Multi-controller cooperative control method and device and vehicle

Abstract

The invention relates to the technical field of vehicle control, in particular to a multi-controller cooperative control method and device and a vehicle, and the method comprises the steps: transmitting the obtained current state information of the vehicle to a deep reinforcement learning model, and generating a multi-order dynamic weight coefficient; receiving a multi-order dynamic weight coefficient, and generating a response instruction through dynamic coupling; and the response instruction is sent to a vehicle subsystem controller, and the vehicle subsystem controller controls an actuator to execute a response action according to the response instruction, obtains an execution result and feeds back the execution result to the deep reinforcement learning model. According to the method, multi-objective optimization and dynamic adaptation can be realized, and the user experience is improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, specifically to a multi-controller collaborative control method, device, and vehicle. Background Technology

[0002] In the field of new energy vehicle technology, energy management strategies for plug-in hybrid / range-extended electric vehicles are continuously iterating and upgrading, aiming to meet users' core needs for dynamic adaptation of vehicle performance and economy under complex road conditions and diverse usage scenarios. However, current vehicle-road-cloud integrated vehicle system control strategies still face significant technical bottlenecks, specifically in two aspects:

[0003] On the one hand, while some technical solutions achieve the integration and comprehensive decision-making of data from the vehicle, road, and cloud, they employ fixed weight rules in their reward function design, with the adjustment of vehicle-side strategy weights entirely dependent on preset logic. This model cannot dynamically adapt to key dynamic factors such as battery (SOH) degradation and changes in user driving habits, making it difficult for vehicle-side decisions to obtain optimal solutions in complex scenarios, resulting in significantly insufficient system environmental adaptability.

[0004] On the other hand, another type of technical solution lacks a dynamic coordination mechanism among multiple controllers, failing to establish a dynamic weight-based coordination strategy between the vehicle control unit (VCU) and key controllers such as the battery management system (BMS), motor controller (MCU), and body control system (BCM). This results in each controller still executing actions according to a fixed priority in complex scenarios, making it difficult to effectively coordinate conflicts between multiple objectives such as energy efficiency optimization, battery life maintenance, driving comfort, and safety risk protection, thus failing to achieve the global optimization of the vehicle system performance.

[0005] In summary, current vehicle-side control strategies are mostly limited to optimizing a single controller or adjusting a fixed architecture, and have not yet broken through the traditional technical paradigm of fixed rules and single-domain optimization. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-controller collaborative control method, device, and vehicle that can achieve multi-objective optimization and dynamic adaptation, thereby improving the user experience.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] In a first aspect, the present invention discloses a multi-controller cooperative control method, comprising:

[0009] The acquired vehicle current state information is sent to the deep reinforcement learning model to generate multi-order dynamic weight coefficients;

[0010] It receives multi-order dynamic weighting coefficients and generates response commands through dynamic coupling;

[0011] The response command is sent to the vehicle subsystem controller, which controls the actuator to perform the response action according to the response command and obtains the execution result to feed back to the deep reinforcement learning model.

[0012] Furthermore, the multi-level dynamic weighting coefficients include energy efficiency optimization weighting coefficients, battery life maintenance weighting coefficients, comfort assurance weighting coefficients, and safety assessment weighting coefficients.

[0013] Furthermore, the initial dynamic weight coefficients are obtained based on the calculation formula, and then the initial dynamic weight coefficients are adjusted according to the current operating conditions of the vehicle to obtain multi-order dynamic weight coefficients.

[0014] The formula for calculating the energy efficiency optimization weight coefficient w1 is as follows: In the formula, F is the fuel consumption rate, in L / h; η is the motor efficiency, in %;

[0015] The formula for calculating the battery life maintenance weighting coefficient w2 is as follows: In the formula, SOH represents the state of health of the power battery, in %; N represents the number of start-stop cycles of the range extender; and C represents the charge-discharge rate.

[0016] The formula for calculating the comfort protection weighting coefficient w3 is as follows:

[0017] In the formula, ΔT is the temperature deviation in different areas of the cabin, in °C; ΔTq is the torque fluctuation rate, in %; Δspd is the speed fluctuation rate, in %; and L is the range extender noise, in dB.

[0018] The formula for calculating the safety assessment weighting coefficient w4 is as follows: In the formula, ΔT cell For the temperature difference of the power battery cells, F flag This is the tire pressure abnormality indicator. P represents the brake hydraulic pressure in bar. normal P is the standard pressure for brake hydraulic pressure. max This is the maximum permissible pressure of the brake hydraulic fluid.

[0019] Furthermore, in response to the safety assessment weight coefficient exceeding the first threshold, the safety assessment weight coefficient is placed with the highest priority, and the energy efficiency optimization weight coefficient, battery life maintenance weight coefficient, and comfort protection weight coefficient are adjusted according to the risk level.

