Smart home pre-activation method based on vehicle driving data

Abstract

The invention provides a smart home pre-activation method based on vehicle driving data, and the method comprises the steps: carrying out the real-time processing of a dynamic state vector through employing a pre-established lightweight neural network model, judging whether a vehicle is in a home way or not, and outputting a home returning probability value; according to the distance estimation value and the current speed data, using a Kalman filtering algorithm to predict the remaining duration of the vehicle arriving at the residence, and outputting an initial duration prediction value; acquiring real-time traffic flow data and a vehicle acceleration change trend, and correcting the initial time length predicted value through a dynamic adaptation module to obtain an optimized time length value; when the optimized time length value reaches a preset triggering opportunity threshold range, a pre-activation instruction is sent to the smart home device through an Internet of Things protocol, and a device preparation state is determined; and for the control instruction sequence, a time sequence analysis algorithm is adopted to evaluate the fluctuation range of the triggering opportunity, and a triggering stability index is obtained.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and particularly to a smart home pre-activation method based on vehicle driving data. Background Art

[0002] Problem Background: The research fields of vehicle intelligent control and Internet of Things technology are of crucial importance in modern society. With the rapid development of intelligent transportation and smart homes, how to improve the convenience and comfort of drivers' lives through technical means has become an important direction for promoting industry progress. This field not only concerns vehicle operation efficiency but also directly affects user experience and energy management efficiency, with significant potential economic and social benefits. However, there are still obvious limitations in the existing methods for achieving seamless linkage between vehicles and home appliances. Many solutions rely on static rules or simple time prediction, making it difficult to adapt to the dynamic changes in vehicle operation status and the variability of user behavior, resulting in inaccurate triggering times of control instructions. The responses of smart home devices are often lagged or advanced, reducing the practicality of the system.

[0003] In this context, core challenges have gradually emerged, mainly focusing on three technical factors: the real-time update and analysis of vehicle state data, the accurate judgment of the vehicle's state on the way home, and the calculation of the remaining duration based on dynamic data. Since vehicle operation state data may change rapidly due to road conditions, driving habits, etc., existing neural network models face difficulties in continuously updating inputs and accurately judging whether the vehicle is on the way home, such as data processing delays or misjudgments. At the same time, how to accurately calculate the distance between the vehicle and the residence through the current speed and position and further deduce a reasonable remaining duration has become complicated due to the lack of a dynamic adaptation mechanism. These unresolved technical factors directly lead to the inability of smart home devices to be activated at the best time, affecting the intelligence level and user satisfaction of the system.

[0004] Therefore, how to continuously update data through a neural network model and accurately judge whether the vehicle is on the way home under the condition of continuous change of vehicle operation state, and at the same time calculate a reasonable remaining duration based on dynamic speed and position to ensure that smart home devices are triggered within an appropriate threshold has become a key issue in improving the linkage efficiency between vehicles and the Internet of Things. The solution to this problem will directly determine whether the system can achieve efficient and accurate intelligent control in complex real-world scenarios. Summary of the Invention

[0006] The present invention provides a smart home pre-activation method based on vehicle driving data, mainly including: Obtain real-time speed, acceleration, and direction data from vehicle sensors, and perform preliminary feature extraction on the vehicle operation state through an edge computing module to obtain a dynamic state vector; For the dynamic state vector, a pre-established lightweight neural network model is used for real-time processing to determine whether the vehicle is on the way home and output the home probability value; If the home probability value exceeds the preset threshold of 0.8, obtain the current speed data and current position data from the navigation system, and calculate the real-time distance between the vehicle and the residence in combination with the historical road condition data to obtain the distance estimation value; According to the distance estimation value and the current speed data, apply the Kalman filter algorithm to predict the remaining time for the vehicle to reach the residence and output the initial time prediction value; Obtain the real-time traffic flow data and the vehicle acceleration change trend, and correct the initial time prediction value through the dynamic adaptation module to obtain the optimized time value; When the optimized time value reaches the preset trigger timing threshold range, send a pre-activation instruction to the smart home device through the Internet of Things protocol to determine the device preparation status; Receive a feedback signal from the smart home device, determine whether the device has completed pre-activation. If it has, adjust the final trigger time according to the optimized time value and output the control instruction sequence; For the control instruction sequence, use the time series analysis algorithm to evaluate the fluctuation range of the trigger timing and obtain the trigger stability index; If the trigger stability index is lower than the preset threshold of 0.9, update the neural network model parameters through the feedback loop to optimize the accuracy of the judgment on the way home and output the adjusted state judgment result.

