Hybrid vehicle driving optimization system and method, cloud server and vehicle

Abstract

The invention relates to the technical field of vehicles, in particular to a hybrid power vehicle driving optimization system and method, a cloud server and a vehicle. The system comprises the cloud server which is deployed with a predictive cruise algorithm; the state data of the hybrid power vehicle and the signal lamp state and traffic flow information of the current area where the hybrid power vehicle is located are input into a predictive cruise algorithm, the predictive cruise algorithm outputs speed planning data of the hybrid power vehicle, and the speed planning data are issued to the hybrid power vehicle; the hybrid power vehicle is provided with a real-time energy management model, the state data and the speed planning data are input into the real-time energy management model, the real-time energy management model outputs engine power of the hybrid power vehicle, and the hybrid power vehicle is controlled based on the engine power. Therefore, the problems of poor decision-making and control effects and the like in related technologies are solved.

Description

Technical Field

[0001] The present application relates to the field of vehicle technology, and in particular to a hybrid vehicle driving optimization system, method, cloud server, and vehicle. Background Art

[0002] With the rapid advancement of global industrialization, environmental pollution is becoming increasingly severe. Exhaust emissions from traditional fuel-powered vehicles in the transportation sector have become a significant source of pollution. Hybrid vehicles, leveraging the long driving range of traditional internal combustion engines and the environmentally friendly nature of electric motors, have become a highly sought-after technological solution. To fully leverage the advantages of hybrid vehicles, effective and scientific management and control of their power source are crucial. Speed planning and energy management are closely intertwined, as changes in vehicle speed and acceleration directly impact energy consumption and distribution. Appropriate speed planning can reduce unnecessary acceleration and deceleration, making energy management more efficient. The integration of these two approaches can significantly improve the energy efficiency of hybrid vehicles.

[0003] Most related technologies adopt a multi-objective decoupling architecture for speed planning and energy management, but are limited to a simple sequential integration method, ignoring the coupling relationship between systems, and mainly implementing speed planning and energy management through centralized algorithms. This makes it difficult to adapt to the distributed computing requirements of the vehicle-road-cloud architecture, resulting in poor decision-making and control effects. Summary of the Invention

[0004] The present application provides a hybrid vehicle driving optimization system, method, cloud server and vehicle to solve problems such as poor decision-making and control effects in related technologies.

[0005] A first aspect embodiment of the present application provides a hybrid vehicle driving optimization system, including: a cloud server, the cloud server is deployed with a predictive cruise algorithm, the status data of the hybrid vehicle, the signal light status and traffic flow information of the current area where the hybrid vehicle is located are input into the predictive cruise algorithm, the predictive cruise algorithm outputs the speed planning data of the hybrid vehicle, and the speed planning data is sent down to the hybrid vehicle; a hybrid vehicle, the hybrid vehicle is deployed with a real-time energy management model, the status data and speed planning data are input into the real-time energy management model, the real-time energy management model outputs the engine power of the hybrid vehicle, and the hybrid vehicle is controlled based on the engine power.

[0006] Optionally, the predictive cruise algorithm includes a first action network and a first evaluation network, wherein the state data of the hybrid vehicle, the traffic light state and traffic flow information of the current area where the hybrid vehicle is located are used as state inputs of the first action network, and the first action network outputs the acceleration of the hybrid vehicle at the next moment based on the current state; the first evaluation network evaluates the combined evaluation value of the current state and the acceleration at the next moment, and optimizes the first action network based on the combined evaluation value and the feedback value of the objective function.

[0007] Optionally, the objective function is used to output a reward signal to optimize the first action network and the first evaluation network, and the input of the objective function is the engine power of the hybrid vehicle at the current moment, the status data of the hybrid vehicle, and the traffic light status and traffic flow information of the current area where the hybrid vehicle is located.

[0008] Optionally, the state data includes: SOC (Stage of Charge), vehicle speed, and the real-time energy management model includes a second action network and a second evaluation network, wherein the SOC of the hybrid vehicle, the vehicle speed at the current moment, and the acceleration at the next moment are used as state inputs of the second action network, and the second action network outputs the engine power of the hybrid vehicle based on the current state; the second evaluation network evaluates the combined evaluation value of the current state and the engine power, and optimizes the second action network based on the combined evaluation value and the optimization target.

[0009] Optionally, the real-time energy management model is trained using a deep deterministic policy gradient algorithm, where

[0010] The update strategy of the deep deterministic policy gradient algorithm is:

[0011] θ′=τθ+(1-τ)θ′;

[0012] Among them, θ′ is the target network parameter, θ is the current network parameter, and τ is the soft update coefficient;

[0013] The loss function of the deep deterministic policy gradient algorithm is:

[0014] L=E[(r+γQ′Q′(s′,μ′(s′|θμ′)|θQ′)-Q(s,a|θQ)) 2 ;

[0015] Where L is the loss function, Q′ is the target evaluation network, μ′ is the target action network, E(·) is the mathematical expectation, r is the immediate reward, γ is the discount factor, Q′(s′,μ′(s′|θμ′)|θQ′) is the value estimate of the target evaluation network Q′ for the next state s′ and the action μ′(s′) generated by the target action network μ′, and Q(s,a|θQ) is the value estimate of the current evaluation network for the state s and action a.

