Automobile start-stop system management method based on intelligent control

By optimizing start-stop strategies through multimodal environmental perception and deep reinforcement learning, the problems of high fuel consumption, short battery life, response delay and non-compliance with emission standards in traditional vehicle start-stop systems have been solved, achieving efficient energy saving and emission reduction and safe and comfortable driving under intelligent control.

CN121084388APending Publication Date: 2025-12-09HUBEI FENGLIN RENEWAL RESOURCE CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202511148677.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-08-18
Publication Date
2025-12-09

AI Technical Summary

Technical Problem

Traditional car start-stop systems lead to increased fuel consumption, shortened battery life, delayed start-stop response, non-compliance with emissions standards, and poor environmental adaptability when frequently started and stopped, affecting driving safety and comfort.

Method used

Employing a multimodal environmental perception system, a deep reinforcement learning decision-making framework, a hierarchical collaborative control execution system, and a real-time health assessment system, combined with a cloud-based collaborative optimization mechanism, the system monitors vehicle status and environment in real time through high-precision sensors and deep learning technology, optimizes start-stop strategies, and achieves intelligent control.

Benefits of technology

It improves start-stop response speed, reduces fuel consumption, extends battery life, reduces emissions, enhances driving safety and comfort, and adapts to various environmental conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121084388A_ABST
    Figure CN121084388A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of automobile power system control, in particular to an automobile start-stop system management method based on intelligent control, which comprises the following steps: S1, a multi-mode environment sensing system; s2, a deep reinforcement learning decision framework; s3, hierarchical cooperative control execution; s4, a real-time health assessment system; s5, a cloud collaborative optimization mechanism; through intelligent start-stop strategy optimization and energy recovery management, under the urban congestion working condition, compared with a traditional start-stop system, the method can further reduce oil consumption by 15-20%; and under the comprehensive working condition, the oil consumption can be reduced by 8-12%. The fuel consumption cost of the vehicle is reduced, the development trend of global energy conservation and emission reduction is also met, and important economic and environmental significance is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of automotive powertrain control technology, specifically to a method for managing an automotive start-stop system based on intelligent control. Background Technology

[0002] Under the current trend of energy conservation and emission reduction in the automotive industry, the start-stop system has become an important technical means to reduce fuel consumption and emissions. However, traditional start-stop systems have many problems that severely limit their application effectiveness and user experience:

[0003] The triggering logic is simplistic: most existing start-stop systems rely solely on simple thresholds such as vehicle speed and brake pedal status for start-stop decisions. For example, when the vehicle speed is below 5 km / h and the brake pedal is depressed, the system shuts off the engine. This simplistic triggering logic fails to adequately consider complex factors such as driving habits, road conditions, and battery status. In congested traffic with frequent start-stop cycles, the system will repeatedly start and stop the engine, not only failing to achieve energy conservation and emission reduction but also increasing engine wear and reducing driving comfort.

[0004] Battery life is significantly affected: frequent start-stop operations place a heavy burden on the starter motor and battery. Traditional lead-acid batteries experience a 30-50% reduction in cycle life under deep discharge conditions. Because traditional start-stop systems cannot effectively control start-stop frequency and battery charge / discharge states, batteries age prematurely, increasing vehicle operating costs and maintenance frequency.

[0005] The issue of start-stop response delay is prominent: the start-up time of traditional starters is typically between 0.3 and 0.5 seconds. When a vehicle needs to accelerate urgently, this delay can lead to untimely power response, affecting driving safety and the driving experience. For example, when a vehicle needs to start quickly after waiting for a green light at an intersection, a long start-up delay may cause it to miss the optimal starting opportunity, or even lead to a traffic accident.

[0006] Inadequate emission control: Traditional start-stop systems primarily focus on fuel savings during idling, neglecting emission changes during the start-stop process. During engine cold starts, incomplete combustion produces significant amounts of pollutants such as nitrogen oxides (NOx), carbon monoxide (CO), and hydrocarbons (HC). Traditional systems are not optimized for these emissions, potentially leading to a surge in emissions under certain operating conditions, failing to meet increasingly stringent environmental requirements.

