Remote summon pre-launch method, system, and medium based on user behavior prediction

Abstract

The application relates to a remote calling pre-starting method and system based on user behavior prediction and a medium, the method comprising collecting multi-dimensional data related to user behavior and vehicle state during vehicle power-off; using an LSTM neural network model to process the multi-dimensional data to predict whether the user will use the remote calling function at the next moment, wherein the processing of the LSTM neural network model comprises operations on a previous moment cell state (ct-1) and a previous moment hidden state (ht-1) to analyze a sequenced user behavior mode; when the prediction indicates that the remote calling function will be used, selectively activating vehicle subsystems required by the remote calling function in a low-power mode; and when the prediction indicates that the remote calling function will not be used, keeping the vehicle in a power-off state. The application realizes fast response of the remote calling function under the premise of reasonable consumption of electric quantity, and effectively solves the double problems of electric quantity consumption and starting time in the traditional scheme.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving, and in particular to a remote summoning pre-start method, system, and medium based on user behavior prediction. Background Technology

[0002] In vehicle remote summoning applications, existing technologies primarily employ two activation timing schemes. The first scheme keeps the vehicle powered on, enabling rapid activation of the remote summoning function and ensuring immediate response. However, continuous power supply leads to excessive battery drain, impacting battery efficiency and range. The second scheme uses a normal vehicle power-off mode, where the vehicle only powers on and activates the remote summoning control module when the user activates it. While this controls power consumption, it results in a lengthy startup time, increasing user waiting time and degrading the user experience.

[0003] Existing remote summoning solutions face a dual challenge: vehicle battery consumption and startup time. Traditional solutions cannot achieve rapid response for remote summoning while ensuring reasonable battery consumption, and lack intelligent pre-start strategies based on user behavior prediction. Furthermore, existing technologies lack the ability to deeply analyze and predict user behavior patterns, and cannot comprehensively model user history and current environmental conditions, making it difficult to determine in advance whether a user will activate the remote summoning function on the timeline between vehicle power-off and power-on. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a remote summoning pre-start method, system and medium based on user behavior prediction, which can achieve rapid response of remote summoning function while ensuring reasonable power consumption, and effectively solve the dual problems of power consumption and start-up time in traditional solutions.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] In a first aspect, the present invention provides a remote summoning pre-launch method based on user behavior prediction, which adopts the following technical solution: Collect multi-dimensional data related to user behavior and vehicle status during vehicle power-off; The LSTM neural network model is used to process the multi-dimensional data to predict whether a user will use the remote summoning function in the next moment. The processing of the LSTM neural network model includes operations on the cell state (ct-1) and the hidden state (ht-1) of the previous moment to analyze serialized user behavior patterns. When the prediction indicates that the remote summoning function will be used, selectively activate the vehicle subsystems required for the remote summoning function in a low-power mode; and Keep the vehicle powered down when the prediction indicates that the remote summon function will not be used.

[0007] Furthermore, in the aforementioned remote summoning pre-start method, the multi-dimensional data includes user information, vehicle information, and intelligent driving usage information.

[0008] Furthermore, in the aforementioned remote summoning pre-start method, the processing of the LSTM neural network model includes: The retention rate of historical information is assessed through forget gate operations; The key information that needs to be written into internal memory is determined by the input gate operation; Update cell state to merge retained historical memory and current candidate content; and The confidence level of whether a user will use the remote summoning function in the next moment is generated by output gate operations.

[0009] Furthermore, in the aforementioned remote summoning pre-start method, the vehicle subsystem includes an intelligent driving control module, a power system module, and a communication module.

[0010] Furthermore, the aforementioned remote summoning pre-start method also includes dynamically adjusting the prediction threshold based on remaining battery life, parking environment, and time period.

[0011] Furthermore, the aforementioned remote summoning pre-start method also includes maintaining the low-power mode in an active state within a limited time window.

[0012] Furthermore, in the aforementioned remote summoning pre-start method, when no user-triggered operation is detected within the time window, the vehicle is automatically powered down and returns to standby mode.

[0013] Furthermore, the aforementioned remote summoning pre-start method also includes adopting a conservative strategy to maintain the power-down state when the model confidence is insufficient or the system is abnormal.

[0014] Furthermore, the aforementioned remote summoning pre-start method also includes online updating of the LSTM neural network model based on usage feedback.

[0015] Furthermore, in the aforementioned remote summoning pre-start method, the collection of multi-dimensional data is based on the complete process of a vehicle going from power-off to power-on in a single instance as an observation interval.

[0016] Secondly, the present invention provides a remote summoning pre-launch system based on user behavior prediction, which adopts the following technical solution: The data acquisition and modeling module is used to collect multi-dimensional data related to user behavior and vehicle status during the period when the vehicle is powered off. A remote summoning pre-launch prediction model is used to process the multi-dimensional data using an LSTM neural network to predict whether a user will use the remote summoning function in the next moment. The LSTM neural network processing includes operations on the previous moment's cell state (ct-1) and the previous moment's hidden state (ht-1) to analyze serialized user behavior patterns. The pre-start decision and power management module is used to selectively activate the vehicle subsystems required for the remote summoning function in a low-power mode when the prediction indicates that the remote summoning function will be used, and to keep the vehicle powered down when the prediction indicates that the remote summoning function will not be used.

[0017] Furthermore, the aforementioned remote summoning pre-start system also includes: The rollback and adaptive strategy module is used to maintain the power-off state by adopting a conservative strategy when the model confidence is insufficient or the system is abnormal. An online update module is used to update the remote summoning pre-launch prediction model online based on usage feedback.

[0018] Thirdly, the present invention provides a readable storage medium, which adopts the following technical solution: A readable storage medium storing computer instructions that, when executed by a processor, implement the method as described in any one of the first aspects above.

[0019] In summary, compared with the prior art, the present invention has at least one of the following beneficial technical effects: The remote summoning pre-start method based on user behavior prediction provided by this invention uses an LSTM neural network model to analyze multi-dimensional data to predict the likelihood of a user using the remote summoning function. When the prediction indicates that the user will use the function, the relevant vehicle subsystems are selectively activated in a low-power mode, thereby significantly shortening the startup time of the remote summoning function while maintaining reasonable power consumption. This effectively solves the contradiction between power consumption and response speed in traditional solutions, improves user experience, and optimizes vehicle energy consumption management. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a specific embodiment of a remote summoning pre-start method based on user behavior prediction according to the present invention.