[0016] Optionally, the optimization goal of the real-time energy management model is to minimize the fuel consumption of the hybrid vehicle and keep the SOC of the hybrid vehicle within a target range.

[0017] A second aspect of the present application provides a hybrid vehicle driving optimization method, which is applied to a cloud server and includes the following steps: obtaining status data of the hybrid vehicle and the signal light status and traffic flow information of the current area where the hybrid vehicle is located; inputting the status data, the signal light status and traffic flow information of the current area into a predictive cruise algorithm, and the predictive cruise algorithm outputs speed planning data of the hybrid vehicle; sending the speed planning data to the hybrid vehicle, wherein the hybrid vehicle determines the engine power of the hybrid vehicle based on the speed planning data, and controls the hybrid vehicle based on the engine power.

[0018] The third aspect of the present application provides a hybrid vehicle driving optimization method, which is applied to a hybrid vehicle and includes the following steps: uploading status data of the hybrid vehicle; receiving speed planning data of the hybrid vehicle sent by a cloud server; inputting the status data and speed planning data into a real-time energy management model, the real-time energy management model outputs the engine power of the hybrid vehicle, and controlling the hybrid vehicle based on the engine power.

[0019] The fourth aspect of the present application provides a cloud server, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to execute the hybrid vehicle driving optimization method as described in the above embodiment.

[0020] The fifth aspect of the present application provides a hybrid vehicle, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the hybrid vehicle driving optimization method as described in the above embodiment.

[0021] Therefore, this application has at least the following beneficial effects:

[0022] In the embodiment of the present application, the status data of the hybrid vehicle, the status of the traffic lights in the current area where the hybrid vehicle is located, and the traffic flow information can be input into the predictive cruise algorithm deployed on the cloud server. The predictive cruise algorithm outputs the speed planning data of the hybrid vehicle, which is sent to the hybrid vehicle. The hybrid vehicle inputs the speed planning data sent by the cloud server and the status data of the hybrid vehicle into the deployed real-time energy management model. The real-time energy management model outputs the engine power of the hybrid vehicle, and controls the hybrid vehicle based on the engine power. Through the hierarchical architecture of collaborative control between the cloud server and the vehicle side, the predictive cruise algorithm is deployed on the cloud side for speed planning and instruction distribution, and the real-time energy management is deployed on the vehicle side for power distribution. The complex collaborative optimization problem is decomposed into two sub-problems: energy management and speed planning, which significantly reduces the computational complexity. The powerful computing resources of the cloud server can make up for the performance bottlenecks of the vehicle side and the road side when processing complex algorithms, improve computing efficiency, and ensure the effectiveness of decision-making and control. Thus, the technical problem of poor decision-making and control effects in the related art is solved.

[0023] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0025] Figure 1 A framework diagram of a hybrid vehicle driving optimization system provided according to an embodiment of the present application;

[0026] Figure 2 This is an architectural diagram of a power-split hybrid vehicle provided according to an embodiment of the present application;

[0027] Figure 3 A schematic diagram of a vehicle simulation model provided according to an embodiment of the present application;

[0028] Figure 4 This is a diagram of a battery equivalent circuit structure provided according to an embodiment of the present application;

[0029] Figure 5 A schematic diagram of a characteristic curve of the open circuit voltage of a battery provided according to an embodiment of the present application;

[0030] Figure 6 A schematic diagram of a charge-discharge internal resistance characteristic curve of a battery provided according to an embodiment of the present application;

[0031] Figure 7Schematic diagram of a collaborative optimization framework for a two-layer DDPG (Deep Deterministic Policy Gradient) algorithm according to an embodiment of the present application;

[0032] Figure 8 A flowchart of generating a real-time energy manager for a hybrid vehicle according to an embodiment of the present application;

[0033] Figure 9 Schematic diagram of the intelligent connected vehicle network domain architecture provided according to an embodiment of the present application;

[0034] Figure 10 Schematic diagram of data transmission provided according to an embodiment of the present application;

[0035] Figure 11 An activity diagram of an algorithm for predictive cruise economy driving according to an embodiment of the present application;

[0036] Figure 12 A flowchart of a hybrid vehicle driving optimization method provided according to an embodiment of the present application;

[0037] Figure 13 The present invention is a flowchart of a hybrid vehicle driving optimization method provided according to another embodiment of the present application. DETAILED DESCRIPTION

[0038] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0039] Before describing the solution of this application, we first introduce the related technologies to better understand the solution.

[0040] Currently, hybrid vehicle energy management strategies can be primarily categorized into three types: rule-based, optimization-based, and learning-based. While rule-based energy management strategies are simple and practical, they have limited adaptability. Optimization-based methods optimize control variables within operational constraints using predefined cost functions. These methods can be further categorized into global optimization methods (such as dynamic programming, genetic algorithms, particle swarm optimization, and the Pontryagin minimum principle) and instantaneous optimization methods (such as equivalent fuel consumption minimization strategies, game theory, and model predictive control). While global optimization strategies can achieve optimal solutions, they rely on complete driving cycle information, making real-time implementation difficult. In contrast, instantaneous optimization methods reduce reliance on global information and enable real-time control, but they cannot guarantee global optimality.