[0007] Poor environmental adaptability: In special scenarios such as high and low temperature environments, congested road conditions, and mountain roads, traditional start-stop systems cannot adaptively adjust their control strategies according to actual conditions. For example, in low temperature environments, battery performance degrades, and traditional start-stop systems may experience difficulty starting due to insufficient battery power; during frequent uphill and downhill driving on mountain roads, the system cannot properly control engine start-stop, affecting vehicle power performance and fuel economy.

[0008] These problems make it difficult for traditional car start-stop systems to achieve the desired energy-saving and emission-reduction effects in practical applications, resulting in low user acceptance. There is an urgent need for a more intelligent and efficient start-stop system management method to solve these problems. To this end, a car start-stop system management method based on intelligent control is proposed. Summary of the Invention

[0009] In view of this, the present invention provides a vehicle start-stop system management method based on intelligent control to solve or alleviate the technical problems existing in the prior art, and at least provides a beneficial option.

[0010] The technical solution of this invention is implemented as follows: A method for managing a car start-stop system based on intelligent control, comprising the following steps:

[0011] S1, Multimodal environment perception system;

[0012] S2, Deep Reinforcement Learning Decision Framework;

[0013] S3, hierarchical collaborative control execution;

[0014] S4, Real-time Health Assessment System;

[0015] S5, cloud-based collaborative optimization mechanism.

[0016] Further preferably, in step S1, high-precision sensors are deployed at key parts of the vehicle to acquire real-time vehicle operating status information. A high-precision accelerometer (accuracy ±0.01g) is used to monitor the vehicle's acceleration and deceleration in real time, accurately capturing subtle speed changes. The battery management system (BMS) acquires key parameters such as the battery's state of charge (SOC), state of health (SOH), and internal resistance (SOC accuracy ±2%), providing a comprehensive understanding of the battery's operating status. Transmission system sensors monitor parameters such as gear position, clutch status, and transmission oil temperature in real time, providing accurate transmission system information for start-stop decisions. Advanced sensors and deep learning technology are used to deeply analyze the driver's operating intentions. The deep learning-based pedal behavior analysis model collects pressure change data of the driver's accelerator and brake pedals through a high-precision pressure sensor (sampling rate 100Hz). Combined with historical driving data and driving habit models, it accurately identifies the driver's pedal patterns such as acceleration, deceleration, and constant speed. The steering wheel angle sensor (accuracy ±0.1°) and gyroscope data are combined to determine the driver's steering intention in real time, including steering angle and steering speed. Equipped with an eye-tracking camera (resolution 1920×1080), it further assists in determining driving intentions by analyzing the driver's gaze direction and attention distribution. For example, when the driver is looking at the intersection ahead, it can predict possible steering or acceleration operations in advance. It integrates multiple sensors and communication technologies to comprehensively perceive the vehicle's external environment. Millimeter-wave radar (detection range 200m, angular resolution 0.1°) monitors the distance, relative speed, and direction of movement of vehicles ahead in real time, providing crucial information for following and stop / start decisions. A vision camera (supporting 8 megapixels) uses computer vision algorithms to identify traffic lights, road signs, and traffic congestion conditions ahead, such as planning engine shutdown strategies in advance when a red light is detected. A V2X communication module enables information exchange between the vehicle and the outside world, acquiring real-time traffic information, including road congestion index and remaining time at traffic lights, helping the system make more reasonable stop / start decisions. The navigation system integrates high-precision maps (10cm accuracy) to provide detailed road feature information such as slope, curvature, and speed limits. When the vehicle is driving on a slope, the system can adjust the stop / start strategy according to the slope, avoiding frequent stops and starts on inclines that could affect power performance.