[0022] Figure 2 This is an LSTM neural network architecture in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Furthermore, it should be understood that the specific embodiments described herein are only for illustration and explanation of this application and are not intended to limit this application.

[0024] It should be noted that the order of description of the following embodiments is not intended to limit the preferred order of the embodiments of this application. Furthermore, the descriptions of each embodiment in the following embodiments have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0025] The method steps described in this embodiment of the invention can be executed in the order described in the specific implementation, or the execution order of each step can be adjusted according to actual needs, provided that the technical problem can be solved. These are not listed one by one here.

[0026] Reference Figure 1 A remote summoning pre-start method based on user behavior prediction is implemented starting from the vehicle's power-off state. This method includes collecting multi-dimensional data related to user behavior and vehicle status during the vehicle's power-off period. The collection of multi-dimensional data covers information such as user usage patterns, vehicle operating parameters, and historical usage of intelligent driving functions.

[0027] Preprocessing operations are performed on the collected multi-dimensional data, including robust handling and normalization of outlier and missing data, so that the data can be directly used for inference by subsequent time series models. The preprocessed data is organized into a serialized format for analysis by the LSTM neural network model.

[0028] This method uses an LSTM neural network model to process multi-dimensional data to predict whether a user will use the remote summoning function in the next moment. The LSTM neural network model's processing includes processing the cell state from the previous moment (c...). t-1 ) and the hidden state of the previous moment (h t-1 The operation is used to analyze serialized user behavior patterns. The LSTM neural network model identifies potential patterns and regularities in users' use of the remote summoning function by analyzing the temporal dependencies in historical behavior sequences.

[0029] Reference Figure 1The remote summoning function prediction module receives processed multi-dimensional data and outputs a prediction of whether a user will use the remote summoning function. The prediction module generates a confidence score based on the calculation results of an LSTM neural network, which reflects the probability that the user will activate the remote summoning function in the next moment.

[0030] When a forecast indicates that the remote summon function will be used, the vehicle subsystems required for the remote summon function are selectively activated in a low-power mode. Activation of the low-power mode avoids fully powering the entire vehicle system, instead activating only the subsystems directly related to the remote summon function, thereby controlling power consumption while ensuring the availability of the function.

[0031] When a prediction indicates that the remote summon function will not be used, the vehicle remains powered down. In this case, the system does not perform any pre-start operation, and the vehicle continues to maintain a low-power standby mode to minimize power consumption.

[0032] This method achieves a balance between power consumption control and response speed optimization through predictive power management strategies, thereby improving the overall user experience of remote summoning functionality.

[0033] The multi-dimensional data includes data from three dimensions: user information, vehicle information, and intelligent driving usage information. The user information dimension covers basic attributes and behavioral characteristics related to vehicle users. The vehicle information dimension includes the vehicle's own status parameters and operational data. The intelligent driving usage information dimension records the historical usage and operating modes of intelligent driving functions.

[0034] User information includes specific data fields such as user ID, age, city, gender, and mobile phone model. The user ID serves as a unique identifier to distinguish different vehicle users. Age information reflects the user's age group, and city information identifies the user's geographical location. Gender information records the user's gender attribute. Mobile phone model information indicates the type of mobile device the user is using, which is related to the remote summoning function's interface and response characteristics.

[0035] Vehicle information includes detailed parameters such as VIN code, purchase date, vehicle model, total mileage, mileage before the last power-off, remaining range, charging status, parking location, underground parking garage status, multi-level underground parking garage status, parking space status, vehicle power mode, current power-off time, and the interval between a single power-off and power-on. The VIN code serves as a unique identifier for each vehicle. Purchase date and vehicle model provide basic vehicle information. Total mileage and mileage before the last power-off reflect vehicle usage intensity. Remaining range and charging status display the vehicle's battery level. Parking location, underground parking garage status, multi-level underground parking garage status, and parking space status describe the vehicle's parking environment. Vehicle power mode, current power-off time, and the interval between a single power-off and power-on record the vehicle's power management status.

[0036] The intelligent driving usage information includes data fields such as intelligent driving usage frequency, historical remote summon usage count, remote summon activation frequency, historical remote summon success rate, and whether valet parking was used before power-off. Intelligent driving usage frequency reflects the user's overall usage of the intelligent driving function. Historical remote summon usage count records the cumulative number of times the user has used the remote summon function. Remote summon activation frequency indicates how frequently the user initiates the remote summon function. Historical remote summon success rate statistics show the percentage of successful remote summon executions. Information on whether valet parking was used before power-off records the parking method before the vehicle was powered off; this information is strongly correlated with subsequent remote summon usage.

[0037] Reference Figure 2 The processing of the LSTM neural network model includes processing the hidden state from the previous time step (h t-1 ) and the cell state at the previous moment (c t-1 The LSTM neural network model uses coordinated operations of the forget gate, input gate, cell state update, and output gate to process time-series data and generate prediction results.

[0038] LSTM neural network models evaluate the retention rate of historical information through a forget gate operation. The forget gate receives the input vector x at the current time step. t and the hidden state h from the previous moment t-1 This calculates the degree of retention of historical memory information. The formula for calculating the forgetting gate is: Where σ represents the sigmoid activation function, W f The weight parameter b represents the weight of the forget gate. f This represents the bias parameter of the forget gate. The output value of the forget gate, ranging from 0 to 1, determines the cell state c at the previous time step. t-1 The percentage of information retained.

[0039] The LSTM neural network model determines the key information from the current input to be written into its internal memory through input gate operations. The input gate consists of two components: the input gate activation function and the candidate value generation function. The formula for calculating the input gate activation function is as follows: W i b represents the weight parameters of the input gate. i This represents the bias parameters of the input gate. The formula for calculating the candidate value generation function is: W c b represents the weight parameter of the candidate value. c represents the bias parameter of the candidate value, and tanh represents the hyperbolic tangent activation function.

[0040] The LSTM neural network model updates the cell state to incorporate retained historical memories and current candidate content. The cell state update operation fuses the historical information processed by the forget gate with the current information filtered by the input gate. The formula for calculating the cell state update is as follows: , where f t Indicates the output of the forget gate, C t-1 This indicates the cell state at the previous moment. Let i represent the candidate value at the current time. t This represents the output of the input gate. The updated cell state C. t It contains filtered historical information and current input information.