[0041] Furthermore, environmental complexity and uncertainty significantly impact the performance of energy management systems, requiring coordination between powertrain dynamics and driving behavior. This has led to the emergence of comprehensive approaches that integrate advanced driver assistance systems with energy management strategies. Speed planning, in particular, exhibits a natural synergy with hybrid vehicle energy management by precisely controlling speed and acceleration, as vehicle dynamics directly influence energy consumption and allocation patterns.

[0042] Research on predictive cruise control primarily relies on onboard perception systems or vehicle-road collaboration to acquire traffic environment information. However, these technologies have significant limitations in communication distance, data latency, and sensing range, making it difficult to provide vehicles with sufficient long-range traffic environment information. This limited sensing range makes it difficult for vehicles to plan ahead for long-range road conditions, particularly in complex road networks, reducing the effectiveness of energy-efficient driving.

[0043] The integration of speed planning and energy management can bring significant energy-saving advantages. However, most current solutions adopt a multi-objective decoupled architecture for speed planning and energy management. Although fuel economy and driving comfort are improved through rule-based energy management algorithms, these studies are limited to a simple sequential integration approach. In this mode, the speed planning module and the energy management module operate relatively independently, with the former focusing on optimizing the vehicle's dynamic following performance and the latter implementing power allocation based solely on the planned speed. Although energy efficiency can be improved by introducing energy consumption indicators into the speed planning objective function, the inherent nonlinear characteristics of the plug-in hybrid vehicle powertrain mean that the optimal acceleration profile determined solely from driving conditions often fails to meet the actual operating requirements of the energy management system. Although this separate optimization method achieves basic functional integration, it ignores the basic coupling relationship between systems, ultimately resulting in suboptimal control performance.

[0044] The related technologies have the following problems:

[0045] 1. Inherent limitations of hierarchical decoupling strategies. Related technologies often use a hierarchical decoupling approach to integrate speed planning and energy management. Specifically, the speed planning module is responsible for executing speed planning, while the energy management module implements power allocation based on the planned speed. While this separate optimization approach achieves basic functional integration, it neglects the coupling relationship between the two systems, resulting in suboptimal control performance. This is particularly true in highly nonlinear powertrains such as plug-in hybrid vehicles, where optimal acceleration determined solely from driving conditions often fails to meet the actual operational requirements of the energy management system.

[0046] 2. Computational complexity. To address the limitations of hierarchical decoupling strategies, some studies have proposed collaborative optimization strategies based on methods such as model predictive control. However, in the process of hybrid electric vehicles following a vehicle, the economic optimization of speed planning and energy management involves a complex set of ordinary differential equations and nonlinear constraints. These nonlinear and non-convex characteristics transform the problem into an NP-hard problem that cannot be solved in polynomial time. Although these integrated control strategies can minimize energy consumption while ensuring safety and SOC constraints, the huge computational burden generated by their simultaneous optimization process severely limits their implementation potential in practical systems.

[0047] 3. Limitations in Perception Range. Current in-vehicle perception systems and vehicle-road cooperative technologies have significant limitations in communication distance, data latency, and perception range, making it difficult to provide vehicles with sufficient long-range traffic environment information. This limited perception range makes it difficult for vehicles to plan ahead for long-range road conditions, reducing the effectiveness of energy-efficient driving. This problem is particularly prominent in complex road network environments.

[0048] 4. Limited training scenarios. While collaborative optimization methods based on deep reinforcement learning have achieved significant breakthroughs in reducing computational load, their training scenarios are still limited to simplified vehicle-following conditions and are incapable of handling complex real-world driving environments. Furthermore, the complexity of the fusion architecture poses significant challenges to system debugging and maintenance, severely restricting the rapid optimization and adaptive adjustment capabilities of each module.

[0049] 5. The conflict between distributed computing requirements and centralized algorithms. While related technologies can achieve theoretically globally optimal fusion optimization methods for speed planning and energy management, these centralized algorithms struggle to adapt to the distributed computing requirements inherent in vehicle-road-cloud architectures. In particular, research based on intelligent vehicle-road cooperative systems, over-reliance on vehicle-road communication and infrastructure perception, has shown significant limitations in both information exchange scope and environmental perception capabilities.