[0017] Further preferably, in step S2, a state space containing a 22-dimensional state vector is constructed to comprehensively describe the vehicle's operating state and environmental information, specifically covering key parameters such as vehicle speed, acceleration, battery SOC, battery temperature, ambient temperature, distance to the vehicle ahead, traffic light status, road gradient, and driving style index. These parameters comprehensively consider multiple dimensions, including the vehicle's own state, driver behavior, and external environment, providing rich and accurate input information for the deep reinforcement learning algorithm. An action space containing both continuous and discrete control dimensions is designed. The continuous control dimension includes starter torque (adjustment range 5-30 Nm, resolution 0.1 Nm) and start-up time window (0.1-0.5 seconds), enabling precise control of the starter's starting process and achieving smooth power output. The discrete control dimension includes start-stop decisions (start / stop / hold) and energy recovery intensity (0-3 levels), flexibly adjusting the engine start-stop state and energy recovery strategy according to different operating conditions and needs to improve energy utilization efficiency. A reward function that comprehensively considers multiple objectives is designed to guide the algorithm in learning the optimal start-stop strategy. Fuel economy (weight 0.4) is evaluated based on fuel savings predicted by the WLTC (Worldwide Harmonized Light Vehicles Test Cycle) model, encouraging the algorithm to choose start-stop strategies that reduce fuel consumption. Driving comfort (weight 0.3) is based on a smoothness index represented by the jerk value (rate of change of acceleration), penalizing start-stop operations that cause vehicle jerking and improving driving comfort. Battery health (weight 0.2) is evaluated based on a battery degradation model using the equivalent full cycle count (EFC), avoiding overcharging and discharging that could damage the battery and extending battery life. Emission control (weight 0.1) considers factors such as cold start emission spikes, establishing a comprehensive emission model to prompt the algorithm to make optimized decisions in emission reduction. It employs a Proximal Policy Optimization (PPO) algorithm combined with a Long Short-Term Memory (LSTM) network architecture to handle sequential decision problems. The PPO algorithm effectively solves the problem of unstable policy updates in traditional reinforcement learning algorithms, improving the convergence speed and stability of the algorithm; the LSTM network can capture the time-series characteristics of the data, better handling the dynamic information during vehicle operation. Meanwhile, a priority experience replay mechanism is introduced, which prioritizes the replay of experiences based on their importance, thereby improving sample utilization efficiency and accelerating the learning speed of the algorithm.

[0018] Further preferably, in step S3, based on the principle of Model Predictive Control (MPC), the start-stop decision sequence is continuously optimized within a prediction time domain of 5-10 seconds. This layer fully considers the thermal management requirements of the vehicle's powertrain system, avoiding engine wear and increased emissions due to frequent cold starts. For example, when it is predicted that the vehicle will enter a congested road section, a reasonable start-stop strategy is planned in advance to reduce unnecessary engine starts, maintain the engine temperature within a suitable range, and use a fuzzy control algorithm to achieve coordination of the working states and smooth torque transition between components such as the starter, generator, battery, and engine. During engine start and stop, the vehicle's jerking sensation is reduced and driving comfort is improved by precisely controlling the working states and torque output of each component. For example, when the engine starts, the work of the starter and generator is coordinated to ensure a smooth engine start; when the engine stops, the power output of the generator and the charging state of the battery are reasonably controlled to ensure stable system operation. The starter uses a brushless DC motor and is equipped with an intelligent controller (response time ≤20ms) to achieve fast and precise start control. Simultaneously, a rapid start-up strategy was developed, reducing the start-up time to less than 0.2 seconds through pre-engagement technology, significantly improving start-stop response speed. Furthermore, an integrated intelligent generator enables optimized energy management during the start-stop process. When the engine stops, the intelligent generator can efficiently generate electricity based on battery status and vehicle power demand, improving energy utilization efficiency.

[0019] More preferably, in step S4, electrochemical impedance spectroscopy (EIS) technology is used to monitor the battery's internal resistance online (measurement frequency range 10mHz-10kHz). By analyzing changes in the battery's internal resistance, the battery's health status is assessed. Simultaneously, a Kalman filter-based SOC / SOH joint estimation algorithm is employed to accurately estimate the battery's state of charge and health status, with an estimation error of <2%. This provides accurate battery status information for start-stop decisions, preventing start-up failures due to insufficient battery power. A current sensor monitors the starter motor's starting current waveform in real time, extracting characteristic parameters such as peak current and start-up time. A starter motor wear prediction model is established using deep learning algorithms. Based on the changing trends of characteristic parameters, the remaining lifespan of the starter motor is predicted, allowing for early detection of potential starter motor faults and timely maintenance and replacement, ensuring the normal operation of the start-stop system. A multi-sensor fusion fault diagnosis method is employed, combined with DS evidence theory for decision-level fusion. Through comprehensive analysis and reasoning of data from multiple sensors, the accuracy and reliability of fault diagnosis are improved. At the same time, a fault tree model is established to classify and analyze possible faults, enabling rapid fault location and isolation. When a system fault occurs, the cause of the fault can be quickly determined, and corresponding measures can be taken to repair it or switch to standby mode to ensure the safe operation of the vehicle.