[0020] In response to the safety assessment weight coefficient exceeding the second threshold, the safety assessment weight coefficient will forcibly take over control, cut off the power supply to unnecessary loads, and limit the motor power.

[0021] Furthermore, the dynamic coupling calculation formula is: M = A × w + b; where M is the response instruction allocation matrix, A is the correction coefficient matrix, w is the multi-order dynamic weight coefficient matrix, and b is the bias term.

[0022] Furthermore, the correction coefficient matrix is ​​reasonably set according to vehicle requirements; the multi-order dynamic weight coefficient matrix also includes a weight coefficient w5 corresponding to the current working state of the range extender.

[0023] Furthermore, the deep reinforcement learning model is deployed in the cloud to receive the vehicle's current state information and the feedback execution results, and to generate multi-level dynamic weight coefficients.

[0024] Furthermore, the vehicle subsystem controller includes a powertrain controller, an energy system controller, and a body system controller.

[0025] Secondly, the present invention discloses a multi-controller cooperative control device, comprising:

[0026] The sending module is used to send the acquired vehicle current state information to the deep reinforcement learning model to generate multi-order dynamic weight coefficients;

[0027] The receiving module is used to receive multi-order dynamic weighting coefficients and generate response commands through dynamic coupling.

[0028] The execution module is used to send the response command to the vehicle subsystem controller, which controls the actuator to perform the response action according to the response command and obtains the execution result to feed back to the deep reinforcement learning model.

[0029] In a first aspect, the present invention discloses a vehicle including a processor and a memory, the memory storing a computer program, wherein the processor executes the computer program stored in the memory to implement the steps in the above-described multi-controller cooperative control method.

[0030] The present invention has the following unexpected beneficial effects:

[0031] This invention generates multi-level dynamic weight coefficients by sending the acquired vehicle current state information to a deep reinforcement learning model, thus eliminating the reliance on preset rules for weight generation. The generated dynamic weights can match the vehicle's current state in real time; for example, they automatically increase energy efficiency weights to reduce range extender start-stop during traffic jams and strengthen battery life maintenance weights in low temperatures. Compared to fixed weights, this invention can flexibly adapt to complex road conditions and real-time vehicle states, solving the problem of insufficient system environmental adaptability and ensuring the rationality of control strategies in different scenarios. Furthermore, response commands are generated through dynamic coupling. The dynamic coupling process integrates multiple objectives such as energy efficiency optimization, battery life, ride comfort, and safety risks, avoiding independent actions by each controller according to fixed priorities. The response commands directly address the vehicle subsystem controllers, achieving synchronous responses from multiple controllers and achieving global optimization of the entire vehicle system performance, rather than local optimization of a single controller or single objective. Attached Figure Description

[0032] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in 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 the present invention.

[0033] Figure 1 A flowchart illustrating the multi-controller collaborative control method according to an embodiment of the present invention is shown.

[0034] Figure 2 A schematic diagram of the structure of the multi-controller collaborative control device according to an embodiment of the present invention is shown.

[0035] Figure 3 A schematic diagram of the architecture of the multi-controller collaborative control device according to an embodiment of the present invention is shown. Detailed Implementation

[0036] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0037] In one embodiment, see Figure 1 As shown, this invention discloses a multi-controller cooperative control method, which includes:

[0038] The acquired vehicle current state information is sent to the deep reinforcement learning model to generate multi-order dynamic weight coefficients;

[0039] It receives multi-order dynamic weighting coefficients and generates response commands through dynamic coupling;

[0040] The response command is sent to the vehicle subsystem controller, which controls the actuator to perform the response action according to the response command and obtains the execution result to feed back to the deep reinforcement learning model.

[0041] This invention generates multi-level dynamic weight coefficients by sending the acquired vehicle current state information to a deep reinforcement learning model, thus eliminating the reliance on preset rules for weight generation. The generated dynamic weights can match the vehicle's current state in real time; for example, they automatically increase energy efficiency weights to reduce range extender start-stop during traffic jams and strengthen battery life maintenance weights in low temperatures. Compared to fixed weights, this invention can flexibly adapt to complex road conditions and real-time vehicle states, solving the problem of insufficient system environmental adaptability and ensuring the rationality of control strategies in different scenarios. Furthermore, response commands are generated through dynamic coupling. The dynamic coupling process integrates multiple objectives such as energy efficiency optimization, battery life, ride comfort, and safety risks, avoiding independent actions by each controller according to fixed priorities. The response commands directly address the vehicle subsystem controllers, achieving synchronous responses from multiple controllers and achieving global optimization of the entire vehicle system performance, rather than local optimization of a single controller or single objective.

