a self-check to investigate the cause of the fault. The edge computing platform is equipped with a 5000mAh backup power supply, automatically switching to backup power when the mains power is interrupted, providing a battery life of ≥4 hours to ensure the normal operation of core functions. The data acquisition devices employ a low-power design; battery-powered devices automatically enter energy-saving mode when the mains power is interrupted, extending battery life.

[0100] V. Prerequisites for using the system: During system initialization, at least 7 days of user historical behavior logs, with no less than 12 hours per day, must be collected to complete the initial construction of user behavior profiles before the spatiotemporal behavior prediction function can be activated. In multi-user home scenarios, each user must be equipped with an independent location tag, and the system must create independent profiles for each user to avoid prediction conflicts caused by confusion of behavioral characteristics among different users. Edge computing devices must maintain stable power supply and heat dissipation conditions to prevent GPU performance degradation due to ambient temperatures exceeding 40°C, which would affect model inference speed. All data transmission links must enable TLS 1.3 encryption, and encryption keys must be changed regularly to fully protect user privacy data. Firmware upgrades must be performed on edge servers and data collection devices every 3 months, and the system must be scanned for vulnerabilities to ensure device protocol compatibility and overall system security. Specific Implementation Example Six: The following are specific use cases of this solution: Use Case 1: Daily Living Scenarios of Elderly People Living Alone This case study focuses on the home-based elderly care scenario for seniors living alone, emphasizing system security and ease of use. The user is 72-year-old Grandma Li, residing in an 80㎡ two-bedroom apartment. She suffers from hypertension, has slightly slower mobility, and while her daily routine is generally regular, occasional emergencies occur. Her children do not live nearby, and she has limited experience operating smart home devices. Her core needs are easy operation, proactive device adaptation, and home security alerts. System deployment includes six UWB positioning sensors in the living room, bedroom, kitchen, bathroom, balcony, and doorway; four comprehensive environmental sensors (temperature, humidity, light, air quality, and human body detection); one smart home gateway; and one edge computing platform. Grandma Li is equipped with a smart bracelet integrating Bluetooth, Wi-Fi, and heart rate monitoring modules. The system also connects to smart lighting, smart air conditioning, a smart water heater, a small kitchen water heater, a smart door lock, and a gas alarm.

[0102] The system's complete execution process consists of six steps. The first step is data acquisition. The system uses a UWB positioning sensor to collect Grandma Li's location coordinates in real time at a sampling frequency of 1Hz. The smart bracelet simultaneously collects heart rate and behavioral data, and environmental sensors collect parameters such as temperature, humidity, and light intensity in various areas. The smart devices upload their operating status in real time. All data is transmitted to the edge computing platform after being timestamped and mapped to spatial coordinates. The second step is metadata conversion. The distributed spatiotemporal behavioral data mart converts the collected location trajectory data into trajectory metadata, identifying Grandma Li's core trajectory from 6:00 to 6:30 AM as getting up from her bedroom and walking to the bathroom. The time-series behavioral data is converted into intent metadata, identifying her intention to wash up and drink water after getting up. Device and environmental data are converted into demand metadata, identifying that she needs 42℃ warm water and 200 lux of light in the bathroom when washing up.

[0103] The third step is behavior prediction. The spatiotemporal behavior prediction model, through hierarchical feature extraction and fusion, accurately predicted at 5:50 AM that Grandma Li would get up around 6:10 AM and then move from the bedroom to the bathroom, indicating an intention to wash up and use warm water. The device demand probabilities for the bathroom light, the small water heater, and the bedroom curtains were 99%, 98%, and 95%, respectively. The fourth step is instruction generation. The active control instruction sequence generation module, through chain matching, identified the target devices as the smart bathroom light, the small water heater, and the smart bedroom curtains. Combining the device activation characteristics and the elderly person's movement characteristics, optimized instructions were generated. The bathroom light was turned on at 6:08 AM, with a brightness of 200 lux and a warm light of 3000K, turned on 2 minutes in advance to adapt to the elderly person's movement speed; the small water heater was turned on at 6:05 AM, heating to 42℃ and keeping warm, turned on 5 minutes in advance to allow for heating time; the bedroom curtains were slowly opened at 6:12 AM to avoid strong light stimulating the elderly person's eyes.