[0050] To this end, the present application provides a hybrid vehicle driving optimization system, which can input the status data of the hybrid vehicle, the signal light status and traffic flow information of the current area where the hybrid vehicle is located into the predictive cruise algorithm deployed on the cloud server. The predictive cruise algorithm outputs the speed planning data of the hybrid vehicle, and sends the speed planning data to the hybrid vehicle. The hybrid vehicle inputs the speed planning data sent by the cloud server and the status data of the hybrid vehicle into the deployed real-time energy management model. The real-time energy management model outputs the engine power of the hybrid vehicle, and controls the hybrid vehicle based on the engine power. Through the hierarchical architecture of collaborative control between the cloud server and the vehicle side, the predictive cruise algorithm is deployed on the cloud for speed planning and command distribution, and the real-time energy management is deployed on the vehicle side for power distribution. The complex collaborative optimization problem is decomposed into two sub-problems of energy management and speed planning, which significantly reduces the computational complexity. The powerful computing resources of the cloud server can make up for the performance bottlenecks of the vehicle side and the road side when processing complex algorithms, improve computing efficiency and ensure the decision-making and control effects.

[0051] Specifically, Figure 1 A schematic diagram of a hybrid vehicle driving optimization system provided in an embodiment of the present application.

[0052] like Figure 1 As shown, the hybrid vehicle driving optimization system 10 includes: a cloud server 11 and a hybrid vehicle 12 .

[0053] Among them, the cloud server 11 is deployed with a predictive cruise algorithm, which inputs the status data of the hybrid vehicle 12, the traffic light status and traffic flow information of the current area where the hybrid vehicle 12 is located into the predictive cruise algorithm, and the predictive cruise algorithm outputs the speed planning data of the hybrid vehicle 12, and sends the speed planning data to the hybrid vehicle 12; the hybrid vehicle 12 is deployed with a real-time energy management model, which inputs the status data and speed planning data into the real-time energy management model, and the real-time energy management model outputs the engine power of the hybrid vehicle 12, and controls the hybrid vehicle 12 based on the engine power.

[0054] The status data of the hybrid vehicle of the present application includes SOC and vehicle speed, which can be obtained through on-board sensors; the signal light status includes traffic light information, traffic flow information includes preceding vehicle information, traffic flow information, etc.; speed planning data includes vehicle speed, etc.; speed planning data includes vehicle speed and acceleration at the next moment, etc.

[0055] It can be understood that the embodiment of the present application constructs a hybrid vehicle driving optimization system 10, which can input the status data of the hybrid vehicle 12, the signal light status and traffic flow information of the current area where the hybrid vehicle 12 is located, into the predictive cruise algorithm deployed by the cloud server 11. The predictive cruise algorithm outputs the speed planning data of the hybrid vehicle 12, and sends the speed planning data to the hybrid vehicle 12. The hybrid vehicle 12 inputs the speed planning data sent by the cloud server 11 and the status data of the hybrid vehicle 12 into the deployed real-time energy management model. The real-time energy management model outputs the engine power of the hybrid vehicle 12, and controls the hybrid vehicle 12 based on the engine power. Through the hierarchical architecture of collaborative control between the cloud server and the vehicle side, the predictive cruise algorithm is deployed on the cloud side for speed planning and command distribution, and the real-time energy management is deployed on the vehicle side for power distribution. The complex collaborative optimization problem is decomposed into two sub-problems: energy management and speed planning, which significantly reduces the computational complexity. The powerful computing resources of the cloud server can make up for the performance bottlenecks of the vehicle side and the road side when processing complex algorithms, improve computing efficiency and ensure the decision-making and control effects.

[0056] In addition, it should be noted that the traffic light status and traffic flow information of the embodiment of the present application can be obtained through the cloud platform, and the beyond-visual-range perception data can be obtained through the multi-source information fusion capability of the cloud platform, breaking through the limitations of the perception range of a single vehicle and realizing accurate perception and optimized control of complex road network environments.

[0057] In an embodiment of the present application, the predictive cruise algorithm includes a first action network and a first evaluation network, wherein the state data of the hybrid vehicle 12, the signal light state and traffic flow information of the current area where the hybrid vehicle 12 is located are used as the state input of the first action network, and the first action network outputs the acceleration of the hybrid vehicle 12 at the next moment based on the current state; the first evaluation network evaluates the combined evaluation value of the current state and the acceleration at the next moment, and optimizes the first action network according to the combined evaluation value and the feedback value of the objective function.

[0058] It can be understood that the predictive cruise algorithm of the embodiment of the present application includes a first action network and a first evaluation network. The first action network (Actor network) is used to determine the action based on the input data, and the first evaluation network (Critic network) is used to evaluate the combined evaluation value of the current state and action, and optimize the first action network according to the combined evaluation value and the feedback value of the objective function.

[0059] In an embodiment of the present application, the objective function is used to output a reward signal to optimize the first action network and the first evaluation network. The input of the objective function is the engine power of the hybrid vehicle 12 at the current moment, the status data of the hybrid vehicle 12, and the traffic light status and traffic flow information of the current area where the hybrid vehicle 12 is located.

[0060] Specifically, the predictive cruise algorithm at the lower level of the embodiment of the present application uses spatiotemporal information such as vehicle status, surrounding vehicle status, and traffic signals as state input. The Actor network outputs the desired speed, and the Critic network evaluates the Q value of the current state-action combination and calculates the acceleration at the next moment based on the current state. Combined with the current SOC value, this information is input into the trained energy management model at the upper level. The engine power output by the energy management model, together with the vehicle status and environmental status, constitutes the objective function, which serves as the reward signal for the predictive cruise algorithm at the lower level. This architectural design of sequential optimization of the upper and lower levels not only ensures the independent optimization of energy management and speed planning, but also forms a closed-loop mechanism through the feedback evaluation of the upper model by the lower level, thereby achieving effective coordination between the two subsystems.