[0041] The LSTM neural network model generates the confidence level of whether a user will use the remote summoning function in the next time step through output gate operations. The output gate controls the degree of information output from the cell state and determines the content of the hidden state at the current time step. The formula for calculating the output gate is as follows: W O b represents the weight parameter of the output gate. O This represents the bias parameter of the output gate. The formula for calculating the hidden state at the current time step is: O t Indicates the output of the output gate, C t This represents the cell state at the current moment. Hidden state h t The subsequent processing layer converts the data into a confidence probability value for the user to use the remote summoning function.

[0042] The gate operations of the LSTM neural network model are optimized through training of weight and bias parameters to learn the temporal dependencies and pattern features in user behavior sequences. When processing serialized input data, the model uses the previous hidden state h... t-1 The cell state c at the previous moment t-1 As a memory carrier, it transmits information from historical moments to the current moment in the calculation process, enabling the capture and analysis of user behavior time-series patterns.

[0043] The vehicle subsystem includes an intelligent driving control module, a powertrain module, and a communication module. The intelligent driving control module is responsible for path planning, obstacle detection, and vehicle control commands for the remote summoning function. The powertrain module manages the power supply and control signals for the vehicle's drive motors, braking system, and steering system. The communication module handles data transmission and command reception between the vehicle and user equipment.

[0044] The activation method of the intelligent driving control module in low-power mode includes starting the basic functions of the core processing unit and sensor interfaces. In low-power mode, the intelligent driving control module maintains the minimum operating state of the perception system, including the standby state of the camera interface, radar interface, and positioning module. The low-power mode of the intelligent driving control module maintains the basic computing power of the path planning algorithm, ensuring that the full autonomous driving function can be quickly activated upon receiving a remote summoning command.

[0045] The powertrain module's activation in low-power mode includes pre-charging the high-voltage system and initializing the motor controller. In low-power mode, the powertrain module preheats the drive motor's inverter, reducing the transition time from a standby state to an operating state. The powertrain module's low-power mode keeps the braking system's hydraulic pump in standby mode, ensuring the vehicle can immediately respond to braking demands upon receiving a remote call command. The powertrain module maintains basic power supply to the electric power steering system in low-power mode.

[0046] The communication module's activation methods in low-power mode include maintaining a connection with the cloud server and listening for user commands. In low-power mode, the communication module maintains minimal bandwidth usage for 4G or 5G network connections, ensuring the ability to receive remote call requests from user mobile devices. The communication module maintains scanning status for in-vehicle Wi-Fi and Bluetooth functions in low-power mode to detect proximity signals from user devices. In low-power mode, the communication module periodically sends vehicle status information to the cloud server, maintaining the communication link for remote monitoring and control.

[0047] The tiered loading method completes the initialization and handshake process between the critical control plane and the basic communication link. It activates each subsystem in an orderly manner according to functional priority and startup time requirements. The communication module, as the first-level loading component, establishes a connection with the external network first during the pre-startup phase, ensuring the ability to receive and respond to remote user commands. The intelligent driving control module, as the second-level loading component, activates the foundational module for perception and decision-making functions after the communication link is established. The powertrain module, as the third-level loading component, performs pre-charging and initialization of the motors and actuators after the control system is ready.

[0048] The initialization process of the key control surfaces includes establishing internal communication protocols and data exchange interfaces between the subsystems. A control command transmission channel is established between the intelligent driving control module and the powertrain module to ensure that path planning results are accurately transmitted to the actuators. A status feedback channel is established between the powertrain module and the communication module to upload vehicle operating status information to the cloud server and user equipment in real time. Each subsystem executes a self-test procedure during initialization to verify the normal operation of hardware functions and software modules.

[0049] The handshake process of the basic communication link includes verifying the connection status and data transmission capabilities of each communication interface. The communication module performs authentication and encryption key exchange with the cloud server to establish a secure data transmission channel. Each subsystem within the vehicle performs a handshake confirmation via the CAN bus or Ethernet interface to verify the communication path for command transmission and status feedback. The handshake process includes testing communication latency and data integrity to ensure that the command response time and control accuracy during the remote summoning function execution meet safety requirements.

[0050] The method also includes dynamically adjusting the prediction threshold based on remaining range, parking environment, and time period. The dynamic threshold adjustment mechanism adaptively modifies the confidence criterion of the LSTM neural network model output by analyzing the vehicle's current state and environmental conditions, thereby achieving a balance between power consumption and response speed.

[0051] Remaining driving range is a primary factor in dynamic threshold adjustment, directly impacting the prediction threshold setting. When the vehicle has sufficient remaining driving range, the system lowers the prediction threshold, increasing the probability of triggering pre-start operations to prioritize the rapid response performance of the remote summon function. When the vehicle has low remaining driving range, the system raises the prediction threshold, reducing unnecessary pre-start operations and prioritizing power consumption to extend the vehicle's standby time.

[0052] In some implementations, when the remaining driving range exceeds 200 kilometers, the prediction threshold is set to 0.6, allowing the system to perform pre-start operations under medium confidence levels. When the remaining driving range is in the range of 100-200 kilometers, the prediction threshold is adjusted to 0.7, increasing the triggering condition for pre-start operations. When the remaining driving range is less than 100 kilometers, the prediction threshold is set to 0.8, and pre-start operations are only performed under high-confidence prediction results.

[0053] The strategy for adjusting the prediction threshold based on the parking environment considers the usage requirements and activation difficulty of the remote summoning function in different parking scenarios. In a surface parking lot environment, vehicle communication signals are good and activation conditions are relatively simple, so the system uses a standard prediction threshold for judgment. In an underground parking garage environment, vehicles face challenges such as communication signal attenuation and a complex activation environment, so the system lowers the prediction threshold to increase the execution frequency of pre-start operations.

[0054] In some implementations, the prediction threshold in underground parking environments is reduced by 0.1 compared to the standard threshold. For single-level underground parking, the prediction threshold is reduced by 0.1 from the standard value to compensate for the impact of communication delays and signal attenuation on remote call response time. For multi-level underground parking environments, the prediction threshold is reduced by 0.15 from the standard value to further increase the trigger probability of pre-start operations, addressing more complex start conditions and communication environments.

[0055] The threshold adjustment strategies differ between parking in a parking space and parking in an open area. When parking in a parking space, vehicles need to perform more complex path planning and obstacle avoidance operations, so the system lowers the prediction threshold by 0.05 to increase the probability of pre-starting operations. When parking in an open area, the space conditions for vehicle starting and movement are better, so the system maintains the standard prediction threshold or appropriately increases the threshold by 0.05 to balance response speed and power consumption.