[0020] Further preferably, in step S5, a lightweight model (less than 1 million parameters) is deployed on the vehicle side to handle tasks with high real-time requirements, such as driving intention recognition and emergency start-stop decisions. Edge computing nodes can process data quickly locally, reducing data transmission latency and improving system response speed and real-time performance. Model training and updates are performed based on massive amounts of vehicle operation data (>1 million vehicles). Utilizing the powerful computing resources of the cloud, the collected large amounts of data are deeply analyzed and mined to continuously optimize the deep reinforcement learning model and other control models, improving model accuracy and adaptability. A federated learning framework is used to achieve a balance between vehicle privacy protection and data sharing. Without disclosing vehicle user privacy data, each vehicle uploads its locally trained model parameters to the cloud. The cloud aggregates and updates the models, and then distributes the optimized models back to the vehicles. In this way, the data remains stationary while the model moves, protecting user privacy while fully utilizing data from multiple vehicles for model optimization, improving overall system performance. Remote upgrades of the control software are achieved through a secure tunnel, supporting differential updates (update package size <500KB). Once the cloud training center has optimized the control strategies and models, the updates can be securely and quickly pushed to the vehicle via OTA technology. The vehicle can complete the system upgrade without having to go to a repair shop, and obtain the latest optimization strategies in a timely manner, maintaining the system's advanced nature and reliability.

[0021] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions:

[0022] I. This invention, through intelligent start-stop strategy optimization and energy recovery management, can further reduce fuel consumption by 15-20% compared to traditional start-stop systems under urban congestion conditions; under combined operating conditions, fuel consumption can be reduced by 8-12%. This not only reduces vehicle fuel costs but also aligns with the global trend of energy conservation and emission reduction, possessing significant economic and environmental implications.

[0023] Second, this invention employs advanced starter technology and a rapid start strategy, reducing the starter's start time from 0.3-0.5 seconds in traditional systems to 0.15-0.2 seconds, improving response speed by 60%. When the vehicle needs emergency acceleration or starting, it can quickly start the engine, providing sufficient power and improving driving safety and smoothness.

[0024] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0026] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0027] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0028] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] like Figure 1 As shown, this embodiment of the invention provides a vehicle start-stop system management method based on intelligent control, including the following steps:

[0030] S1, Multimodal environment perception system;

[0031] S2, Deep Reinforcement Learning Decision Framework;

[0032] S3, hierarchical collaborative control execution;

[0033] S4, Real-time Health Assessment System;

[0034] S5, cloud-based collaborative optimization mechanism.

[0035] In one embodiment, in S1, high-precision sensors are deployed at key parts of the vehicle to acquire real-time vehicle operating status information. A high-precision accelerometer (accuracy ±0.01g) is used to monitor the vehicle's acceleration and deceleration in real time, accurately capturing subtle speed changes. The Battery Management System (BMS) acquires key parameters such as the battery's state of charge (SOC), state of health (SOH), and internal resistance (SOC accuracy ±2%) to comprehensively understand the battery's operating status. Transmission system sensors monitor parameters such as gear position, clutch status, and transmission oil temperature in real time, providing accurate transmission system information for start-stop decisions. Advanced sensors and deep learning technology are used to deeply analyze the driver's operating intentions. The deep learning-based pedal behavior analysis model collects pressure change data of the driver's accelerator and brake pedals through a high-precision pressure sensor (sampling rate 100Hz). Combined with historical driving data and driving habit models, it accurately identifies the driver's pedal patterns such as acceleration, deceleration, and constant speed. The steering wheel angle sensor (accuracy ±0.1°) and gyroscope data are combined to determine the driver's steering intention in real time, including steering angle and steering speed. Equipped with an eye-tracking camera (resolution 1920×1080), it further assists in determining driving intentions by analyzing the driver's gaze direction and attention distribution. For example, when the driver is looking at the intersection ahead, it can predict possible steering or acceleration operations in advance. It integrates multiple sensors and communication technologies to comprehensively perceive the vehicle's external environment. Millimeter-wave radar (detection range 200m, angular resolution 0.1°) monitors the distance, relative speed, and direction of movement of vehicles ahead in real time, providing crucial information for following and stop / start decisions. A vision camera (supporting 8 megapixels) uses computer vision algorithms to identify traffic lights, road signs, and traffic congestion conditions ahead, such as planning engine shutdown strategies in advance when a red light is detected. A V2X communication module enables information exchange between the vehicle and the outside world, acquiring real-time traffic information, including road congestion index and remaining time at traffic lights, helping the system make more reasonable stop / start decisions. The navigation system integrates high-precision maps (10cm accuracy) to provide detailed road feature information such as slope, curvature, and speed limits. When the vehicle is driving on a slope, the system can adjust the stop / start strategy according to the slope, avoiding frequent stops and starts on inclines that could affect power performance.