[0061] In an embodiment of the present application, the real-time energy management model includes a second action network and a second evaluation network, wherein the SOC of the hybrid vehicle 12, the vehicle speed at the current moment, and the acceleration at the next moment are used as state inputs of the second action network, and the second action network outputs the engine power of the hybrid vehicle 12 based on the current state; the second evaluation network evaluates the combined evaluation value of the current state and the engine power, and optimizes the second action network based on the combined evaluation value and the optimization target.

[0062] The optimization goal of the real-time energy management model is to minimize the fuel consumption of the hybrid vehicle and keep the SOC of the hybrid vehicle within the target range.

[0063] It can be understood that the real-time energy management model of the embodiment of the present application includes a second action network and a second evaluation network. The SOC of the hybrid vehicle, the vehicle speed at the current moment, and the acceleration at the next moment are used as state inputs of the second action network. The second action network outputs the engine power of the hybrid vehicle based on the current state. The second evaluation network evaluates the combined evaluation value of the current state and the engine power, and optimizes the second action network based on the combined evaluation value and the optimization target.

[0064] Specifically, the real-time energy management model of the embodiment of the present application uses the vehicle's speed, acceleration and SOC as state inputs to the second action network. The second action network outputs the engine power of the hybrid vehicle based on the current state. The second evaluation network is used for the combined evaluation value of the current state and the engine power. The second action network is optimized based on the combined evaluation value and the optimization target.

[0065] In the embodiment of the present application, the real-time energy management model is trained using a deep deterministic policy gradient algorithm, wherein:

[0066] The update strategy of the deep deterministic policy gradient algorithm is:

[0067] θ′=τθ+(1-τ)θ′; (1)

[0068] Among them, θ′ is the target network parameter, θ is the current network parameter, and τ is the soft update coefficient;

[0069] The loss function of the deep deterministic policy gradient algorithm is:

[0070] L=E[(r+γQ′Q′(s′,μ′(s′|θμ′)|θQ′)-Q(s,a|θQ)) 2 ; (2)

[0071] Where L is the loss function, Q′ is the target evaluation network, μ′ is the target action network, E(·) is the mathematical expectation, r is the immediate reward, γ is the discount factor, Q′(s′,μ′(s′|θμ′)|θQ′) is the value estimate of the target evaluation network Q′ for the next state s′ and the action μ′(s′) generated by the target action network μ′, and Q(s,a|θQ) is the value estimate of the current evaluation network for the state s and action a.

[0072] Specifically, the DDPG algorithm in the embodiments of this application is a deep reinforcement learning method that cleverly combines deep learning with deterministic policy gradients to provide an effective solution to decision-making problems in continuous action spaces. As a representative algorithm of the actor-critic architecture, DDPG adopts a dual network structure: the actor network is responsible for directly outputting deterministic actions, while the critic network is dedicated to evaluating the value of actions.

[0073] During the training of the real-time energy management model, DDPG introduces several innovative mechanisms to improve learning efficiency and stability. First, it draws on the experience replay technology of DQN to build an experience replay pool to store transfer samples (s, at, rt, st+1), breaking the temporal correlation between samples. Second, the algorithm adopts a target network mechanism to establish a slow-updating target network for the actor and the critic respectively. Through soft updates, the main network parameters are gradually tracked, effectively suppressing the excessive bias of value estimation.

[0074] This "soft update" strategy can be expressed as:

[0075] θ′=τθ+(1-τ)θ′; (1)

[0076] In terms of policy updates, DDPG uses deterministic policy gradients to directly maximize the gradient of the Q value with respect to the policy parameters. The critic network is optimized by minimizing the temporal difference error, and its loss function is:

[0077] L=E[(r+γQ′Q′(s′,μ′(s′|θμ′)|θQ′)-Q(s,a|θQ)) 2 ;(2)

[0078] Where Q′ is the target critic network and μ′ is the target actor network. To enhance exploration capabilities, DDPG adds Ornstein-Uhlenbeck noise to actions during training. This time-dependent noise process can produce more coherent exploration behavior.

[0079] Due to the independent action network, the DDPG algorithm can map states to a set of deterministic continuous action sequences, which makes DDPG suitable for problems with continuous state and action spaces. The specific process of the DDPG algorithm is as follows:

[0080] 1. Initialization: The weights of the Critic network and the Actor network are θ Q and θ μ , the target network Q′ and μ′, the weights are μ Q′ and θ μ′ , replay experience pool R, and random process N are used for action exploration.

[0081] 2. From episode 1 to episode M, perform the following steps for each episode:

[0082] 2.1 Obtain the initial state of the hybrid vehicle: initial speed v1, initial acceleration acc1, initial battery state of charge SoC1.