[0042] Meanwhile, the method described in this invention feeds the execution results back to the deep reinforcement learning model, possessing sustainable optimization capabilities compared to existing static control strategies. The execution results of the vehicle-to-vehicle subsystem can feed back into the deep reinforcement learning model, iteratively updating the weight generation logic through algorithms such as gradient descent. Therefore, during long-term use, the system can adapt to long-term dynamic factors such as changes in user driving habits and battery SOH degradation, without requiring manual rule adjustments, continuously improving control accuracy and adaptability, and avoiding the performance degradation problem that occurs after existing technical strategies become fixed.

[0043] Compared to the shortcomings of existing technologies that struggle to meet real-time user needs, this solution leverages dynamic weights and collaborative commands to translate technological optimization into tangible improvements in user experience. Dynamic weights can indirectly reflect potential user needs, such as increasing energy recovery intensity for economy-conscious users in congested areas and lowering the range extender activation threshold for power-oriented users on inclines, based on historical driving preferences. Response commands precisely control actuators through subsystem controllers, avoiding sacrificing user experience for a single optimization goal and achieving a balance between technical indicators and user experience.

[0044] In a preferred embodiment of the present invention, the multi-level dynamic weighting coefficients include energy efficiency optimization weighting coefficients, battery life maintenance weighting coefficients, comfort assurance weighting coefficients, and safety assessment weighting coefficients.

[0045] This preferred implementation breaks down the complex control requirements of the vehicle into quantifiable and adjustable target dimensions through a clear division of four-level weights: the energy efficiency optimization weight coefficient w1 focuses on the efficiency of the power system, such as the fuel consumption rate of the range extender and the efficiency of the motor, directly corresponding to the user's core demand for vehicle economy; the battery life maintenance weight coefficient w2 anchors to key parameters such as the state of health (SOH) of the power battery and the charge / discharge rate, solving the problem of long-term performance degradation caused by battery degradation; the comfort assurance weight coefficient w3 is related to comfort indicators such as cabin temperature deviation, torque fluctuation rate, and range extender noise, avoiding the design defect of prioritizing performance over user experience; the safety assessment weight coefficient w4 is based on safety signals such as battery cell temperature difference, braking pressure, and tire pressure, ensuring that safety goals are not diluted by other requirements. The four-level weights form a comprehensive system with quantifiable targets, enabling the vehicle-side VCU (Vehicle Control Unit) to clearly weigh the priorities of multiple targets during dynamic coupling calculations, completely solving the problem of existing technologies where multiple target conflicts lack clear coordination basis.

[0046] In a preferred embodiment of the present invention, an initial dynamic weight coefficient is obtained according to the calculation formula, and then the initial dynamic weight coefficient is adjusted according to the current operating conditions of the vehicle to obtain a multi-order dynamic weight coefficient.

[0047] The formula for calculating the energy efficiency optimization weight coefficient w1 is as follows: In the formula, F represents the fuel consumption rate in L / h, and η represents the motor efficiency in %. This calculation formula integrates two energy efficiency indicators: fuel consumption rate F and motor efficiency η. This demonstrates that lower fuel consumption results in a greater contribution to energy efficiency, directly addressing users' economic needs; higher motor efficiency (η) leads to a greater contribution to the energy efficiency of the electric drive system, covering efficiency optimization in pure electric drive scenarios. Coefficients a1 and a2 can be flexibly configured according to vehicle model positioning, achieving precise energy efficiency optimization through hybrid powertrain coordination, avoiding the problem of current technologies focusing solely on either fuel or motor, leading to a one-sided view of system energy efficiency.

[0048] The formula for calculating the battery life maintenance weighting coefficient w2 is as follows: In the formula, SOH represents the health status of the power battery, in percentage; N represents the number of start-stop cycles of the range extender; and C represents the charge / discharge rate.

[0049] The higher the State of Health (SOH) of a power battery, the greater the fundamental weight of battery life maintenance.

[0050] e -0.1*N This demonstrates that the more frequently the range extender starts and stops, the greater the indirect loss to battery life. Frequent starts and stops can easily cause fluctuations in battery charging and discharging. This factor decreases as the number of starts and stops increases, prompting the system to reduce unnecessary range extender starts and stops.

[0051] (1-C) reflects that the higher the charge / discharge rate, the faster the battery life is depleted. This factor decreases when the charge / discharge rate is high, which drives the system to optimize the charge / discharge strategy.

[0052] The three technologies work together to comprehensively cover the impact of the battery's own state, the influence of external systems, and the effects of charging and discharging behavior on lifespan. This addresses the shortcomings of existing technologies that only focus on a single battery parameter and ignore the damage caused by the coupling of multiple factors, enabling a multi-dimensional and accurate assessment of battery lifespan maintenance.

[0053] The formula for calculating the comfort protection weighting coefficient w3 is as follows:

[0054] In the formula, ΔT is the temperature deviation of different areas of the cabin, in °C; ΔTq is the torque fluctuation rate, in %; Δspd is the speed fluctuation rate, in %; and L is the range extender noise, in dB.