[0036] In one embodiment, in S2, a state space containing a 22-dimensional state vector is constructed to comprehensively describe the vehicle's operating state and environmental information, specifically covering key parameters such as vehicle speed, acceleration, battery SOC, battery temperature, ambient temperature, distance to the vehicle ahead, traffic light status, road gradient, and driving style index. These parameters comprehensively consider multiple dimensions, including the vehicle's own state, driver behavior, and external environment, providing rich and accurate input information for the deep reinforcement learning algorithm. An action space containing continuous and discrete control dimensions is designed. The continuous control dimension includes starter torque (adjustment range 5-30 Nm, resolution 0.1 Nm) and start time window (0.1-0.5 seconds), which can precisely control the starter's starting process and achieve smooth power output. The discrete control dimension includes start-stop decision (start / stop / hold) and energy recovery intensity (0-3 levels), which flexibly adjust the engine start-stop state and energy recovery strategy according to different operating conditions and needs to improve energy utilization efficiency. A reward function that comprehensively considers multiple objectives is designed to guide the algorithm to learn the optimal start-stop strategy. Fuel economy (weight 0.4) is evaluated based on fuel savings predicted by the WLTC (Worldwide Harmonized Light Vehicles Test Cycle) model, encouraging the algorithm to choose start-stop strategies that reduce fuel consumption. Driving comfort (weight 0.3) is based on a smoothness index represented by the jerk value (rate of change of acceleration), penalizing start-stop operations that cause vehicle jerking and improving driving comfort. Battery health (weight 0.2) is evaluated based on a battery degradation model using the equivalent full cycle count (EFC), avoiding overcharging and discharging that could damage the battery and extending battery life. Emission control (weight 0.1) considers factors such as cold start emission spikes, establishing a comprehensive emission model to prompt the algorithm to make optimized decisions in emission reduction. It employs a Proximal Policy Optimization (PPO) algorithm combined with a Long Short-Term Memory (LSTM) network architecture to handle sequential decision problems. The PPO algorithm effectively solves the problem of unstable policy updates in traditional reinforcement learning algorithms, improving the convergence speed and stability of the algorithm; the LSTM network can capture the time-series characteristics of the data, better handling the dynamic information during vehicle operation. Meanwhile, a priority experience replay mechanism is introduced, which prioritizes the replay of experiences based on their importance, thereby improving sample utilization efficiency and accelerating the learning speed of the algorithm.

[0037] In one embodiment, in S3, based on the principle of Model Predictive Control (MPC), the start-stop decision sequence is continuously optimized within a 5-10 second prediction time domain. This layer fully considers the thermal management requirements of the vehicle's powertrain system, avoiding engine wear and increased emissions due to frequent cold starts. For example, when it is predicted that the vehicle will enter a congested road section, a reasonable start-stop strategy is planned in advance to reduce unnecessary engine starts and maintain the engine temperature within a suitable range. A fuzzy control algorithm is used to achieve coordination of the working states and smooth torque transition between components such as the starter, generator, battery, and engine. During engine start-up and stop, the vehicle's jerking sensation is reduced and driving comfort is improved by precisely controlling the working states and torque output of each component. For example, when the engine starts, the work of the starter and generator is coordinated to ensure a smooth engine start; when the engine stops, the generator's power output and the battery's charging state are reasonably controlled to ensure stable system operation. The starter uses a brushless DC motor and is equipped with an intelligent controller (response time ≤20ms) to achieve fast and precise start control. Simultaneously, a rapid start-up strategy was developed, reducing the start-up time to less than 0.2 seconds through pre-engagement technology, significantly improving start-stop response speed. Furthermore, an integrated intelligent generator enables optimized energy management during the start-stop process. When the engine stops, the intelligent generator can efficiently generate electricity based on battery status and vehicle power demand, improving energy utilization efficiency.