[0083] 2.2 In each episode, from the 1st time step to the Tth time step, perform the following steps for each time step t:

[0084] 2.2.1 According to the current strategy (by the Actor network according to the current state and weight θ μ generated action) and exploration noise Nt to select the action: at=μ(st|θμ)+Nt.

[0085] 2.2.2 Execute action at and observe the reward rt obtained by the hybrid vehicle and the new state st+1 it transfers to.

[0086] 2.2.3 Transfer samples (st, at, rt, st+1) and store them in the experience replay pool R.

[0087] 2.2.4 Sample a small batch of transfer samples (st, at, rt, st+1) from the experience replay pool R, using prioritized experience replay.

[0088] 2.2.5 Set target value: y i =r i +γQ′(st+1,μ′(st+1|θμ′)|θQ′).

[0089] 2.2.6 Update the Critic network by minimizing the loss function: L = 1 / N∑(y i -Q(s i ,a i |θQ)) 2 .

[0090] 2.2.7 Update the Actor network policy using sampled policy gradients:

[0091]

[0092] 2.2.8 Update target network:

[0093] θ Q′ ←τθ Q +(1-τ)θ Q′ ;

[0094] θ μ′ ←τθ μ +(1-τ)θ μ′ .

[0095] The following describes the hybrid vehicle part of this application, the construction process of the real-time energy management model and how to realize the generator power distribution process through a specific embodiment. This application adopts a power split hybrid vehicle, and its architecture is as follows: Figure 2 shown.

[0096] This application uses the on-board sensors installed in networked modified vehicles to obtain real-time vehicle operation data, and transmits information such as speed, acceleration, and battery state of charge (SOC) to the cloud in real time via the on-board communication terminal (TBOX). This application uses the DDPG algorithm to build an energy management controller, using the speed sequence, acceleration sequence, and current SOC under multiple standard working conditions as the input state quantity of the strategy network to output the optimal power distribution of the engine. The controller takes minimizing comprehensive fuel consumption as its optimization goal, while ensuring that the battery SOC value remains within a reasonable range, achieving real-time and efficient energy management.

[0097] Given that predictive cruise control primarily focuses on the vehicle's longitudinal driving characteristics, this study only models the vehicle's longitudinal dynamics and does not consider lateral dynamics. During vehicle motion, longitudinal traction must overcome four main types of resistance: rolling resistance caused by tire-ground contact, air resistance from the vehicle body, grade resistance caused by hilly driving, and acceleration resistance. These resistances are calculated using the following formulas:

[0098]

[0099] Where m represents the mass of the car (kg); g represents the acceleration due to gravity (m / s 2 ); f represents the rolling resistance coefficient; α represents the road slope angle; C d represents the air resistance coefficient; A represents the frontal area (m 2 ); ρ represents the air density (kg / m 3 ); v represents the driving speed (m / s); δ represents the rotational mass conversion coefficient.

[0100] Then the longitudinal dynamic equation is:

[0101] F d =F f +F w +F i +F j ;(4)

[0102] Simulation methods are categorized as forward and backward, depending on whether the simulation data and vehicle power transmission directions are consistent. During the simulation process, information transmission follows the same direction as power transmission, meaning both information and power are transferred from the driver model to the powertrain and finally to the dynamics model. This is called a forward simulation method.

[0103] This application adopts the backward simulation method, which calculates the required power by using the vehicle dynamics model and wheel model to calculate the target vehicle speed information of the driving condition. The vehicle control system determines the vehicle driving mode and power source distribution strategy according to the energy distribution strategy, and then solves the actual vehicle speed through the power output of the corresponding power source. The vehicle simulation model is as follows Figure 3 shown.

[0104] Using the vehicle dynamics parameters and power transmission ratio, the vehicle speed is mapped to the drive motor speed and torque. The mathematical model expression is:

[0105] nm2=v*60*2*pi*r*i m ;(5)

[0106] Tm2=F d *r*eff_t-T d *k p / (1+k p) / i m ;(6)

[0107] Where, nm2 represents the speed of the driving motor (r / min); v represents the driving speed (m / s); pi represents the circumference of the circle; r represents the rolling radius of the wheel; i m Indicates the motor reduction ratio; Tm2 indicates the output torque of the drive motor (N*m); F d Indicates the required wheel torque (N*m); eff_t indicates the efficiency of the transmission system; i m Indicates the motor reduction ratio; T d Indicates the engine output torque (N*m); k p is the characteristic parameter of the planetary gear.

[0108] Based on the obtained parameters of the engine and drive motor, the parameters of the generator can be calculated. The mathematical model expression is as follows:

[0109] nm1=(1+k p )*T n -k p *nm2*i0;(7)

[0110] Tm1=-T s / (1+k p );(8)

[0111] Where, nm1 represents the generator speed (r / min); T n represents the engine speed (r / min); i0 represents the main drive ratio; Tm1 represents the generator torque (N*m).