[0007] The technical solution provided by the embodiment of the present invention may include the following beneficial effects: The present invention discloses a smart home pre-activation method based on vehicle driving data. This method analyzes vehicle sensor data in real time and uses a lightweight neural network model to determine whether the vehicle is on the way home. When the home probability exceeds the threshold, in combination with the navigation system and historical road condition data, the Kalman filter algorithm is applied to predict the arrival time at home. Subsequently, the present invention considers the real-time traffic flow and vehicle acceleration changes and dynamically adjusts the prediction duration. At the appropriate time, the present invention sends a pre-activation instruction to the smart home device and adjusts the final trigger time according to the device feedback. The trigger stability is evaluated through time series analysis, and the neural network model parameters are updated if necessary to improve the judgment accuracy. The present invention realizes the seamless connection between the vehicle driving state and the smart home system, improving the intelligent level of home automation and the user experience. Description of the Drawings

[0008] Figure 1 It is a flowchart of a smart home pre-activation method based on vehicle driving data of the present invention.

[0009] Figure 2Schematic diagram of a smart home pre-activation method based on vehicle driving data according to the present invention.

[0010] Figure 3 Another schematic diagram of a smart home pre-activation method based on vehicle driving data according to the present invention. Detailed implementation manners

[0011] Next, the technical solutions of the present invention will be described clearly and completely in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0012] As Figures 1 - 3 , a smart home pre-activation method based on vehicle driving data in this embodiment may specifically include: S101. Obtain real-time speed, acceleration, and direction data from vehicle sensors, and perform preliminary feature extraction on the vehicle running state through an edge computing module to obtain a dynamic state vector.

[0013] Obtain real-time speed, acceleration, and direction data from the sensors, and perform preliminary processing through the edge computing module to obtain an original data set. For the original data set, a data cleaning method is used to remove noise to obtain a clean data set. Through the clean data set, a feature extraction algorithm is used to analyze the change trends of the real-time speed and acceleration to obtain a state feature set. According to the state feature set, if the real-time speed and acceleration exceed the preset thresholds, the vehicle running state is judged through the direction data to obtain a state classification result. Using the state classification result, the vehicle's motion trajectory is calculated in combination with the dynamic vector to obtain a trajectory feature set. Through the trajectory feature set, a clustering algorithm is used to group the vehicle running states to obtain a set of running modes. According to the set of running modes, the vehicle's dynamic state vector is determined to obtain a final state description.

[0014] Specifically, the vehicle sensors collect real-time data through the CAN bus at a period of 100 milliseconds, such as the current speed of 72 km / h, the longitudinal acceleration of 3 m / s², and the steering wheel angle of 15 degrees. The edge computing module processes the time-series data using the sliding window method, sets the window length to 10 sampling points (i.e., 1 second), and denoises the original data through the Kalman filtering algorithm, where the process noise covariance matrix Q is set to [1 0; 0 05], and the observation noise covariance matrix R is fixed at 01. In the feature extraction stage, the statistical features of the data within the window are calculated, including the speed standard deviation of 2 km / h, the acceleration mean of 28 m / s², and the direction change rate of 4 degrees / second. At the same time, the frequency domain features of the acceleration signal are analyzed through the fast Fourier transform, and the energy ratio of the 5 - 2 Hz frequency band extracted as the vibration feature is 35%. The dynamic state vector is finally constructed as a 7-dimensional feature space [72, 3, 15, 2, 28, 4, 35]. When using the principal component analysis method for dimensionality reduction, the first 3 principal components are retained, and the cumulative contribution rate reaches 92%. In the state evaluation session, the feature vector is input into a pre-trained XGBoost model (learning rate 01, tree depth 5), and the vehicle stability score of 87 is output (a threshold of 8 or above is determined as a stable state). At the same time, the DBSCAN clustering algorithm (eps = 5, min_samples = 3) is used to detect abnormal driving patterns, and the Euclidean distance between the current data point and the core cluster is 21, which is lower than the threshold of 3.