[0055] This calculation integrates four comfort dimensions: cabin temperature uniformity (ΔT), power smoothness (ΔTq), engine speed fluctuation (Δspd), and range extender noise (L). Each dimension is weighted according to the logic that the smaller the deviation or fluctuation, the greater its contribution; for example, the smaller ΔT, the greater the weighting. The closer the value is to 1, the more accurately it quantifies the user-perceived comfort index. Compared to existing technologies that rely on fixed logic for comfort control, such as fixed air conditioning settings or fixed power smoothness thresholds, this formula can dynamically adjust comfort weights based on real-time deviations. For example, when traffic congestion causes the range extender noise L to increase, By reducing noise, the system will prioritize optimizing noise control to achieve multi-dimensional comfort and coordination that adapts to different operating conditions.

[0056] The formula for calculating the safety assessment weighting coefficient w4 is as follows: In the formula, ΔT cell For the temperature difference of the power battery cells, F flag This is the tire pressure abnormality indicator. P represents the brake hydraulic pressure in bar. normal P is the standard pressure for brake hydraulic pressure. max This is the maximum permissible pressure of the brake hydraulic fluid.

[0057] ΔT cell Quantifying the risk of battery thermal runaway: the greater the temperature difference between individual cells, the higher the thermal risk; F flag Abnormal tire pressure is a veto item, and a high-weight event is triggered directly when an abnormality occurs. Quantify the deviation of the brake hydraulic pressure to reflect the braking safety status.

[0058] To ensure that the safety weight does not exceed the limit, and when any safety indicator is abnormal, w4 will quickly rise and be forced to take priority, thus solving the problems of lagging safety response and insufficient coordination of multiple safety factors in existing technologies, and realizing a safety guarantee that combines dynamic monitoring and rigid constraints.

[0059] The calculation formulas for the four weights are all objectively quantified using measurable physical quantities, and each variable is a real-time dynamic parameter of the vehicle, enabling the weight coefficients to change in real time with operating conditions and vehicle status. This provides a clear and calculable multi-objective weight benchmark for subsequent dynamic coupling to generate response commands. When coordinating the conflicts between multiple objectives such as energy efficiency, battery life, comfort, and safety, controllers such as the VCU no longer rely on fixed rules or subjective priorities, but instead make globally optimal decisions based on precisely quantified weights, ultimately achieving dynamic coordination and globally optimal performance of the entire vehicle system.

[0060] In a preferred embodiment of the present invention, in response to the safety assessment weight coefficient exceeding a first threshold, the safety assessment weight coefficient is placed with the highest priority, and the energy efficiency optimization weight coefficient, battery life maintenance weight coefficient, and comfort protection weight coefficient are adjusted according to the risk level; in response to the safety assessment weight coefficient exceeding a second threshold, the safety assessment weight coefficient is forcibly taken over control, cutting off the power supply to unnecessary loads and limiting the motor power. The second threshold is greater than the first threshold.

[0061] Existing technologies lack a tiered response logic for safety risks. For minor risks, they excessively sacrifice other objectives, such as significantly limiting power due to slight tire pressure abnormalities. Conversely, for major risks, they fail to respond adequately, such as failing to intervene in time before battery thermal runaway. This preferred embodiment achieves full-scenario coverage through a tiered design of a first threshold (dynamic priority adjustment) and a second threshold (forced takeover).

[0062] In the first threshold scenario, safety assessment is given the highest priority, but the weights for energy efficiency, battery life, and comfort can still be dynamically adjusted. For example, if the tire temperature is slightly high during high-speed driving, causing the safety assessment weight coefficient to exceed the first threshold, the system will prioritize tire pressure safety, but energy efficiency optimization can still be considered to a certain extent, and basic comfort can be maintained, such as not forcibly shutting off the air conditioner, but only adjusting the fan speed to assist in heat dissipation. This ensures that safety is prioritized while avoiding meaningless sacrifices of economy and comfort in low-risk scenarios, achieving flexible coordination of multiple objectives under safety constraints.

[0063] The second threshold scenario involves a safety assessment weighting coefficient that forces the takeover of control. By rigidly cutting off power to unnecessary loads (such as entertainment systems and seat heaters) and limiting motor power, the system directly reduces risk sources at the hardware level. For example, when the temperature difference of a battery cell suddenly increases and approaches thermal runaway, the system can instantly cut off unnecessary electrical loads and limit motor power to a safe range, buying time for risk management and completely solving the problem of delayed response to extreme safety risks in existing technologies.

[0064] In a preferred embodiment of the present invention, the dynamic coupling calculation formula is: M = A × w + b; where M is the response instruction allocation matrix, A is the correction coefficient matrix, w is the multi-order dynamic weight coefficient matrix, and b is the bias term.