[0038] In one embodiment, in S4, electrochemical impedance spectroscopy (EIS) technology is used to monitor the battery's internal resistance online (measurement frequency range 10mHz-10kHz). By analyzing changes in the battery's internal resistance, the battery's health status is assessed. Simultaneously, a Kalman filter-based SOC / SOH joint estimation algorithm is employed to accurately estimate the battery's state of charge and health status, with an estimation error of <2%. This provides accurate battery status information for start-stop decisions, preventing start-up failures due to insufficient battery power. A current sensor monitors the starter motor's starting current waveform in real time, extracting characteristic parameters such as peak current and start-up time. A starter motor wear prediction model is established using deep learning algorithms. Based on the changing trends of characteristic parameters, the remaining lifespan of the starter motor is predicted, allowing for early detection of potential starter motor faults and timely maintenance and replacement, ensuring the normal operation of the start-stop system. A multi-sensor fusion fault diagnosis method is employed, combined with DS evidence theory for decision-level fusion. Through comprehensive analysis and reasoning of data from multiple sensors, the accuracy and reliability of fault diagnosis are improved. At the same time, a fault tree model is established to classify and analyze possible faults, enabling rapid fault location and isolation. When a system fault occurs, the cause of the fault can be quickly determined, and corresponding measures can be taken to repair it or switch to standby mode to ensure the safe operation of the vehicle.

[0039] In one embodiment, in S5, a lightweight model (less than 1 million parameters) is deployed on the vehicle side to handle tasks with high real-time requirements, such as driving intention recognition and emergency start-stop decisions. Edge computing nodes can process data quickly locally, reducing data transmission latency and improving system response speed and real-time performance. Model training and updates are performed based on massive vehicle operation data (>1 million vehicles). Utilizing the powerful computing resources of the cloud, the collected large amount of data is deeply analyzed and mined to continuously optimize the deep reinforcement learning model and other control models, improving the model's accuracy and adaptability. A federated learning framework is adopted to achieve a balance between vehicle privacy protection and data sharing. Without disclosing vehicle user privacy data, each vehicle uploads the locally trained model parameters to the cloud. The cloud aggregates and updates the model, and then distributes the optimized model back to the vehicle side. In this way, the data remains stationary while the model moves, protecting user privacy while fully utilizing data from multiple vehicles for model optimization, improving the overall system performance. Remote upgrades of the control software are achieved through a secure tunnel, supporting differential updates (update package size <500KB). Once the cloud training center has optimized the control strategies and models, the updates can be securely and quickly pushed to the vehicle via OTA technology. The vehicle can complete the system upgrade without having to go to a repair shop, and obtain the latest optimization strategies in a timely manner, maintaining the system's advanced nature and reliability.

[0040] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for managing a vehicle start-stop system based on intelligent control, characterized in that: Includes the following steps: S1, Multimodal environment perception system; S2, Deep Reinforcement Learning Decision Framework; S3, hierarchical collaborative control execution; S4, Real-time Health Assessment System; S5, cloud-based collaborative optimization mechanism.