[0056] The adjustment of the prediction threshold based on time period factors reflects the behavioral patterns of users using the remote summoning function at different times. Weekday mornings from 7-9 AM and evenings from 5-7 PM are high-frequency usage periods for the remote summoning function; during these periods, the system lowers the prediction threshold by 0.1 to increase the trigger sensitivity of pre-start operations. Weekday noon from 12-2 PM and all day on weekends are medium-frequency usage periods, and the system uses the standard prediction threshold for determination.

[0057] In some implementations, the period from 10 PM to 6 AM the following day is considered a low-frequency usage period for the remote summoning function. The system increases the prediction threshold by 0.15, reduces the frequency of pre-start operations, and prioritizes controlling standby power consumption at night. During holidays, the system adjusts the threshold strategy for different time periods based on historical usage data to adapt to changes in user travel patterns.

[0058] The comprehensive threshold adjustment strategy weights and calculates multiple factors, including remaining range, parking environment, and time period, to generate the final dynamic prediction threshold. The base prediction threshold is set at 0.7, and the remaining range, parking environment, and time period factors each adjust the base threshold. The final prediction threshold is calculated by adding the weighted adjustment value of each factor to the base threshold. The adjustment value is limited to a range of -0.2 to +0.2 to ensure that the prediction threshold varies within a reasonable range.

[0059] The dynamic threshold adjustment mechanism includes two operating modes: real-time monitoring and periodic updates. In real-time monitoring mode, the system continuously tracks changes in remaining driving range and immediately updates the predicted threshold when the remaining range crosses the preset interval boundary. In periodic update mode, the system reassesses parking environment and time period factors every 30 minutes and adjusts the corresponding threshold parameters. The two operating modes work together to ensure that the predicted threshold can respond promptly to changes in vehicle status and environmental conditions.

[0060] The method also includes maintaining the low-power mode in an active state within a limited time window. The time window is set based on statistical analysis of user response time and consideration of power consumption balance to ensure that the pre-startup operation remains effective within a reasonable time range. The length of the time window is dynamically adjusted according to different application scenarios and user behavior patterns.

[0061] The time window setting is based on the confidence level of historical data analysis and user behavior prediction. When the confidence level of the LSTM neural network model output is high, the system sets a longer time window, allowing the low-power mode to be maintained for a longer period to accommodate user operation latency. When the confidence level is at a medium level, the system uses a standard time window length to balance response preparation time and power consumption. When the confidence level is close to the prediction threshold boundary, the system shortens the time window to reduce the duration of the pre-startup state.

[0062] In some implementations, the time window for high-confidence predictions is set to 15 minutes to provide users with ample preparation time. The time window for medium-confidence predictions is set to 10 minutes, striking a balance between responsiveness and power consumption. The time window for low-confidence predictions is set to 5 minutes to minimize power loss due to uncertain predictions.

[0063] The time window maintenance mechanism consists of two components: countdown monitoring and status detection. The countdown monitoring module starts the countdown function from the beginning of the pre-start operation and continuously tracks the remaining duration of the time window. The status detection module monitors user operation signals, vehicle status changes, and external trigger events during the time window to determine whether the time window needs to be ended early or extended.

[0064] The countdown monitoring module uses a high-precision clock source for time measurement to ensure the accuracy of time window control. It updates the remaining time display every 30 seconds, providing time window status information to other modules in the system. When the remaining time approaches a preset warning threshold, the countdown monitoring module sends a warning signal to the power management module indicating that it is preparing to power off.

[0065] The status detection module monitors events that may affect the time window through various sensors and communication interfaces. It monitors application activity on the user's mobile device, detecting preparations for the remote summon function. The module tracks changes in the vehicle's surrounding environment, including the approach of other vehicles or pedestrians. It receives instruction updates and user account activity logs from the cloud server.

[0066] In the low-power ready state, the system operates using a hierarchical management architecture, classifying and managing each subsystem according to functional priority and power consumption level. The first layer includes the communication module and basic monitoring functions, maintaining minimum power consumption and basic responsiveness. The second layer includes the core processing unit of the intelligent driving control module, maintaining moderate computing power and perception capabilities. The third layer includes the pre-charging function of the powertrain module, preparing for rapid startup of the drive and execution systems.

[0067] In the first-level operating mode, the communication module maintains a heartbeat connection with the cloud server, sending a status acknowledgment signal every 2 minutes. The communication module maintains the minimum data transmission rate of the 4G or 5G network connection to ensure the ability to receive user commands and system updates. Basic monitoring functions include battery level detection, vehicle location updates, and safety status monitoring; these functions operate in a low-frequency mode to control power consumption.

[0068] In the second-level operating mode, the core processing unit of the intelligent driving control module maintains 25% of its computing performance, preserving the basic computational capabilities for path planning and decision-making algorithms. The sensor interfaces of the intelligent driving control module are in standby mode, and the camera and radar systems maintain a minimum power consumption monitoring mode. The storage system of the intelligent driving control module preloads map data and navigation information, reducing data read time during full startup.

[0069] In the third-level operating mode, the power system module pre-charges the high-voltage battery and inverter to maintain the voltage level in a start-up ready state. The motor controller of the power system module keeps the temperature control system operating at low power to prevent critical components from becoming too cold and affecting starting performance. The braking system of the power system module maintains the intermittent operation mode of the hydraulic pump to ensure timely braking response.

[0070] The resource management strategy achieves coordinated operation of each subsystem through dynamic power allocation and load balancing. The strategy adjusts the power allocation ratio of each level based on the remaining power level. When power is sufficient, it increases the power supply to the second and third levels; when power is insufficient, it prioritizes ensuring the basic functions of the first level.

[0071] The dynamic power allocation mechanism monitors the real-time power consumption of each subsystem. When the power consumption of a subsystem exceeds a preset limit, it automatically reduces the operating power of that subsystem. The dynamic power allocation mechanism adjusts the power allocation strategy according to the remaining duration of the time window, gradually reducing the power supply of each level in the later part of the time window to prepare for the upcoming power-down operation.

[0072] The load balancing strategy distributes computational and communication tasks across time, preventing multiple subsystems from reaching peak power consumption simultaneously. It prioritizes tasks directly related to remote summoning functions, delaying non-critical tasks until the end of the time window or after full startup. Through task queue management and priority scheduling, the load balancing strategy ensures system responsiveness and stability in a low-power ready state.