[0015] S102. For the dynamic state vector, use a pre-established lightweight neural network model for real-time processing to determine whether the vehicle is on the way home and output the probability value of being on the way home.

[0016] For the dynamic state, obtain real-time data through sensors to get the state vector. Through the state vector, use a lightweight model for feature analysis to obtain the analysis result. According to the analysis result, use a neural network to process the data input to obtain a preliminary judgment value. If the preliminary judgment value exceeds the preset threshold, then make a vehicle judgment through the attribute of being on the way home to obtain an adjusted judgment value. Through the adjusted judgment value, combined with the real-time processing flow, determine the probability value. According to the probability value, use the output result to generate the final state to obtain the probability value of being on the way home. Through the probability value of being on the way home, combined with the vehicle judgment logic, determine the vehicle dynamic behavior classification.

[0017] Specifically, in the processing of the dynamic state vector, first, data such as the vehicle's speed, acceleration, and direction angle are collected in real-time through in-vehicle sensors. These data are updated at a frequency of 10 times per second, forming a multi-dimensional time series. Then, a pre-established lightweight neural network model is used. This model adopts the convolutional neural network (CNN) structure, including 3 convolutional layers and 2 fully connected layers. The input layer receives time series data with a length of 100, and the output layer is a Sigmoid activation function used to output the probability value of going home. During the training phase of the model, a dataset containing 100,000 historical driving records is used. Each record contains information such as the vehicle's driving trajectory, timestamp, and destination. The dataset is divided into a training set and a validation set through the cross-validation method. During the training process, the Adam optimizer is used, and the learning rate is set to 0.01. After 50 epochs of training, the accuracy of the model on the validation set reaches 95%. In the real-time processing phase, the system inputs the currently collected dynamic state vector into the trained model. The model extracts spatial features in the time series through the convolutional layer, and the fully connected layer further fuses these features. Finally, a probability value between 0 and 1 is output, indicating the likelihood that the vehicle is on the way home.

[0018] For example, when the model outputs a probability value of 85, the system determines that the vehicle is highly likely to be on the way home, combines map data to plan the optimal route, and synchronizes this information to the user's mobile device so that the user can understand the vehicle's status in real-time. In addition, the system will also dynamically adjust the model parameters according to historical data and real-time road conditions to ensure the accuracy and real-time nature of the prediction results.

[0019] S103: If the probability value of going home exceeds the preset threshold of 0.8, obtain the current speed data and current location data from the navigation system, and calculate the real-time distance between the vehicle and the residence in combination with historical road condition data to obtain a distance estimate value.

[0020] S104: According to the distance estimate value and the current speed data, apply the Kalman filter algorithm to predict the remaining time for the vehicle to reach the residence, and output an initial time prediction value.

[0021] Obtain the distance estimation value and the current speed data through sensors, use the Kalman filtering algorithm to calculate the initial duration prediction value, and obtain the preliminary remaining duration. Extract the speed change trend from the preliminary remaining duration, update the state equation of the Kalman filter for the speed change, and determine the adjusted duration prediction value. According to the adjusted duration prediction value combined with the vehicle position data, calculate the real-time distance estimation value through the position update formula to obtain the updated remaining duration. If the difference between the updated remaining duration and the preliminary remaining duration exceeds the preset threshold, smooth the speed change through the Kalman filter and judge the optimized duration output. After obtaining the optimized duration output, for the real-time change of the vehicle position, use the interpolation method to calculate the continuous duration adjustment value and determine the final remaining A duration. Update the observation equation of the Kalman filter through the final remaining duration combined with the dynamic change of the current speed to obtain a stable duration prediction result. Extract the position update data from the stable duration prediction result and output the accurate remaining duration value through the iterative calculation of distance estimation.