2. The vehicle start-stop system management method based on intelligent control according to claim 1, characterized in that: In step S1, high-precision sensors are deployed at key parts of the vehicle to acquire real-time vehicle operating status information. A high-precision accelerometer (accuracy ±0.01g) monitors the vehicle's acceleration and deceleration in real time, accurately capturing subtle speed changes. The Battery Management System (BMS) acquires key parameters such as the battery's state of charge (SOC), state of health (SOH), and internal resistance (SOC accuracy ±2%), providing a comprehensive understanding of the battery's operating status. Transmission system sensors monitor parameters such as gear position, clutch status, and transmission oil temperature in real time, providing accurate transmission system information for start-stop decisions. Advanced sensors and deep learning technology are used to deeply analyze the driver's operational intentions. The deep learning-based pedal behavior analysis model collects pressure change data of the driver's accelerator and brake pedals through a high-precision pressure sensor (sampling rate 100Hz). Combined with historical driving data and driving habit models, it accurately identifies the driver's pedal patterns such as acceleration, deceleration, and constant speed. The steering wheel angle sensor (accuracy ±0.1°) and gyroscope data are combined to determine the driver's steering intention in real time, including steering angle and steering speed. Equipped with an eye-tracking camera (resolution 1920×1080), it further assists in determining driving intentions by analyzing the driver's gaze direction and attention distribution. For example, when the driver is looking at the intersection ahead, it can predict possible steering or acceleration operations in advance. Integrating multiple sensors and communication technologies, it comprehensively perceives the vehicle's external environment. Millimeter-wave radar (detection range 200m, angular resolution 0.1°) monitors the distance, relative speed, and direction of movement of vehicles ahead in real time, providing crucial information for following and stop / start decisions; The vision camera (supports 8 megapixels) uses computer vision algorithms to identify information such as traffic lights, road signs, and traffic congestion ahead. For example, it can plan engine shutdown strategies in advance when it recognizes a red light. The V2X communication module enables information exchange between the vehicle and the outside world, obtaining real-time traffic information, including road congestion index and remaining traffic light time, to help the system make more reasonable start-stop decisions. The navigation system integrates high-precision maps (10cm accuracy) to provide detailed road feature information such as road slope, curvature, and speed limit. When the vehicle is driving on a slope, the system can adjust the start-stop strategy according to the slope to avoid frequent start-stop on slopes affecting power performance.

3. The vehicle start-stop system management method based on intelligent control according to claim 1, characterized in that: In S2, a state space containing 22-dimensional state vectors is constructed to comprehensively describe the vehicle's operating state and environmental information, specifically covering key parameters such as vehicle speed, acceleration, battery SOC, battery temperature, ambient temperature, distance to the vehicle ahead, traffic light status, road gradient, and driving style index. These parameters comprehensively consider multiple dimensions, including the vehicle's own state, driver behavior, and external environment, providing rich and accurate input information for the deep reinforcement learning algorithm. An action space containing continuous and discrete control dimensions is designed. The continuous control dimension includes starter torque (adjustment range 5-30 Nm, resolution 0.1 Nm) and start time window (0.1-0.5 seconds), which can precisely control the starter's starting process and achieve smooth power output. The discrete control dimension includes start-stop decision (start / stop / hold) and energy recovery intensity (0-3 levels), which flexibly adjusts the engine start-stop state and energy recovery strategy according to different operating conditions and needs to improve energy utilization efficiency. A reward function that comprehensively considers multiple objectives is designed to guide the algorithm to learn the optimal start-stop strategy. Fuel economy (weight 0.4) is evaluated based on fuel savings predicted by the WLTC (Worldwide Harmonized Light Vehicles Test Cycle) model, encouraging the algorithm to choose start-stop strategies that reduce fuel consumption. Driving comfort (weight 0.3) is based on a smoothness index represented by the jerk value (rate of change of acceleration), penalizing start-stop operations that cause vehicle jerking and improving driving comfort. Battery health (weight 0.2) is evaluated based on a battery degradation model using the equivalent full cycle count (EFC), avoiding overcharging and discharging that could damage the battery and extending battery life. Emission control (weight 0.1) considers factors such as cold start emission spikes, establishing a comprehensive emission model to prompt the algorithm to make optimized decisions in emission reduction. It employs a Proximal Policy Optimization (PPO) algorithm combined with a Long Short-Term Memory (LSTM) network architecture to handle sequential decision problems. The PPO algorithm effectively solves the problem of unstable policy updates in traditional reinforcement learning algorithms, improving the convergence speed and stability of the algorithm; the LSTM network can capture the time-series characteristics of the data, better handling the dynamic information during vehicle operation. Meanwhile, a priority experience replay mechanism is introduced, which prioritizes the replay of experiences based on their importance, thereby improving sample utilization efficiency and accelerating the learning speed of the algorithm.