[0073] In some implementations, the resource management strategy includes a battery level warning and an emergency power-down mechanism. When the vehicle's remaining battery level falls below 5%, the resource management strategy automatically shortens the time window to 2 minutes and reduces the power supply level of all subsystems. When the vehicle's remaining battery level falls below 2%, the resource management strategy immediately terminates the low-power ready state and forces an emergency power-down operation to protect the battery system.

[0074] System status monitoring within the time window includes power tracking, temperature monitoring, and communication quality assessment. The power tracking module records the actual power consumption of each subsystem in low-power ready mode and compares it with a preset power consumption model. The temperature monitoring module detects the operating temperature of key components to ensure effective thermal management in low-power operation mode. The communication quality assessment module tests the stability of network connections and data transmission latency, verifying the reliability of remote call command reception.

[0075] If no user-triggered operation is detected within the time window, the system automatically powers down the vehicle and returns it to standby mode. The automatic power-down mechanism continuously monitors user operation signals and system state changes, executing an orderly power-off process at the end of the preset time window to ensure the vehicle safely returns to low-power standby mode.

[0076] User-triggered operation detection involves comprehensive monitoring of multiple signal sources. The system monitors the launch signal of the remote summon application on the user's mobile device and detects the user's operation command of clicking the remote summon function button. The system tracks changes in the Bluetooth or Wi-Fi connection status between the user's device and the vehicle, and identifies user behavior patterns when approaching the vehicle. The system receives user account activity records from the cloud server, including login status, application usage, and records of remote control command transmissions.

[0077] In some implementations, the user-triggered operation detection mechanism employs a multi-layered verification strategy. The first layer of detection includes the mobile application's interface operations; the system identifies the user's sequence of actions: opening the remote summoning function interface, selecting the summoning target location, and confirming the summoning command. The second layer of detection includes device connection status monitoring; the system detects signals indicating a direct communication connection between the user's device and the vehicle. The third layer of detection includes location information analysis; the system uses GPS positioning to determine whether the user is moving towards the vehicle's location.

[0078] The system continuously monitors command signals from user devices via a communication module. The communication module scans Bluetooth and Wi-Fi signals every 10 seconds to detect connection requests from paired user devices. It maintains real-time data synchronization with the cloud server, receiving updates on user account operation status and remote control commands. The communication module records all received signal types, timestamps, and signal strengths to provide data support for triggering operation decisions.

[0079] The intelligent driving control module participates in the user operation detection process, recognizing user approach behavior through environmental perception of the vehicle's surroundings. The module's camera system intermittently acquires images in low-power mode to detect human activity around the vehicle. The radar system monitors moving objects around the vehicle, identifying the trajectories of people approaching it. The intelligent driving control module correlates and analyzes environmental perception information with user device signals to improve the accuracy of triggered operation detection.

[0080] The timing precision control of trigger operation detection ensures that the system can respond to user commands in a timely manner. The system adopts an event-driven detection mechanism, immediately verifying and confirming any possible user operation signal received. The system sets a detection response time limit of 3 seconds to ensure that user operation signals can be identified and processed by the system before the time window ends. The system records the response time and processing result of each detection operation to optimize the performance of the detection algorithm.

[0081] When no valid user-triggered operation is detected within the time window, the system initiates an automatic power-down process. The automatic power-down process executes according to a predefined shutdown sequence, ensuring that each subsystem stops operating safely and orderly. The execution order of the power-down process is the reverse of the hierarchical loading order during pre-startup, prioritizing the shutdown of subsystems with higher power consumption, and finally shutting down basic monitoring and communication functions.

[0082] The first stage of the automatic power-down process includes shutting down the power system module. The power system module stops the pre-charging process of the high-voltage battery and switches the inverter and motor controller to a completely shut-off state. The power system module shuts down the intermittent operation mode of the braking system hydraulic pump, restoring the braking system to a stationary state. The power system module's temperature control system stops heating critical components, allowing the component temperatures to naturally decrease to ambient temperature.

[0083] The second phase of the automatic power-down process includes shutting down the intelligent driving control module. The intelligent driving control module stops the core processing unit's computational tasks and clears the temporarily stored path planning and decision data. It also disables the sensor interfaces in standby mode and switches the camera and radar systems to a completely off state. Finally, the intelligent driving control module saves the current system state and configuration parameters to non-volatile memory in preparation for the next startup.

[0084] The third stage of the automatic power-down process includes shutting down the communication module. Before shutting down, the communication module sends vehicle status update information to the cloud server, notifying the server that the vehicle is about to enter a complete power-down state. The communication module disconnects from the user's Bluetooth and Wi-Fi connections, and stops scanning and monitoring the local network. The communication module disables 4G or 5G network connections, retaining only the most basic Wake-on-LAN function for receiving emergency commands.

[0085] In some implementations, the automatic power-down process includes system state saving and recovery preparation operations. The system saves the current predictive model parameters, user behavior data, and environmental state information to the vehicle's permanent storage device. The system records the execution time, duration, and termination reason of this pre-start operation for subsequent model optimization and threshold adjustment. The system cleans up temporary files and cached data, freeing up storage space and optimizing performance for the next startup.

[0086] The orderly power-down process includes status verification and safety checks for each subsystem. Each subsystem performs a self-test procedure before shutdown to verify the integrity and safety of the shutdown operation. Subsystems report their shutdown status and any abnormalities to the main control module, ensuring the traceability of the power-down process. The main control module collects shutdown confirmation signals from all subsystems to verify the successful execution of the entire power-down procedure.

[0087] The timing control of the power-down process ensures that each stage is executed at predetermined time intervals. The first stage, the power system shutdown operation, takes 30 seconds to allow for safe discharge of the high-voltage system and a smooth shutdown of the temperature control system. The second stage, the intelligent driving control module shutdown operation, takes 20 seconds to ensure data preservation and complete shutdown of the sensor system. The third stage, the communication module shutdown operation, takes 10 seconds to complete the disconnection from external systems and status reporting.

[0088] Once the vehicle returns to standby mode, it maintains minimal power consumption and basic functional operation. In standby mode, the vehicle only maintains the operation of the battery management system, safety monitoring system, and remote wake-up receiver. The battery management system monitors battery charge, temperature, and health status to prevent over-discharge or abnormal conditions. The safety monitoring system includes basic safety functions such as anti-theft alarm, door lock status detection, and vehicle tilt detection.

[0089] The remote wake-up receiver maintains its ability to listen for specific wake-up signals while in standby mode. Employing an ultra-low power design, it consumes only microwatts of power to maintain basic communication functions. The remote wake-up receiver can receive direct wake-up commands from user devices and remote wake-up signals from cloud servers. Upon receiving a valid wake-up signal, the remote wake-up receiver initiates the vehicle's complete power-on process, restoring all system functions.