[0022] Specifically, assume that the initial distance estimation value of the current vehicle from the residence is 5000 meters, and the current speed provided by the GPS system is 60 km / h (converted to 16.7 m / s). The state vector of the Kalman filtering algorithm is set as X = [distance, speed]ᵀ, where the distance observation noise variance R = 25 square meters, and the process noise covariance matrix Q = [[1, 1], [1, 01]]. In the initialization stage, the state covariance matrix P takes the value of [[100, 0], [0, 10]], indicating the uncertainty of the initial distance and speed estimation. In the prediction step, according to the kinematic equation X K = FX K ₋1 + W K , where the state transition matrix F = [[1, -Δt], [0, 1]] (taking Δt = 1 second), the prior state estimation X K ⁻ = [5000 - 16.7×1, 16.7]ᵀ = [4983.3, 16.7]ᵀ is calculated. Then update the prior covariance P K ⁻ = FP K ₋1Fᵀ + Q, and the new covariance value is obtained as [[100.1, -9], [-9, 01]] through matrix operations. When the new observation value Z K = [4950 meters, 15 m / s] arrives, calculate the Kalman gain K = P K ⁻Hᵀ(HP K ⁻Hᵀ + R)⁻¹ (the observation matrix H takes the identity matrix), and the gain matrix [[80.1, 0], [0, 092]] is obtained. The final state is updated to X K = X K ⁻ + K(Z K - HX K⁻) = [4978, 159]ᵀ. At this time, the remaining distance of 4978 meters divided by the corrected speed of 159 meters per second gives the first predicted arrival time of 298 seconds (about 5 minutes). During the process, the system continuously monitors that the standard deviation of speed fluctuations is 3 meters per second. Through the covariance update formula P K =(I - KH)P K ⁻ reduces the error covariance to [[18, 0], [0, 92]], providing a more accurate basis for the next iteration.

[0023] S105. Obtain real-time traffic flow data and the trend of vehicle acceleration changes, and correct the initial duration prediction value through the dynamic adaptation module to obtain an optimized duration value.

[0024] Obtain real-time data from road nodes through the sensor network to obtain the traffic flow distribution. Extract vehicle acceleration information based on the traffic flow distribution to determine the acceleration sequence. Use time series analysis methods to process the acceleration sequence to obtain a trend curve. Analyze the trend curve through a preset dynamic adaptation module to judge the correction coefficient. If the correction coefficient exceeds the preset threshold, adjust it in combination with the initial duration data to obtain an adjusted duration value. Use a linear regression algorithm to fit the adjusted duration value and the predicted value to obtain the optimized duration result. Determine the final duration value through module processing of the optimized duration result.

[0025] Specifically, through sensors and in-vehicle devices deployed at major traffic intersections, real-time traffic flow data and vehicle acceleration information are collected. For example, at a certain intersection, the number of vehicles passing through per minute during the morning rush hour is 120, and the average vehicle acceleration is 5 m / s². Use the Kalman filter algorithm to process the collected data to eliminate noise interference and obtain a smooth traffic flow curve and the trend of acceleration changes. Based on historical data and real-time data, use a long short-term memory network (LSTM) model to predict the initial duration. For example, predict the travel time of a certain section of the road to be 15 minutes. Through the dynamic adaptation module, compare and analyze the real-time traffic flow and the trend of acceleration changes with the initial prediction value. If it is found that the current traffic flow has increased by 20% compared to the prediction, and the vehicle acceleration has decreased by 2 m / s², then adjust the initial prediction value according to the preset correction coefficient to obtain an optimized travel duration of 18 minutes. This optimized value will be updated in real time to the navigation system to provide more accurate route planning suggestions for drivers.

[0026] S106. When the optimized duration value reaches the preset trigger timing threshold range, send a pre-activation instruction to the smart home device through the Internet of Things protocol to determine the device preparation status.

[0027] By monitoring and optimizing the duration value, determine whether it reaches the threshold range corresponding to the trigger timing to obtain a determination result. If the determination result shows that the threshold range is reached, generate a pre-activation instruction through the Internet of Things protocol and determine the instruction content. Use the protocol transmission module to process the pre-activation instruction, send the instruction to the smart home device, and obtain the sending status. Confirm whether the instruction is successfully transmitted to the smart home according to the sending status, and judge the device reception situation. Extract the device preparation information from the data returned by the device to determine the device status. Classify and process the device status to obtain a status classification result. Adjust the subsequent instruction sending frequency according to the status classification result to determine the optimized sending strategy.