[0112] To facilitate simulation calculations, the engine fuel consumption rate curve is normalized and fitted with a two-dimensional quintic polynomial:

[0113]

[0114] Among them, b e is the fuel consumption rate, g / kWh; A 1 / 2(i+1)(i+2)-i-1+j are the coefficients of each item, T e 、T e_max is the engine output torque and maximum torque (N*m); n e 、n e_max is the engine output speed and maximum speed (r / min).

[0115] Points can be obtained, and the fuel consumption per unit time is:

[0116]

[0117] Among them, B e is the fuel consumption (g / s); Pe is the engine output power (kWh).

[0118] Similarly, in order to facilitate simulation calculations, two-dimensional quintic polynomial fitting is performed on the working efficiency of the generator and drive motor respectively:

[0119]

[0120] Among them, η MG1 is the efficiency of motor No. 1; C 1 / 2(i+1)(i+2)-i-1+j are the coefficients of each item, T MG1 、T MG1_max The output torque and maximum torque of motor No. 1 (N*m); n MG1 、n MG1_max is the output speed and maximum speed of motor No. 1 (r / min); η MG2 is the efficiency of motor No. 2; B 1 / 2(i+1)(i+2)-i-1+j are the coefficients of each item, T MG2 、T MG2_max The output torque and maximum torque of the second motor (N*m); n MG2 、n MG2_max The output speed and maximum speed of motor No. 1 (r / min).

[0121] The power battery modeling methods are mainly divided into two categories: electrochemical model and equivalent circuit model. The electrochemical model is modeled by describing the internal reaction mechanism of the battery, but it is highly complex and difficult to guarantee accuracy. This paper adopts the equivalent circuit model, based on the quasi-static assumption, ignoring the influence of the electrochemical reaction process and environmental factors, and describes the input and output characteristics of the battery in the form of an ideal voltage source and a resistor in series. Figure 4 As shown, Figure 4 This is the battery equivalent circuit structure diagram. The model includes open-circuit voltage Voc (V), internal resistance R (Ω), current I (A), and terminal voltage V (V). The positive and negative values of the current correspond to discharge and charge conditions, respectively.

[0122] According to the equivalent circuit diagram, by Kirchhoff's law, the load voltage can be expressed as

[0123] U=U oc -I*R int ; (13)

[0124] Without considering the influence of external conditions such as temperature, it can be considered that the open circuit voltage is only affected by the change of the power battery SOC, while the internal resistance is mainly affected by the change of the power battery SOC and the charge and discharge state. Through charge and discharge experimental tests, the characteristic curves of open circuit voltage and internal resistance under different battery SOC can be obtained. The characteristic curve of open circuit voltage is as follows: Figure 5 As shown, the battery charge and discharge internal resistance characteristic curve is as follows Figure 6 shown.

[0125] Based on the load voltage and current, the power battery output power can be expressed as:

[0126] P bat =U*I;(14)

[0127] According to the ampere-hour integration method, the battery SOC change can be expressed as:

[0128]

[0129] Among them, SOC(t) and SOC(t0) are the battery SOC values at the end time and the start time respectively, Q nom is the power battery capacity (As).

[0130] By combining formulas (11), (12) and (13), the formula for calculating the battery SOC change is:

[0131]

[0132] Regarding the life attenuation factors of hybrid vehicle power batteries, studies have shown that for batteries of the same capacity, the greater the ampere-hour flux, the greater the attenuation. In addition, for high-discharge rate batteries, the discharge rate is the main factor affecting the battery life attenuation rate. Therefore, the degree of battery loss can be characterized by the battery's effective ampere-hour flux, using the formula:

[0133]

[0134] Where Ah represents the effective ampere-hour flux of the battery (As); σ represents the battery working intensity coefficient related to the charge and discharge rate, which can be estimated according to the empirical formula:

[0135]

[0136] Among them, I c It is the charge and discharge rate, which represents the ratio of the battery operating current to the battery capacity.

[0137] According to the mechanical performance limitations of the actual configuration power components, each power component must meet the following constraints:

[0138]

[0139] Among them, n e_min 、n e_max 、n MG1_min 、n MG1_max 、n MG2_min 、n MG2_max are the minimum and maximum speeds of the engine, motor 1 and motor 2, r / min, T e_min 、T e_max 、TMG1_min 、T Mg1_max 、T MG2_min 、T MG2_max Respectively, they are the minimum and maximum torque (KW), which are related to the current working speed; SOC min , SOC max The minimum and maximum values of the power battery SOC set by the system; P bat_min 、P bat_max It is the minimum and maximum power of the power battery (KW).

[0140] The hybrid vehicle driving optimization system of the present application is described below through specific embodiments.

[0141] This application proposes a collaborative optimization framework based on a two-layer DDPG, such as Figure 7 As shown in the figure, at the upper level, a deep DDPG algorithm is used to train a real-time energy management model. This model uses the vehicle's speed, acceleration, and SOC value as state inputs, and uses fuel consumption and SOC loss as reward functions. It outputs engine power through an Actor-Critic network structure, where the initial strategy of the Actor network is derived from the BSFC optimal working curve. After training, the model can output an optimized energy management strategy in real time. The flow chart of the upper-level hybrid vehicle generating a real-time energy manager is shown in the figure. Figure 8 shown.