[0065] The correction coefficient matrix A can be flexibly adjusted based on differences in vehicle hardware (such as range extender power, battery capacity, and motor performance) and scenario requirements (urban commuting, long-distance highway travel, and extreme temperature conditions).

[0066] If the vehicle model prioritizes economy, the coefficient corresponding to the energy efficiency optimization weighting coefficient w1 in A can be increased, making the response command more inclined to optimize fuel and electric energy efficiency; if the vehicle model is positioned as high-end comfort, the coefficient corresponding to the comfort assurance weighting coefficient w3 in A can be increased, making the command prioritize ensuring cabin temperature uniformity and power smoothness.

[0067] For extreme low temperature scenarios, the coefficient corresponding to the battery life maintenance weight coefficient w2 in A can be temporarily adjusted to strengthen the weight ratio of battery preheating related instructions.

[0068] The bias term 'b' can handle baseline control requirements independent of dynamic weights. During system initialization, when the vehicle is powered on, the dynamic weights have not yet been calculated in real time. 'b' can provide basic instructions (such as minimum battery insulation power and minimum motor torque) to ensure basic system functionality upon startup. If a sensor malfunction causes abnormal dynamic weight calculation, 'b' can trigger fallback instructions, such as limiting power output to a safe range or maintaining power supply to core loads, to prevent vehicle loss of control due to weight failure and significantly improve the system's fault tolerance.

[0069] In a preferred embodiment of the present invention, the correction coefficient matrix A is reasonably set according to the vehicle requirements; the multi-order dynamic weight coefficient matrix w also includes a weight coefficient w5 corresponding to the current working state of the range extender.

[0070] For example, response instruction allocation matrix Among them, P_batt is the response command sent to the energy system controller, T_motor and P_eng are the response commands sent to the power system controller, and P_ac is the response command sent to the body system controller.

[0071] The correction coefficient matrix A is

[0072] The meanings of each parameter in the correction coefficient matrix A are shown in Table 1.

[0073] Table 1. Meaning of each parameter in the correction coefficient matrix A

[0074] Correction coefficient meaning Principles and basis <![CDATA[0.8(P_batt-w1)]]> The effect of energy efficiency on battery power High energy efficiency allows for larger battery discharge. <![CDATA[-0.2(P_batt-w2)]]> The effect of battery life on battery power suppression Limit discharge power when protecting the battery <![CDATA[0.5(P_batt-w4)]]> The role of safety in boosting battery power More power support is needed in emergencies <![CDATA[0.6(T_motor-w1)]]> The effect of energy efficiency on motor torque Allowing for greater torque at high efficiency <![CDATA[-0.1(T_motor-w3)]]> Comfort's effect on suppressing motor torque To ensure comfort, sudden torque changes must be limited. <![CDATA[0.9(P_ac-w3)]]> The effect of comfort on air conditioning power Increase air conditioning power when high comfort requirements are needed <![CDATA[0.3(P_ac-w4)]]> The effect of safety on air conditioner power Maintain basic temperature control in safe mode <![CDATA[-0.3(P_batt-w5)]]> Range extender's effect on battery power suppression Reduce battery discharge when the range extender is working <![CDATA[0.4(T_motor-w5)]]> The effect of range extenders on motor phase torque Greater torque is allowed when the range is extended. <![CDATA[0.2(P_ac-w5)]]> The role of range extenders in supporting air conditioning power Range extender power supply enhances thermal management capabilities <![CDATA[0.1(P_eng-w1)]]> Energy efficiency's effect on suppressing the start-up of range extenders High energy efficiency prompts range extender intervention <![CDATA[0.4(P_eng-w4)]]> Safety's role in promoting range extender startup Force the range extender to full power in an emergency

[0075] The meaning of the parameters of the bias term b is shown in Table 2.

[0076] Table 2. Parameter meanings of bias term b

[0077]

[0078] In a preferred embodiment of the present invention, the deep reinforcement learning model is deployed in the cloud to receive the current state information of the vehicle and the feedback execution results, and to generate multi-level dynamic weight coefficients.

[0079] Deep reinforcement learning models require processing massive amounts of parameters and complex state spaces, resulting in extremely high computational intensity. Vehicle-side systems, limited by computing power, power consumption, and cost, struggle to support real-time training and inference. Deploying the model in the cloud allows data center-level computing power to quickly calculate multi-order dynamic weight coefficients from vehicle state information, ensuring low latency and high real-time performance in weight generation, and providing timely decision-making support for dynamic collaborative control of the entire vehicle.