[0028] Specifically, when the optimized duration value reaches the preset trigger threshold (such as the predicted remaining time is less than 10 minutes), the system sends a pre-activation instruction to the smart home gateway through the MQTT protocol. The instruction includes the device type, the expected start time (such as 5 minutes later), and the operating parameters (such as the target temperature of the air conditioner is 26°C). After receiving the instruction, the gateway first checks the current status (standby / running) of the target device (such as the living room air conditioner). If the device is in the standby state, it sends a wake-up signal and starts a self-check program to detect whether the data of the sensor (such as the temperature and humidity sensor) is normal (the current room temperature is 30°C, and the humidity is 60%). If the deviation exceeds the threshold (±2°C), a calibration process is triggered. At the same time, the gateway queries the device power (remaining 80%) and network latency (average 50 ms) through the Zigbee protocol. If the conditions are met (power > 50%, latency < 100 ms), a ready signal is returned. After receiving the confirmation, the system marks the device status as "pre-activated" and starts a countdown (300 seconds). During this period, the environmental changes (such as the room temperature drops 5°C per minute) are continuously monitored. If an unexpected abnormality occurs (such as the temperature suddenly rises 3°C), the trigger conditions are re-evaluated. When the countdown reaches zero, the system sends a final execution instruction, and the device starts according to the preset parameters, and the real-time operation data (the current power is 800W, and the outlet temperature is 24°C) is fed back to the cloud database for closed-loop optimization.

[0029] S107. Receive a feedback signal from the smart home device, determine whether the device has completed pre-activation. If it has, adjust the final trigger time according to the optimized duration value and output a control instruction sequence.

[0030] Obtain a feedback signal from a smart home device, and determine whether the device has completed pre-activation through a signal processing module to obtain a determination result. If the determination result shows that the pre-activation is completed, the time calculation module is used to process the optimized duration value to determine the adjusted trigger time. A control instruction is generated according to the adjusted trigger time, and the instruction content is determined through an instruction encoding module. The transmission protocol is used to process the control instruction, send the instruction to the smart home device, and obtain the sending status. Whether the instruction is successfully transmitted to the device is judged through the sending status to determine the device reception situation. The device feedback information is extracted from the data returned by the device, and the device status is judged through an information parsing module to obtain a status classification result. The subsequent instruction generation frequency is adjusted according to the status classification result, and the optimized sending strategy is determined through a frequency optimization module.

[0031] Specifically, the smart home device sends a feedback signal through a wireless communication protocol such as ZigBee. The signal contains a device status code and the current timestamp. For example, the status code 0x01 indicates that the pre-activation is completed, and the timestamp is 2023-10-25T14:30:45. The central controller uses a sliding window algorithm to verify the status codes for 5 consecutive times. When the number of occurrences of 0x01 in the window is ≥4, it is determined to be effectively completed. Subsequently, the system calls an optimization algorithm to calculate the trigger time. Assuming that the preset basic trigger delay is 10 seconds, the historical operation data is analyzed through a Bayesian optimization model. This model takes user habit parameters (such as an average response delay of 3 seconds) and device performance parameters (such as a relay switching time of 5 seconds) as inputs and outputs an optimal adjustment value ΔT = 7 seconds. The final trigger time is corrected to 17 seconds. The control instruction sequence is generated using a priority queue. For example, [0xA1 close the curtain, 0xB2 turn on the air conditioner] is arranged in ascending order of device response delay. Each instruction is attached with an execution time deviation tolerance of ±3 seconds and is triggered synchronously by a real-time clock chip. In the instruction sending stage, the CRC-16 checksum is used to ensure the transmission integrity. The checksum polynomial is 0x8005. When the check fails 3 consecutive times, the retransmission mechanism of the 8014 protocol is triggered.