[0090] Power consumption management in standby mode employs a time-sharing power supply and intermittent operation strategy. The battery management system performs a power check and status update every 5 minutes, remaining in sleep mode the rest of the time. The security monitoring system uses an event-triggered mode, activating corresponding processing functions only when an anomaly is detected. The remote wake-up receiver uses a periodic scanning mode, checking the wake-up signal every 30 seconds to balance responsiveness and power consumption control.

[0091] In some implementations, the standby state includes periodic system health checks and maintenance operations. The system performs a complete self-test every 24 hours to verify the functional status of critical components and the integrity of electrical connections. The system periodically updates timestamps and location information to ensure the accuracy of vehicle status data. While in standby mode, the system receives remote push notifications of software updates and configuration parameters to prepare for functional improvements upon the next startup.

[0092] The automatic power-down mechanism includes abnormal situation handling and emergency protection functions. When an abnormal drop in battery power or an abnormal rise in system temperature is detected, the automatic power-down mechanism immediately executes the emergency shutdown procedure, skipping the normal phased shutdown process. When a communication system malfunctions or external interference occurs, the automatic power-down mechanism adopts a local decision-making mode, performing a shutdown operation based on preset rules. The automatic power-down mechanism records all abnormal situations and processing results, providing data support for system maintenance and fault diagnosis.

[0093] The method also includes employing a conservative strategy to maintain the vehicle's power-off state when model confidence is insufficient or the system is abnormal. This conservative strategy evaluates the reliability of the LSTM neural network model's output and the stability of the system's operating state through a multi-layered decision-making mechanism. When uncertainties or abnormal conditions are detected, the system prioritizes maintaining the vehicle's power-off state to avoid unnecessary power consumption and potential risks.

[0094] The judgment criterion for insufficient model confidence is based on the probability value output by the LSTM neural network and a preset confidence threshold range. When the confidence value of the model output is lower than the minimum acceptable threshold, the system determines that the prediction result lacks the reliability to perform pre-startup operations. In some implementations, the minimum confidence threshold is set to 0.5; when the model output confidence is lower than this value, the system automatically adopts a conservative strategy. When the model output confidence is between 0.5 and 0.6, the system makes a comprehensive judgment based on other environmental factors, including the remaining battery level, the complexity of the parking environment, and historical prediction accuracy.

[0095] System anomaly detection covers a variety of abnormal situations, including hardware failures, software errors, and communication interruptions. Hardware failure detection includes abnormal sensor data, power supply voltage fluctuations, and temperatures exceeding normal ranges. Software error detection includes algorithm calculation errors, data processing errors, and memory overflows. Communication interruption detection includes network connection failures, data transmission delays, and insufficient signal strength.

[0096] The conservative strategy's execution mechanism includes two operations: immediately stopping the pre-start process and forcibly maintaining the power-down state. When insufficient model confidence is detected, the system cancels the planned pre-start operation to avoid power consumption based on unreliable prediction results. When a system anomaly is detected, the system immediately terminates the ongoing pre-start process and switches the activated subsystems back to a completely shut-down state.

[0097] A multi-tiered fallback mechanism provides tiered anomaly handling and fault recovery capabilities. The first-tier fallback handles minor system fluctuations and temporary anomalies, restoring normal operation through reassessment and a short waiting period. The second-tier fallback handles moderate system anomalies, including communication quality degradation and partial subsystem malfunctions, with the system operating in a degraded mode to maintain basic functionality. The third-tier fallback handles severe system failures and security risks, executing emergency shutdown procedures and logging detailed fault information.

[0098] Automatic cancellation of pre-startup operations in case of communication anomalies includes handling network connection interruptions and data transmission errors. When a connection interruption with the cloud server is detected for more than 30 seconds, the system automatically cancels the pre-startup operation to avoid maintaining a low-power ready state when user commands cannot be received. When a data transmission error rate exceeds 10%, the system determines that the communication quality is insufficient to support the reliable execution of the remote summoning function and immediately terminates the pre-startup process.

[0099] Automatic cancellation of pre-start operations in case of power supply abnormalities includes monitoring and handling of voltage fluctuations and current anomalies. When a battery voltage fluctuation exceeding 5% of the rated value is detected, the system immediately stops all pre-start operations to prevent voltage instability from damaging the vehicle's electronic systems. When an abnormal increase in subsystem current consumption is detected, the system automatically disconnects the power supply to the abnormal subsystem and cancels the entire pre-start process to ensure battery safety.

[0100] User overriding functionality provides manual control through explicit local and remote commands. Users can send forced start or forced stop commands via the vehicle's control panel or a mobile application. Upon receiving the explicit command, the system immediately executes the corresponding action, overriding the judgments of automatic prediction and conservative strategies. User overriding functionality includes authentication and operation confirmation mechanisms to ensure that only authorized users can perform overriding operations.

[0101] The method also includes online updates to the LSTM neural network model based on usage feedback. The online update mechanism identifies and labels samples with large prediction deviations by collecting and comparing actual usage data with prediction results, and continuously optimizes the model's prediction accuracy and adaptability through retraining.

[0102] The identification of samples with large prediction bias is based on the assessment of the difference between the actual user behavior and the model's prediction results. When the LSTM neural network model predicts that a user will use the remote summoning function but the user does not perform the relevant operation within a preset time window, the system marks the sample as a false positive bias sample. When the model predicts that a user will not use the remote summoning function but the user subsequently initiates the remote summoning operation immediately, the system marks the sample as a false negative bias sample.

[0103] In some implementations, the system sets a deviation threshold for samples where the difference between the predicted confidence level and the actual result exceeds 0.3. False positive deviation samples typically have a confidence level higher than 0.7 but did not actually use the function, while false negative deviation samples typically have a confidence level lower than 0.4 but actually used the function. The system provides detailed annotations for these deviation samples, recording the environmental conditions, user status, and vehicle parameters at the time of the deviation.

[0104] The annotation process includes feature analysis and cause classification of deviation samples. The system analyzes user information, vehicle information, and intelligent driving usage information in the deviation samples to identify key feature combinations that lead to prediction errors. The system categorizes the causes of deviations into categories such as changes in user behavior patterns, abnormal environmental conditions, special vehicle status conditions, and model parameter mismatches, providing targeted guidance for subsequent model adjustments.