[0080] The vehicle only needs to retain lightweight communication functions such as uploading status information and receiving weight coefficients (relying on vehicle networking technologies such as 5G / V2X), without the need to deploy complex AI inference and training hardware, which greatly reduces the hardware cost, power consumption and heat dissipation design pressure of the vehicle, and avoids cost runaway caused by excessive upgrades to vehicle hardware.

[0081] Furthermore, the cloud can aggregate vehicle status information and execution result data from multiple plug-in hybrid and range-extended vehicles of the same type (or different types). Through federated learning or large-scale centralized training, the model can learn a wider range of scenario patterns, avoiding overfitting caused by insufficient data for a single vehicle or a single scenario. This allows the generated multi-level dynamic weight coefficients to accurately adapt to complex and diverse real-world vehicle usage scenarios.

[0082] After the vehicle-side execution results (such as actual energy efficiency, battery SOH changes, comfort experience feedback, and safety incident handling effects) are transmitted back to the cloud, the deep reinforcement learning model can continuously iterate and optimize through the reinforcement learning reward mechanism, adapt to the long-term changes in user driving habits, dynamically match the degradation trend of battery SOH, solve the problem that traditional fixed rule strategies cannot evolve and gradually degrade in long-term performance, and realize the self-learning and self-optimization of control strategies.

[0083] In a preferred embodiment of the present invention, the vehicle subsystem controller includes a powertrain controller, an energy system controller, and a body system controller.

[0084] In one embodiment, the present invention discloses a multi-controller collaborative control device, see [link to relevant documentation]. Figure 2 As shown, the cooperative control device includes:

[0085] The sending module is used to send the acquired vehicle current state information to the deep reinforcement learning model to generate multi-order dynamic weight coefficients;

[0086] The receiving module is used to receive multi-order dynamic weighting coefficients and generate response commands through dynamic coupling.

[0087] The execution module is used to send the response command to the vehicle subsystem controller, which controls the actuator to perform the response action according to the response command and obtains the execution result to feed back to the deep reinforcement learning model.

[0088] In a first aspect, the present invention discloses a vehicle including a processor and a memory, the memory storing a computer program, wherein the processor executes the computer program stored in the memory to implement the steps in the above-described multi-controller cooperative control method.

[0089] For example, see Figure 3 As shown, the vehicle's current state information and execution results are collected by the vehicle's VCU, and then sent to the deep reinforcement learning model in the cloud.

[0090] The following signals will be collected at the vehicle end. If existing signals are missing, high-precision sensors will be installed at relevant locations on the range-extended electric vehicle to collect the signals, including but not limited to:

[0091] Powertrain: Range extender speed, fuel flow rate, exhaust temperature; motor torque and efficiency; battery pack cell voltage, temperature, and total current;

[0092] Safety systems: brake hydraulic pressure; Electronic Stability Program (ESP) acceleration and roll angle;

[0093] Comfort system: multi-point temperature in the cabin, real-time air conditioning load, entertainment system power, cabin and interior lighting brightness and illumination system power;

[0094] Environmental and safety monitoring: tire pressure; vehicle GPS location data and driving speed data.

[0095] A distributed control network centered on the vehicle control unit (VCU) is constructed, comprising a powertrain controller, an energy system controller, and a body system controller connected to the VCU for data transmission. The powertrain controller includes a motor controller (MCU) and a range extender controller (EMS), responsible for dynamic torque adjustment and energy recovery control, as well as range extender start-stop control and operating point optimization. The energy system controller includes a battery management system (BMS), used to execute battery charging and discharging strategies and thermal management control. The body system controller (BCM) manages the power supply to cabin comfort equipment.

[0096] A multi-objective deep reinforcement learning model is deployed on a cloud server, constructing a data processing unit and a weight generation unit. The data processing unit is responsible for receiving, cleaning, and storing a large amount of operational data from range-extended electric vehicles, including battery current, voltage, temperature, depth of charge / discharge, motor power, regenerative braking energy, cabin environmental parameters, braking deceleration, tire pressure, and other information. Based on the processed data, the weight generation unit uses a multi-objective deep reinforcement learning algorithm to generate fourth-order dynamic weight coefficients: energy efficiency optimization weight coefficient w1, battery life maintenance weight coefficient w2, comfort assurance weight coefficient w3, and safety assessment weight coefficient w4.

[0097] Regarding dynamic weight generation and adjustment, the multi-objective deep reinforcement learning model in the cloud server uses millions of range-extended electric vehicle operating condition data as training samples, setting four optimization objectives: energy efficiency optimization, battery life maintenance, driving comfort assurance, and safety risk suppression. Through continuous iterative training, it learns the optimal weight allocation for each objective under different operating conditions, generating initial fourth-order dynamic weight coefficients.

[0098] Then, the initial dynamic weight coefficients are adjusted according to the current operating conditions of the vehicle to obtain multi-order dynamic weight coefficients.