[0104] The fifth step is command issuance and execution. Commands are sent to each device via the Zigbee protocol, and each device executes the operation precisely at the specified time. The water heater preheats to the set temperature and keeps it warm, the bathroom lights turn on on time, and the curtains slowly open. The sixth step is feedback and optimization. After execution, the devices upload their operating status in real time, including whether the lights are on, the water temperature is 42℃, and the curtains are 30% open. The execution feedback data is transmitted to the data mart, and the system updates the metadata association weights based on the feedback. Considering that elderly people have slower mobility, the advance execution time of subsequent devices is fine-tuned to 3 minutes to further optimize the user experience.

[0105] In actual use, Grandma Li doesn't need to do anything manually. When she gets up and walks to the bathroom, the lights are already on, warm water is ready, and the curtains are slowly drawn to avoid strong light stimulation. The entire process is adapted to the elderly woman's rhythm and physical condition. At the same time, the smart bracelet's heart rate data is uploaded to the system in real time. When the heart rate exceeds 100 beats per minute, the system immediately sends an alert to her children and the community elderly care center, ensuring the elderly woman's home safety while providing convenience.

[0106] Use Case 2: Weekend Leisure Scenario for Multi-User Families This case study focuses on a multi-user family's leisure home scenario, emphasizing the system's personalized adaptation and conflict-free scheduling capabilities for multiple devices. The user background is Mr. Wang's family of three, living in a 120㎡ four-bedroom apartment. The family consists of a 38-year-old husband, a 36-year-old wife, and a 6-year-old son. They spend weekends relaxing at home, each member with different habits: the husband likes watching sports in the living room, the wife likes reading in the study, and the son likes playing with toys in his room. The core needs are personalized device control and conflict-free scheduling of multiple devices across multiple areas. For system deployment, eight UWB positioning sensors, six environmental sensors, and one edge computing platform are deployed within the home. Each family member is equipped with an independent positioning tag, connecting to smart lights, smart air conditioners, smart TVs, smart toys in the children's room, and an air purifier in the study, among other home devices.

[0107] In the complete system execution process, the data acquisition phase uses independent positioning tags to collect the location trajectories of three family members, capturing real-time location information of the husband in the living room, the wife in the study, and the son in the children's room. Environmental sensors collect temperature, humidity, and light parameters for each area, and each smart device uploads its operating status. All data is collected independently and modeled separately. In the metadata conversion phase, the data mart generates an independent metadata chain for each family member. For the husband, the generated trajectory metadata is "living room," the intent metadata is "watching a sports game," and the demand metadata is "TV on in the living room, air conditioner at 25°C, and lighting at 40% brightness." For the wife, the generated trajectory metadata is "study," the intent metadata is "reading," and the demand metadata is "lighting at 30% brightness in the study and air purifier on." For the son, the generated trajectory metadata is "children's room," the intent metadata is "playing with toys," and the demand metadata is "lighting at 50% brightness in the children's room and toy power on."

[0108] In the behavior prediction phase, the spatiotemporal behavior prediction model independently predicts the behavior of the three family members, predicting that the husband will watch a ball game in the living room for 1-2 hours, the wife will read in the study for 2-3 hours, and the son will play with toys in the children's room for 1 hour. The probability of demand for the corresponding equipment in each area is ≥90%. In the instruction generation phase, the control instruction generation module uses a multi-objective optimization algorithm to generate conflict-free, personalized control instruction sequences. In the living room area, the TV is turned on and tuned to a sports channel, the air conditioner is set to 25°C, and the lighting brightness is 40%. In the study area, the lighting is set to 30%, the warm light is 3500K, and the air purifier is turned on in silent mode. In the children's room area, the lighting is set to 50%, the toys are turned on, and the air conditioner is set to 27°C. All device instructions are executed independently by area without any scheduling conflicts.