[0105] The retraining process employs an incremental learning method to update the weight parameters of the LSTM neural network model. The system fuses the labeled biased samples with historical training data to form an updated training set. The retraining process adjusts the weight matrices W of the forget gate, input gate, and output gate. f W i W c W O and bias parameter b f b i b c b O This improves the model's ability to identify specific user behavior patterns and environmental conditions.

[0106] The threshold and time window configuration are dynamically adjusted based on prediction accuracy statistics and user satisfaction feedback. The system performs a threshold evaluation after collecting every 100 valid samples. When the false positive rate exceeds 15%, the prediction threshold is increased by 0.05; when the false negative rate exceeds 10%, the prediction threshold is decreased by 0.05. The time window configuration is adjusted based on user response time statistics. When the average response time exceeds 80% of the current time window, the window is extended by 2 minutes.

[0107] The continuous adaptive mechanism achieves long-term predictive capability improvement through regular model performance evaluation and parameter optimization. The system performs a comprehensive model evaluation monthly, calculating performance metrics such as prediction accuracy, recall, and F1 score. Based on the characteristics of different user groups and usage scenarios, the system establishes personalized model parameter configurations to achieve prediction optimization for specific users. The adaptive mechanism includes model version management and rollback functionality, automatically reverting to the previous stable version when the performance of a new model version degrades.

[0108] Multi-dimensional data collection uses the complete process of a vehicle's power-off and power-on cycle as an observation interval. The principle for dividing this observation interval is based on the complete lifecycle of the vehicle's power state and the temporal continuity of user behavior patterns. The observation interval begins recording from the moment the vehicle performs a power-off operation and continues until the moment the vehicle is powered on and restarted, forming a closed data collection time window.

[0109] The data serialization method within the observation interval organizes and encodes user information, vehicle information, and intelligent driving usage information in chronological order. At the moment of power-off, the system records the vehicle's initial state parameters, including remaining driving range, parking location, power mode, and intelligent driving function usage. During power-off, the system periodically collects environmental change data, including parking environment status, charging status changes, and time period information. Approaching the moment of power-on, the system records the event type that triggered power-on and user operation information.

[0110] Data serialization processing samples and standardizes all data points within the observation interval at fixed time intervals. In some implementations, the system performs data sampling every 15 minutes, recording the vehicle status and environmental parameters at the current moment. The system normalizes the sampled data, converting parameters with different dimensions and numerical ranges into a unified numerical range. The system then arranges the processed data in a time series format, forming a vectorized input format.

[0111] The complete data packaging method organizes the serialized data within the observation interval into an input format suitable for LSTM neural network processing. The system packages the data sequence of each observation interval into a training sample, which includes the input feature sequence and corresponding label information. The input feature sequence records multi-dimensional data at each time point within the observation interval, and the label information indicates whether the user initiated the remote summoning function at the end of the observation interval. The system performs batch processing on the packaged samples to form a dataset for time series model inference.

[0112] The boundary definition of the observation interval employs power status monitoring and event triggering mechanisms to ensure the integrity and accuracy of data acquisition. The system monitors the vehicle's power-down and power-up events through the battery management system, automatically marking the start and end times of the observation interval. The system records the duration and number of data collection points for each observation interval, ensuring data format consistency and comparability between different observation intervals.

[0113] This invention also discloses a remote summoning pre-start system.

[0114] A remote summoning pre-start system based on user behavior prediction achieves a balanced optimization of vehicle power consumption and response speed through the collaborative work of multiple functional modules. The system includes a data acquisition and modeling module, a remote summoning pre-starting prediction model, and a pre-starting decision and power management module. These modules exchange information and coordinate functions through standardized data interfaces and control protocols.

[0115] The data acquisition and modeling module collects multi-dimensional data related to user behavior and vehicle status during vehicle power-off periods. It acquires user information, vehicle information, and intelligent driving usage information through the onboard sensor network, communication interfaces, and storage system. The module performs preprocessing operations on the raw data, including data cleaning, outlier detection, and normalization, converting the processed data into a standardized format for use by subsequent modules. The module also maintains historical data storage and management functions, establishing a user behavior database and vehicle status profiles.

[0116] The remote summoning pre-start prediction model uses an LSTM neural network to process multi-dimensional data to predict whether a user will use the remote summoning function in the next moment. The remote summoning pre-start prediction model receives serialized data input from the data acquisition and modeling module. The LSTM neural network's processing includes the cell state (c) from the previous moment. t-1 ) and the hidden state of the previous moment (h t-1 The operation is used to analyze serialized user behavior patterns. The remote summoning pre-start prediction model generates a confidence probability value of a user using the remote summoning function through coordinated calculations of the forget gate, input gate, and output gate, and passes the prediction results to the pre-start decision and power management module.

[0117] The pre-startup decision and power management module is used to selectively activate the vehicle subsystems required for the remote summon function in low-power mode when predictions indicate that the remote summon function will be used, and to keep the vehicle powered down when predictions indicate that the remote summon function will not be used. The pre-startup decision and power management module receives the output of the remote summon pre-startup prediction model and evaluates and determines the prediction confidence level in conjunction with a dynamic threshold adjustment strategy. The pre-startup decision and power management module controls the power state of the intelligent driving control module, powertrain module, and communication module, enabling tiered loading and switching between low-power ready modes.

[0118] The data flow between the system modules adopts an architecture combining unidirectional transmission and bidirectional feedback. The data acquisition and modeling module provides real-time and historical data input to the remote summoning pre-start prediction model, with data transmission using a high-speed internal bus to ensure low-latency delivery. The remote summoning pre-start prediction model outputs prediction results and confidence assessments to the pre-start decision and power management module, with the prediction results including probability values, timestamps, and relevant feature weight information. The pre-start decision and power management module feeds back execution status and result information to the data acquisition and modeling module for model performance evaluation and parameter optimization.

[0119] The control flow between the system's modules is managed uniformly and autonomously through a main controller and distributed control nodes. The main controller is responsible for overall system status monitoring, inter-module coordination, and anomaly handling, while the distributed control nodes are responsible for the function execution and status management within each module. The control flow includes three levels: startup control, operation control, and shutdown control. Startup control manages system initialization and module activation sequences, operation control manages data processing and decision execution under normal operating conditions, and shutdown control manages system shutdown and resource release processes.