[0099] For example, if the vehicle is currently operating in urban congestion: when the vehicle speed is consistently below 30km / h and there are frequent starts and stops, the cloud system identifies this condition through data analysis and increases the energy efficiency optimization weight coefficient w1 by 0.10-0.20 to reduce unnecessary starts of the range extender; increases the battery life maintenance weight coefficient w2 by 0.05-0.10 to avoid overcharging and discharging of the battery; keeps the comfort assurance weight coefficient w3 at its default value to ensure basic comfort; and dynamically calculates the safety assessment weight coefficient w4 based on real-time road conditions and sensor data.

[0100] The vehicle is currently operating under high-speed cruising conditions: When the vehicle speed is stable at 80-120km / h, the energy efficiency optimization weight coefficient w1 is maintained at a high level, such as accounting for 40% of the total weight, to optimize the engine-motor coordination efficiency; the battery life maintenance weight coefficient w2 is moderately reduced by 0.05-0.08; the comfort assurance weight coefficient w3 is given lower priority and its weight ratio can be adjusted to 0.15-0.20; the safety assessment weight coefficient w4 focuses on monitoring risk signals such as tire pressure and braking system.

[0101] The vehicle is currently operating under low-temperature conditions: When the ambient temperature is below 0°C, the battery life maintenance weight coefficient w2 is increased by 0.15 to 0.20 to strengthen battery preheating management; the comfort assurance weight coefficient w3 is increased by 0.10 to 0.15 to prioritize cabin heating; and the energy efficiency optimization weight coefficient w1 is appropriately reduced by 0.10 to 0.15 to ensure basic functions.

[0102] The vehicle is currently operating under high temperature conditions: when the ambient temperature is above 35℃, the battery life maintenance weight coefficient w2 is used to control battery cooling, increasing by 0.10 to 0.15; the comfort protection weight coefficient w3 balances air conditioning cooling and energy efficiency, with the weight maintained at 0.25 to 0.30; and the safety assessment weight coefficient w4 focuses on monitoring the risk of battery thermal runaway.

[0103] When the safety assessment weight coefficient w4 exceeds 0.3, a risk warning mode is triggered. The vehicle-side VCU readjusts the weight priorities, placing the safety assessment weight coefficient w4 as the highest priority, while adjusting other weights according to the risk level. For example, when the safety assessment weight coefficient w4 ≥ 0.7, the safety assessment weight coefficient w4 forcibly takes over control, immediately cutting off unnecessary load power, limiting motor power, etc., to ensure vehicle safety.

[0104] The lightweight inference engine is a micro inference framework deployed on automotive-grade chips. It can convert dynamic weight parameters in the cloud into executable parameters for the controller, thereby generating corresponding control commands. The specific generation process is as follows: after normalizing the dynamic weight coefficients input from the cloud, calculation is performed through multi-weight dynamic coupling. The coupling calculation results are converted into physical quantities and mapped to controller commands before being sent to each controller for execution.

[0105] Multi-weighted dynamic coupling is a multi-objective optimization decision-making mechanism that simultaneously considers the needs of four dimensions—energy efficiency (w1), lifespan (w2), comfort (w3), and safety (w4)—in complex scenarios, and generates optimal control commands through mathematical modeling.

[0106] After dynamic coupling calculation, an allocation matrix is ​​generated. Following conflict detection and safety boundary detection, this matrix is ​​sent to the execution controllers—namely, the powertrain controller, energy system controller, and body system controller—to convert the matrix calculation results from abstract numerical values ​​into actual engineering parameters. This process considers different hardware characteristics and operating conditions by defining standardized conversion formulas (such as multiplication coefficients and offset additions) and dynamically adjusting the conversion parameters based on equipment type (battery model, motor specifications). Furthermore, real-time environmental data (temperature, altitude) is used to correct the conversion coefficients, ensuring the parameters are adapted to actual operating conditions.

[0107] Based on the functional characteristics of each controller, the engineering parameters are parsed into specific control commands:

[0108] BMS command generation: Calculates charging and discharging current based on battery power and sets current limit thresholds in conjunction with battery state of charge. Generates thermal management commands based on battery temperature to adjust heating / cooling power.

[0109] MCU instruction generation: Converts motor torque demand into control signals, adjusting torque output based on factors such as vehicle speed and gradient. Develops energy recovery strategies, optimizing energy efficiency through regenerative braking level adjustment.

[0110] BCM instruction generation: Set the cooling / heating level according to the air conditioning power requirements, and combine comfort weight with additional functions such as ambient temperature control, seat heating, and steering wheel heating.

[0111] EMS instruction generation: Optimizes the operating range based on the range extender's power requirements, adjusts the control mode, and coordinates with other controller energy sources (MCU, BCM).

[0112] The actual operating data after each controller on the vehicle executes the control command, including but not limited to changes in battery status, motor output power, vehicle load adjustment, and safety risk response effect, are encrypted and transmitted back to the cloud in real time.