[0032] S108. For the control instruction sequence, a time series analysis algorithm is used to evaluate the fluctuation range of the trigger timing to obtain a trigger stability index.

[0033] The instruction sequence is processed through a time series analysis algorithm to obtain fluctuation range data. The change trend of the trigger timing is calculated based on the fluctuation range data to determine the preliminary stability evaluation value. If the change trend exceeds the preset threshold, the sliding window method is used to adjust the fluctuation range to obtain the optimized trigger timing distribution. The stability index is calculated through the optimized trigger timing distribution to obtain the timing characteristics of the index. Fourier transform processing is performed on the timing characteristics to obtain the fluctuation components in the frequency domain. The trigger stability is judged based on the fluctuation components in the frequency domain to obtain the final stability index value. The key features are extracted from the final stability index value to determine the adjustment basis for the control instruction.

[0034] Specifically, in the time series analysis of the control instruction sequence, first, 1000 consecutive trigger time interval samples are collected at a sampling frequency of 1 kHz to ensure that the data accuracy reaches the millisecond level. The sliding window method is used to preprocess the original data. The window width is set to 50 samples, and the step size is 10 samples. Outliers are eliminated through median filtering. For example, the outlier points exceeding ±3σ are replaced with the 25th percentile within the window. An ARIMA(2,1,1) model is established for the preprocessed data. Among them, the autoregressive order 2 reflects the linear relationship between the current trigger interval and the previous two intervals. The differencing order 1 eliminates the trend term, and the moving average order 1 is used to model the residual correlation. The model parameters are obtained through maximum likelihood estimation: φ1 = 45 ± 03, φ2 = -21 ± 02, θ1 = 33 ± 04. The prediction interval is calculated based on the fitted model. When the confidence level is 95%, the fluctuation range of the trigger timing is [92 ms, 108 ms]. Further, the stability index SDI = 1 - (actual fluctuation range / design tolerance) is calculated. When the design tolerance is ±5 ms, SDI = 932, indicating that the system has high stability. To verify the effectiveness of the model, a residual Ljung-Box test (Q = 16, p = 21 > 05) is performed to confirm that the residuals are a white noise sequence. Finally, a multiple regression relationship is established between the SDI index and the system response delay (mean 12 ms), and it is found that for every 1 increase in stability, the delay can be reduced by 2 ms (R² = 89), providing a quantitative basis for parameter tuning.

[0035] S109. If the trigger stability index is lower than the preset threshold of 0.9, the neural network model parameters are updated through a feedback loop to optimize the accuracy of the judgment during the return journey, and the adjusted state judgment result is output.

[0036] If the stability index is lower than the preset threshold, the real-time state information is obtained through sensor data to determine the preliminary state judgment. The deviation data is extracted from the preliminary state judgment through a feedback loop to update the neural network model parameters. The updated model parameters are used to recalculate the state during the return journey to obtain the adjusted state data. If there is a difference between the adjusted state data and the expected value, the model parameters are optimized through the gradient descent algorithm to judge the optimized state result. The trend of the state change during the return journey is generated based on the optimized state result, and the trend data is output. The input weights of the neural network are adjusted through the trend data to determine the final state judgment result. After obtaining the final state judgment result, the result consistency is verified through preset rules, and the verified state information is output.

[0037] As described above, this is only the specific implementation manner of this specification. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein. It should be understood that the protection scope of this specification is not limited thereto. Any person skilled in the art within the technical scope disclosed in this specification can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of this specification.