[0120] The inter-module collaborative working mechanism employs a scheduling strategy combining event-driven and time-triggered approaches. The data acquisition and modeling module executes data acquisition and processing tasks at fixed time intervals to ensure the continuity and real-time nature of data input. The remote summoning pre-start prediction model adopts an event-driven mode, immediately executing predictive calculations upon receiving new data input, thus improving system response speed. The pre-start decision and power management module combines time-triggered and event-driven mechanisms to periodically assess system status and respond to changes in prediction results, achieving precise control and rapid response in power management.

[0121] The remote summoning pre-startup system also includes a rollback and adaptive strategy module and an online update module. These additional modules work together to enhance the system's stability and adaptability. The rollback and adaptive strategy module is used to maintain the power-down state using a conservative strategy when the model confidence is insufficient or the system is abnormal. By monitoring the output confidence of the LSTM neural network model and the system's operating status in real time, the rollback and adaptive strategy module immediately triggers the conservative strategy execution mechanism when it detects insufficient reliability of the prediction results or hardware / software abnormalities.

[0122] The rollback and adaptive strategy module employs a hierarchical anomaly detection architecture. The first layer detects model confidence anomalies; when the confidence level of the LSTM neural network output falls below a preset threshold, a conservative mode is activated. The second layer detects system operational anomalies, including sensor data anomalies, communication interruptions, and power fluctuations. The third layer detects environmental anomalies, including changes in parking location, external interference, and abnormal temperatures. The rollback and adaptive strategy module selects appropriate conservative strategies based on the anomaly type and severity, including extending the evaluation time, increasing the prediction threshold, and forcibly maintaining the power-down state.

[0123] The online update module is used to update the remote call pre-launch prediction model online based on user feedback. It collects comparison data between actual user behavior and prediction results, identifies samples with large prediction deviations, and classifies and labels them. The online update module uses an incremental learning algorithm to adjust the weight parameters of the LSTM neural network, updating the weight matrices of the forget gate, input gate, and output gate using a mini-batch gradient descent method. The online update module maintains model version management functionality, recording parameter changes and performance improvements for each update.

[0124] The session caching and reuse mechanism enables a rapid switch from a low-power ready state to full-function operation through preloading and state preservation techniques. In the low-power ready state, the system pre-establishes an authentication session with the cloud server, including authentication tokens, encryption keys, and communication protocol parameters. The system caches the intelligent driving control module's path planning algorithm parameters, map data, and sensor calibration information to avoid repeated loading processes during full-function startup.

[0125] Upon receiving a remote summoning command from a user, the system directly reuses pre-initialized sessions and cached data, reducing startup time from the traditional 60-90 seconds to 15-20 seconds. Session reuse includes skipping time-consuming steps such as network connection establishment, authentication, and data synchronization. Cache reuse includes directly loading pre-processed map data, sensor configurations, and control algorithm parameters, reducing computational resource consumption and response latency. The system ensures the reliability and security of the reuse mechanism by periodically refreshing cache content and verifying session validity.

[0126] This invention also discloses a readable storage medium.

[0127] A computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method described in any of the above embodiments. The computer-readable storage medium may include any entity or device capable of carrying a computer program, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc. The computer program includes computer program code. The computer program code may be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable storage medium may include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc.

[0128] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0129] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a system including a processing module or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0130] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote summoning pre-launch method based on user behavior prediction, characterized in that, include: Collect multi-dimensional data related to user behavior and vehicle status during vehicle power-off; The LSTM neural network model is used to process the multi-dimensional data to predict whether a user will use the remote summoning function in the next moment. The processing of the LSTM neural network model includes operations on the cell state (ct-1) and the hidden state (ht-1) of the previous moment to analyze serialized user behavior patterns. When the prediction indicates that the remote summoning function will be used, the vehicle subsystems required for the remote summoning function are selectively activated in a low-power mode; as well as Keep the vehicle powered down when the prediction indicates that the remote summon function will not be used.

2. The remote summoning pre-start method according to claim 1, characterized in that, The multi-dimensional data includes user information, vehicle information, and intelligent driving usage information.

3. The remote summoning pre-start method according to claim 1, characterized in that, The processing of the LSTM neural network model includes: The retention rate of historical information is assessed through forget gate operations; The key information that needs to be written into internal memory is determined by the input gate operation; Update cell state to merge retained historical memory and current candidate content; and The confidence level of whether a user will use the remote summoning function in the next moment is generated by output gate operations.

4. The remote summoning pre-start method according to claim 3, characterized in that, The vehicle subsystem includes an intelligent driving control module, a power system module, and a communication module.

5. The remote summoning pre-start method according to claim 1, characterized in that, It also includes dynamically adjusting the prediction threshold based on remaining range, parking environment, and time period.

6. The remote summoning pre-start method according to claim 1, characterized in that, It also includes maintaining the low-power mode in an active state within a limited time window.

7. The remote summoning pre-start method according to claim 6, characterized in that, If no user-triggered operation is detected within the time window, the vehicle will automatically power down and return to standby mode.

8. The remote summoning pre-start method according to claim 1, characterized in that, It also includes adopting a conservative strategy to maintain the power-off state when the model confidence is insufficient or the system is abnormal.

9. The remote summoning pre-start method according to claim 1, characterized in that, It also includes online updates of the LSTM neural network model based on usage feedback.

10. The remote summoning pre-start method according to any one of claims 1 to 9, characterized in that, The collection of the multi-dimensional data takes the complete process of a vehicle going from power-off to power-on as an observation interval.

11. A remote summoning pre-launch system based on user behavior prediction, characterized in that, The system includes: The data acquisition and modeling module is used to collect multi-dimensional data related to user behavior and vehicle status during the period when the vehicle is powered off. A remote summoning pre-launch prediction model is used to process the multi-dimensional data using an LSTM neural network to predict whether a user will use the remote summoning function in the next moment. The LSTM neural network processing includes operations on the previous moment's cell state (ct-1) and the previous moment's hidden state (ht-1) to analyze serialized user behavior patterns. The pre-start decision and power management module is used to selectively activate the vehicle subsystems required for the remote summoning function in a low-power mode when the prediction indicates that the remote summoning function will be used, and to keep the vehicle powered down when the prediction indicates that the remote summoning function will not be used.

12. The remote summoning pre-start system according to claim 11, characterized in that, The system also includes: The rollback and adaptive strategy module is used to maintain the power-off state by adopting a conservative strategy when the model confidence is insufficient or the system is abnormal. An online update module is used to update the remote summoning pre-launch prediction model online based on usage feedback.

13. A readable storage medium, characterized in that, The readable storage medium stores computer instructions that, when executed by a processor, implement the remote summoning pre-start method as described in any one of claims 1-10.

Images