[0113] The cloud-based multi-objective deep reinforcement learning model receives data from the vehicle and combines it with the latest operating condition information and control performance evaluation indicators. It then uses optimization algorithms such as gradient descent to iteratively update the weight generation strategy. For example, if the current weight settings lead to excessive battery energy consumption under a certain operating condition, the model will adjust the relevant parameters of w1; if the safety risk response is not timely, the calculation and triggering mechanism of w4 will be optimized. Through continuous iteration, the system can adapt to different operating conditions and user needs, continuously improving the control performance and rationality of the multi-controller collaborative system.

[0114] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. A multi-controller cooperative control method, characterized in that, include: The acquired vehicle current state information is sent to the deep reinforcement learning model to generate multi-order dynamic weight coefficients; It receives multi-order dynamic weighting coefficients and generates response commands through dynamic coupling; The response command is sent to the vehicle subsystem controller, which controls the actuator to perform the response action according to the response command and obtains the execution result to feed back to the deep reinforcement learning model.

2. The multi-controller cooperative control method according to claim 1, characterized in that: The multi-level dynamic weighting coefficients include energy efficiency optimization weighting coefficients, battery life maintenance weighting coefficients, comfort assurance weighting coefficients, and safety assessment weighting coefficients.

3. The multi-controller cooperative control method according to claim 2, characterized in that: The initial dynamic weight coefficients are obtained based on the calculation formula, and then the initial dynamic weight coefficients are adjusted according to the current operating conditions of the vehicle to obtain multi-order dynamic weight coefficients. The formula for calculating the energy efficiency optimization weight coefficient w1 is as follows: a1+a2=1, where F is the fuel consumption rate in L / h; η is the motor efficiency in %. The formula for calculating the battery life maintenance weighting coefficient w2 is as follows: b1+b2+b3=1, where SOH is the state of health of the power battery, in %; N is the number of start-stop cycles of the range extender; and C is the charge-discharge rate. The formula for calculating the comfort protection weighting coefficient w3 is as follows: c1+c2+c3+c4=1, where ΔT is the temperature deviation of different areas of the cabin, in °C; ΔTq is the torque fluctuation rate, in %; Δspd is the speed fluctuation rate, in %; and L is the range extender noise, in dB. The formula for calculating the safety assessment weighting coefficient w4 is as follows: d1+d2+d3=1, where ΔT cell For the temperature difference of the power battery cells, F flag This is the tire pressure abnormality indicator. P represents the brake hydraulic pressure in bar. normal P is the standard pressure for brake hydraulic pressure. max This is the maximum permissible pressure of the brake hydraulic fluid.

4. The multi-controller cooperative control method according to claim 2, characterized in that: In response to the safety assessment weight coefficient exceeding the first threshold, the safety assessment weight coefficient is placed with the highest priority, and the energy efficiency optimization weight coefficient, battery life maintenance weight coefficient, and comfort protection weight coefficient are adjusted according to the risk level. In response to the safety assessment weight coefficient exceeding the second threshold, the safety assessment weight coefficient will forcibly take over control, cut off the power supply to unnecessary loads, and limit the motor power.

5. The multi-controller cooperative control method according to claim 1, characterized in that: The dynamic coupling calculation formula is: M = A × w + b; where M is the response instruction allocation matrix, A is the correction coefficient matrix, w is the multi-order dynamic weight coefficient matrix, and b is the bias term.

6. The multi-controller cooperative control method according to claim 5, characterized in that: The correction coefficient matrix is ​​set appropriately according to vehicle requirements; The multi-order dynamic weight coefficient matrix also includes a weight coefficient w5 corresponding to the current operating state of the range extender.

7. The multi-controller cooperative control method according to claim 1, characterized in that: The deep reinforcement learning model is deployed in the cloud and is used to receive the current state information of the vehicle and the feedback execution results to generate multi-level dynamic weight coefficients.

8. The multi-controller cooperative control method according to claim 1, characterized in that: The vehicle subsystem controller includes a powertrain controller, an energy system controller, and a body system controller.

9. A multi-controller collaborative control device, characterized in that, include: The sending module is used to send the acquired vehicle current state information to the deep reinforcement learning model to generate multi-order dynamic weight coefficients; The receiving module is used to receive multi-order dynamic weighting coefficients and generate response commands through dynamic coupling. The execution module is used to send the response command to the vehicle subsystem controller, which controls the actuator to perform the response action according to the response command and obtains the execution result to feed back to the deep reinforcement learning model.

10. A vehicle comprising a processor and a memory, the memory storing a computer program, characterized in that, The processor executes the computer program stored in the memory to implement the steps in the multi-controller cooperative control method as described in any one of claims 1 to 8.

Images