Claims

1. A smart home pre-activation method based on vehicle driving data, characterized in that The method includes: Obtain real-time speed, acceleration, and direction data from vehicle sensors, and perform preliminary feature extraction on the vehicle's operating state through an edge computing module to obtain a dynamic state vector. For the dynamic state vector, use a pre-established lightweight neural network model for real-time processing to determine whether the vehicle is on the way home and output a home probability value. If the home probability value exceeds the preset threshold of 0.8, obtain the current speed data and current position data from the navigation system, and calculate the real-time distance between the vehicle and the residence in combination with historical road condition data to obtain a distance estimate value. According to the distance estimate value and the current speed data, apply the Kalman filter algorithm to predict the remaining time for the vehicle to reach the residence and output an initial time prediction value. Obtain real-time traffic flow data and the vehicle acceleration change trend, and correct the initial time prediction value through a dynamic adaptation module to obtain an optimized time value. When the optimized time value reaches the preset trigger timing threshold range, send a pre-activation instruction to the smart home device through the Internet of Things protocol to determine the device preparation status. Receive a feedback signal from the smart home device, determine whether the device has completed pre-activation. If it has, adjust the final trigger time according to the optimized time value and output a control instruction sequence. For the control instruction sequence, use a time series analysis algorithm to evaluate the fluctuation range of the trigger timing and obtain a trigger stability index. If the trigger stability index is lower than the preset threshold of 0.9, update the neural network model parameters through a feedback loop to optimize the accuracy of the judgment on the way home and output an adjusted state judgment result.

2. The method according to claim 1, wherein The step of obtaining real-time speed, acceleration, and direction data from vehicle sensors, and performing preliminary feature extraction on the vehicle's operating state through an edge computing module to obtain a dynamic state vector includes: Obtain real-time speed, acceleration, and direction data from the sensors, and perform preliminary processing through the edge computing module to obtain an original data set. For the original data set, use a data cleaning method to remove noise and obtain a clean data set. Through the clean data set, use a feature extraction algorithm to analyze the change trends of the real-time speed and acceleration to obtain a state feature set. According to the state feature set, if the real-time speed and acceleration exceed the preset threshold, judge the vehicle's operating state through the direction data to obtain a state classification result. Use the state classification result, combine with the dynamic vector to calculate the vehicle's motion trajectory to obtain a trajectory feature set. Through the trajectory feature set, use a clustering algorithm to group the vehicle's operating state to obtain an operating mode set. According to the operating mode set, determine the vehicle's dynamic state vector and obtain the final state description.

3. The method according to claim 1, wherein The step of using a pre-established lightweight neural network model for real-time processing of the dynamic state vector to determine whether the vehicle is on the way home and output a home probability value includes: For the dynamic state, obtain real-time data through the sensors to obtain a state vector. Through the state vector, use a lightweight model for feature analysis to obtain an analysis result. According to the analysis result, use the neural network to process the data input to obtain a preliminary judgment value. If the preliminary judgment value exceeds the preset threshold, the vehicle is judged through the attributes during the journey home to obtain an adjusted judgment value; Based on the adjusted judgment value and combined with the real-time processing flow, determine the probability value; According to the probability value, adopt the output result to generate the final state and obtain the probability of arriving home; Based on the probability of arriving home and combined with the vehicle judgment logic, determine the classification of vehicle dynamic behavior.

4. The method according to claim 1, characterized in that, The prediction of the remaining duration for the vehicle to reach the residence by applying the Kalman filtering algorithm based on the distance estimation value and the current speed data, and output the initial duration prediction value, including: Obtain the distance estimation value and the current speed data through sensors, and calculate the initial duration prediction value using the Kalman filtering algorithm to obtain the preliminary remaining duration; Extract the speed change trend from the preliminary remaining duration, update the state equation of the Kalman filter for the speed change, and determine the adjusted duration prediction value; According to the adjusted duration prediction value and combined with the vehicle position data, calculate the real-time distance estimation value through the position update formula to obtain the updated remaining duration; If the difference between the updated remaining duration and the preliminary remaining duration exceeds the preset threshold, perform smoothing processing on the speed change through the Kalman filter to judge the optimized duration output; After obtaining the optimized duration output, for the real-time change of the vehicle position, use the interpolation method to calculate the continuous duration adjustment value to determine the final remaining duration A; Based on the final remaining duration and combined with the dynamic change of the current speed, update the observation equation of the Kalman filter to obtain a stable duration prediction result; Extract the position update data from the stable duration prediction result, and output the accurate remaining duration value through iterative calculation of distance estimation.