4. The vehicle start-stop system management method based on intelligent control according to claim 1, characterized in that: In S3, based on the principle of model predictive control (MPC), the start-stop decision sequence is rolled over and optimized within a prediction time domain of 5-10 seconds. This layer fully considers the thermal management requirements of the vehicle's powertrain system, avoiding engine wear and increased emissions caused by frequent cold starts. For example, when it predicts that the vehicle will enter a congested area, a reasonable start-stop strategy is planned in advance to reduce unnecessary engine starts and maintain the engine temperature within a suitable range. Fuzzy control algorithms are used to coordinate the working states and smooth torque transitions between components such as the starter motor, generator, battery, and engine. During engine start-up and shutdown, the working states and torque output of each component are precisely controlled to reduce vehicle jerking and improve driving comfort. For example, when the engine starts, the work of the starter motor and generator is coordinated to ensure a smooth engine start; when the engine stops, the generator's power output and battery charging status are rationally controlled to ensure stable system operation. The starter motor uses a brushless DC motor and is equipped with an intelligent controller (response time ≤20ms) to achieve fast and precise start control. At the same time, a fast start strategy has been developed, using pre-engagement technology to shorten the start time to less than 0.2 seconds, greatly improving the start-stop response speed. In addition, an integrated intelligent generator enables optimized energy management during start-stop processes. When the engine stops, the intelligent generator can generate electricity efficiently based on the battery status and the vehicle's power demand, thereby improving energy utilization efficiency.

5. The vehicle start-stop system management method based on intelligent control according to claim 1, characterized in that: In step S4, electrochemical impedance spectroscopy (EIS) is used to monitor the battery's internal resistance online (measurement frequency range 10mHz-10kHz). By analyzing changes in the battery's internal resistance, the battery's health status is assessed. Simultaneously, a Kalman filter-based SOC / SOH joint estimation algorithm is employed to accurately estimate the battery's state of charge and health status, with an estimation error of <2%. This provides accurate battery status information for start-stop decisions, preventing start-up failures due to insufficient battery power. A current sensor monitors the starter motor's starting current waveform in real time, extracting characteristic parameters such as peak current and start-up time. A starter motor wear prediction model is established using deep learning algorithms. Based on the changing trends of characteristic parameters, the remaining lifespan of the starter motor is predicted, allowing for early detection of potential starter motor faults and timely maintenance and replacement, ensuring the normal operation of the start-stop system. A multi-sensor fusion fault diagnosis method is employed, combined with DS evidence theory for decision-level fusion. Through comprehensive analysis and reasoning of data from multiple sensors, the accuracy and reliability of fault diagnosis are improved. At the same time, a fault tree model is established to classify and analyze possible faults, enabling rapid fault location and isolation. When a system fault occurs, the cause of the fault can be quickly determined, and corresponding measures can be taken to repair it or switch to standby mode to ensure the safe operation of the vehicle.

6. The vehicle start-stop system management method based on intelligent control according to claim 1, characterized in that: In S5, a lightweight model (less than 1 million parameters) is deployed on the vehicle side to handle tasks with high real-time requirements, such as driving intention recognition and emergency start-stop decisions. Edge computing nodes can process data quickly locally, reducing data transmission latency and improving system response speed and real-time performance. Model training and updates are based on massive vehicle operation data (>1 million vehicles). Utilizing the powerful computing resources of the cloud, deep analysis and mining are performed on the large amount of collected data to continuously optimize deep reinforcement learning models and other control models, improving model accuracy and adaptability. A federated learning framework is adopted to achieve a balance between vehicle privacy protection and data sharing. Without disclosing vehicle user privacy data, each vehicle uploads the locally trained model parameters to the cloud. The cloud aggregates and updates the models, and then distributes the optimized models back to the vehicles. In this way, the data remains stationary while the model moves, protecting user privacy while fully utilizing data from multiple vehicles for model optimization, improving overall system performance. Remote upgrades of the control software are achieved through a secure tunnel, supporting differential updates (update package size <500KB). Once the cloud training center has optimized the control strategies and models, the updates can be securely and quickly pushed to the vehicle via OTA technology. The vehicle can complete the system upgrade without having to go to a repair shop, and obtain the latest optimization strategies in a timely manner, maintaining the system's advanced nature and reliability.

Citation Information

Cited By

  • Method and system for quickly starting hybrid electric vehicle at low temperature based on microwave heating

    CN121473949A

  • Microwave-based rapid start method and system for hybrid electric vehicles in low temperatures

    CN121473949B