[0109] During the command issuance and execution phase, commands are sent to devices in each area via the Matter+Zigbee protocol. Each device executes the operation precisely according to its personalized parameters, fully meeting the usage needs of different family members. In the feedback and optimization phase, after execution, the devices upload their operating status, and the feedback data updates the metadata chain of each family member. The system can capture changes in family members' behavior in real time, such as a son moving from his room to the living room, and dynamically adjust device parameters accordingly, such as turning off the lights in the children's room or adjusting the living room lights to a children's mode brightness, achieving dynamic adaptation of device control.

[0110] In actual use, three family members can enjoy personalized smart services tailored to their individual habits within the same home environment. Devices in each area are independently managed without any operational conflicts, and users do not need to manually adjust device parameters. The system can capture changes in family members' behavior in real time and dynamically optimize device control strategies, balancing individual needs with family collaboration, achieving a user comfort score of 9.1. Specific Implementation Example 7: The following are the specific experimental data for this scheme: It should be noted that prediction accuracy refers to the degree of matching between the predicted user's future spatiotemporal behavior and the actual behavior; control command response time refers to the time from the predicted output to the device's completion of execution; energy saving rate is calculated by comparing with traditional rule-based control systems; and user comfort score is obtained through subjective evaluations from 30 real users. Experimental results show that this solution can achieve high accuracy and low latency active control scheduling in various home scenarios.

[0112] Please see Figure 8 The image shown is a screenshot of the system's main interface, used to demonstrate the system's practical application; please refer to [link / reference]. Figure 9 The diagram shows the system data acquisition and interface, illustrating the data types and actual acquisition status.

[0113] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0114] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A proactive home appliance control and scheduling method based on spatiotemporal behavior prediction, characterized in that: Includes the following steps: Sp1. Acquire multi-source spatiotemporal data of the home environment, including user location trajectory data, time-series behavior data, home device status data, environmental parameter data, and historical behavior logs; Sp2. Construct at least one distributed spatiotemporal behavior data mart and receive the multi-source spatiotemporal data. In the distributed spatiotemporal behavior data mart, transform the multi-source spatiotemporal data into a weighted spatiotemporal behavior metadata chain. The weighted spatiotemporal behavior metadata chain includes trajectory metadata, intent metadata, and demand metadata. Sp3. Input the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation, and output the prediction results of the user's future spatiotemporal behavior through forward propagation; Sp4. Generate an active home device control command sequence based on the predicted results of the user's future spatiotemporal behavior, and send it to the corresponding home devices for execution; Sp5. Obtain the execution feedback data of the home devices and feed it back to the distributed spatiotemporal behavior data mart.

2. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method for obtaining multi-source spatiotemporal data of the home environment in Sp1 is as follows: Sp1.1 uses IoT sensors and Wi-Fi / Bluetooth positioning modules deployed in each room to collect user location trajectory data and time-series behavior data in real time, and performs timestamp synchronization and spatial coordinate mapping processing. SP1.2: Obtain real-time status data of home devices through the smart home gateway interface; SP1.3, synchronously collect environmental parameter data; SP1.4 Extract user's past spatiotemporal behavior records from the historical behavior log database; Sp1.5 uses a multimodal data preprocessing module to perform denoising, normalization, and spatiotemporal sequence alignment on the acquired multi-source spatiotemporal data.

3. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method for constructing a distributed spatiotemporal behavior data mart and transforming a weighted spatiotemporal behavior metadata chain in Sp2 is as follows: Sp2.1, Construct a distributed spatiotemporal behavior data mart; Sp2.2, convert user location trajectory data into trajectory metadata; Sp2.3, transform time-series behavioral data into intent metadata; Sp2.

4. Transform home appliance status data and environmental parameter data into demand metadata; Sp2.5: Establish weighted associations between trajectory metadata, intent metadata, and demand metadata to form a weighted spatiotemporal behavior metadata chain.

4. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method for inputting the weighted spatiotemporal behavior metadata chain into the spatiotemporal behavior prediction model for calculation and activation in Sp3 is as follows: Sp3.1 Input the weighted spatiotemporal behavior metadata chain into the trajectory layer neurons, intention layer neurons, and demand layer neurons of the spatiotemporal behavior prediction model; Sp3.2, perform spatiotemporal feature extraction, fusion calculation and activation in neurons of each layer; Sp3.3 outputs predictions of users' future spatiotemporal behavior through forward propagation, including user location movement trajectory, behavioral intent chain, and device demand probability.

5. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method for generating and executing a sequence of proactive home device control commands based on the prediction of the user's future spatiotemporal behavior in Sp4 is as follows: SP4.1: Perform chain matching of user location movement trajectory, behavioral intent chain and device demand probability to determine target home device and optimal control time; SP4.2 uses a multi-objective optimization algorithm to generate a sequence of proactive home appliance control instructions that includes device start / stop timing, operating parameter adjustment values, and advance execution amounts; Sp4.

3. The active home device control command sequence is sent to the corresponding home devices for execution via the Matter protocol or Zigbee protocol to achieve predictive-driven active control scheduling.

6. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method for obtaining and feeding back home device execution feedback data in Sp5 is as follows: SP5.1: Collect actual operating status data from the corresponding home appliances after execution; SP5.2, Use the actual operating status data as the feedback data for home appliances; Sp5.

3. Feedback data from the home devices is fed back to the distributed spatiotemporal behavior data mart to update the weighted spatiotemporal behavior metadata chain.

7. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method includes a two-layer metadata-driven mechanism, which includes distributed spatiotemporal behavior data mart metadata-driven mechanism and spatiotemporal behavior prediction model metadata-driven mechanism. The two-layer metadata-driven mechanism is used to drive spatiotemporal behavior prediction and proactive home device control and scheduling.

8. The active home appliance control and scheduling method based on spatiotemporal behavior prediction according to claim 1, characterized in that: The method is divided into the following three stages based on the prediction time length of the user's future spatiotemporal behavior prediction results: During the short-term trajectory prediction period, metadata of the user's location movement sequence is generated in the first distributed spatiotemporal behavior data mart and trajectory layer neurons. During the mid-term intent prediction period, user behavior intent sequence metadata is generated in the second distributed spatiotemporal behavior data mart and intent layer neurons. During the long-term demand forecasting period, metadata of home appliance demand sequences is generated using the third distributed spatiotemporal behavioral data mart and demand layer neurons.

9. The system corresponding to the active home appliance control and scheduling method based on spatiotemporal behavior prediction according to any one of claims 1-8, characterized in that: The system includes: The multi-source spatiotemporal data acquisition module is used to execute step Sp1 to acquire multi-source spatiotemporal data of the home environment; The distributed spatiotemporal behavior data mart module is connected to the multi-source spatiotemporal data acquisition module and is used to perform the Sp2 step, receiving the multi-source spatiotemporal data and converting it into a weighted spatiotemporal behavior metadata chain; The spatiotemporal behavior prediction model module is connected to the distributed spatiotemporal behavior data mart module and is used to execute the Sp3 step, receive the weighted spatiotemporal behavior metadata chain for calculation and activation, and output the prediction result of the user's future spatiotemporal behavior. An active home device control command sequence generation module is connected to the spatiotemporal behavior prediction model module and is used to execute the Sp4 step, which generates an active home device control command sequence based on the user's future spatiotemporal behavior prediction results and sends it to the corresponding home devices for execution. The execution feedback data acquisition module is used to execute the Sp5 step, acquire the execution feedback data of the home device, and feed it back to the distributed spatiotemporal behavior data mart module.

10. The proactive home appliance control and scheduling system based on spatiotemporal behavior prediction according to claim 9, characterized in that: The system also includes a model optimization module, which is connected to the spatiotemporal behavior prediction model module and the execution feedback data acquisition module, and is used to perform backpropagation training and weight adjustment on the spatiotemporal behavior prediction model.

Images