5. The method according to claim 1, wherein The acquisition of real-time traffic flow data and the vehicle acceleration change trend, and the correction of the initial duration prediction value through the dynamic adaptation module to obtain the optimized duration value, including: Obtain real-time data from road nodes through the sensor network to obtain the traffic flow distribution; Extract the vehicle acceleration information according to the traffic flow distribution to determine the acceleration sequence; Adopt the time series analysis method to process the acceleration sequence to obtain the change trend curve; Analyze the change trend curve through the preset dynamic adaptation module to judge the correction coefficient; If the correction coefficient exceeds the preset threshold, make adjustments in combination with the initial duration data to obtain the adjusted duration value; Adopt the linear regression algorithm to fit the adjusted duration value and the prediction value to obtain the optimized duration result; Determine the final duration value through module processing of the optimized duration result.

6. The method according to claim 1, wherein When the optimized duration value reaches the preset trigger timing threshold range, send a pre-activation instruction to the smart home device through the Internet of Things protocol to determine the device preparation status, including: Monitor the optimized duration value to judge whether it reaches the threshold range corresponding to the trigger timing to obtain the judgment result; If the judgment result shows that the threshold range is reached, generate a pre-activation instruction through the Internet of Things protocol to determine the instruction content; Adopt the protocol transmission module to process the pre-activation instruction, send the instruction to the smart home device, and obtain the sending status; According to the sending status, confirm whether the instruction is successfully transmitted to the smart home to judge the device reception situation; Extract the device preparation information from the data returned by the device to determine the device status; Classify the device status to obtain the status classification result; Adjust the subsequent instruction sending frequency according to the status classification result to determine the optimized sending strategy.

7. The method according to claim 1, characterized in that Receive a feedback signal from the smart home device, determine whether the device has completed pre-activation. If so, adjust the final trigger time according to the optimized duration value and output a control instruction sequence, including: Obtain a feedback signal from the smart home device, and determine whether the device has completed pre-activation through a signal processing module to obtain a determination result; If the determination result shows that the pre-activation is completed, process the optimized duration value using a time calculation module to determine the adjusted trigger time; Generate a control instruction according to the adjusted trigger time, and determine the instruction content through an instruction encoding module; Process the control instruction using a transmission protocol, send the instruction to the smart home device, and obtain the sending status; Judge whether the instruction has been successfully transmitted to the device through the sending status to determine the device reception situation; Extract the device feedback information from the data returned by the device, and judge the device status through an information parsing module to obtain the status classification result; Adjust the subsequent instruction generation frequency according to the status classification result, and determine the optimized sending strategy through a frequency optimization module.

8. The method according to claim 1, characterized in that Regarding the control instruction sequence, use a time series analysis algorithm to evaluate the fluctuation range of the trigger timing to obtain a trigger stability index, including: Process the instruction sequence through a time series analysis algorithm to obtain the fluctuation range data; Calculate the change trend of the trigger timing according to the fluctuation range data to determine the preliminary stability evaluation value; If the change trend exceeds the preset threshold, use the sliding window method to adjust the fluctuation range to obtain the optimized trigger timing distribution; Calculate the stability index through the optimized trigger timing distribution to obtain the timing characteristics of the index; Perform Fourier transform processing on the timing characteristics to obtain the fluctuation components in the frequency domain; Judge the trigger stability according to the fluctuation components in the frequency domain to obtain the final stability index value; Extract the key features from the final stability index value to determine the adjustment basis of the control instruction.

9. The method according to claim 1, characterized in that If the trigger stability index is lower than the preset threshold of 0.9, update the neural network model parameters through a feedback loop to optimize the accuracy of the judgment on the way home, and output the adjusted state judgment result, including: If the stability index is lower than the preset threshold, obtain the real-time state information through sensor data to determine the preliminary state judgment; Extract the deviation data from the preliminary state judgment through a feedback loop to update the neural network model parameters; Recalculate the status on the way home using the updated model parameters to obtain the adjusted status data; If there is a difference between the adjusted status data and the expected value, optimize the model parameters through the gradient descent algorithm and judge the optimized status result; Generate the status change trend on the way home according to the optimized status result and output the change trend data; Adjust the input weights of the neural network through the change trend data to determine the final state judgment result; After obtaining the final state judgment result, verify the result consistency through a preset rule and output the verified state information.

Images