[0142] The lower-level DDPG algorithm deploys the real-time energy management model trained in the upper layer to achieve coordinated control of speed planning and energy management. The lower-level algorithm uses spatiotemporal information such as vehicle status, surrounding vehicle status, and traffic signals as state input. The Actor network outputs the desired speed, and the Critic network evaluates the Q-value of the current state-action combination. The system calculates the acceleration at the next moment based on the current state and, combined with the current SOC value, inputs this information into the trained energy management model in the upper layer. The engine power output of this model, along with the vehicle and environmental conditions, constitutes the objective function and serves as the reward signal for the lower-level DDPG algorithm. This sequential optimization architecture ensures independent optimization of energy management and speed planning while also forming a closed-loop mechanism through feedback evaluation from the lower layer to the upper-level model, enabling effective coordination between the two subsystems.

[0143] Specifically, this application also proposes a network domain architecture for intelligent connected vehicles such as Figure 9 As shown in the figure, it mainly consists of three core components: the vehicle-side, the cloud server, and its communication network. This architecture adopts a three-tiered deployment model of "cloud-edge-end", enabling multi-level data exchange through the 5G network and C-V2X dual channels. The cloud server needs to integrate traffic light status and traffic flow information from multiple intersections.

[0144] At the network communication level, a diversified communication network system has been built. The 5G network provides wide-area coverage, supports remote communication between vehicles and edge clouds, regional clouds, and central clouds, and realizes vehicle-platform-vehicle and vehicle-platform-road information services; the C-V2X direct communication network is responsible for key area coverage, supports short-range communication between vehicle-vehicle, vehicle-road and other end-side devices, and ensures the real-time and efficient broadcasting of key information such as traffic signals and roadside perception. In addition, the road-cloud transmission network realizes the interconnection of roadside equipment and data exchange with cloud servers through optical fiber or 5G backhaul, and data transmission such as Figure 10 shown.

[0145] The regional cloud, serving as the data aggregation center for multiple edge clouds, plays the crucial role of coordinating edge clouds at adjacent intersections. Each edge cloud at each intersection collects real-time traffic dynamics information for its region and uploads it to the regional cloud for unified scheduling. Specifically, the edge cloud in the area serving the vehicle can access signal phases and traffic flow data at multiple intersections ahead through the regional cloud. Combined with supporting platforms such as ground-based augmented positioning systems and high-precision dynamic maps, the edge cloud employs the DDPG algorithm for real-time speed optimization calculations. This algorithm takes as input multi-dimensional information such as signal timing at multiple intersections, vehicle status, and traffic flow, and solves for the optimal speed trajectory through deep reinforcement learning modeling.

[0146] In terms of the vehicle-side architecture, the vehicle networking control unit (T-BOX) serves as a core component, interconnecting with modules such as the vehicle's drivetrain, chassis, safety, body, and sensors via the CAN bus. Based on this architecture, the system supports multiple functions, including real-time energy management, CPCC control, ADAS assistance, and intelligent navigation. It also provides the driver with a user-friendly human-machine interface through intelligent HMI and smart instruments, enabling the collection, processing, and display of vehicle-side data, as well as two-way data exchange with the cloud.

[0147] In the speed planning of lower-level economic driving, such as Figure 11 As shown, Figure 11 The activity diagram for the underlying predictive cruise control algorithm is based on an actor-critic architecture. The DDPG algorithm employed is responsible for generating speed planning strategies, while the critic network evaluates the state-action value function. The experience replay mechanism and soft updates of the target network effectively improve the algorithm's convergence and learning efficiency. The planned speed, acceleration, and current state of charge (SOC) value are input into the hybrid vehicle's real-time energy management model to obtain engine power output. Fuel consumption and battery SOC changes are calculated based on the vehicle dynamics model, and the reward value of the current state-action pair is then evaluated.

[0148] The defined reward function comprehensively considers multiple key factors: vehicle fuel economy, time efficiency at traffic lights, driving comfort (by limiting acceleration changes), maintaining a safe distance from the vehicle ahead, and maintaining the battery SOC value within a reasonable range. This multi-objective optimization ensures the comprehensive performance of the system in actual operation.

[0149] Ultimately, the optimized speed command is sent in real time to the service vehicle via the roadside unit, enabling economical cruise control. This intelligent control architecture, based on cloud-edge-vehicle collaboration, not only ensures real-time system performance but also significantly improves the economy and reliability of predictive cruise control, providing a new solution for the development of future intelligent transportation systems.

[0150] In addition, it should be noted that in order to avoid excessive differences between vehicle data and actual road conditions, this application uses the intelligent driver model (IDM) proposed by Treiber to construct a basic traffic flow diagram. The IDM model, with its streamlined parameters and clear physical meaning, can accurately depict the traffic state transition process from free flow to blocked flow. Studies have shown that the basic traffic flow diagram constructed based on the IDM model shows good consistency with the measured data, whether in highway or urban road scenarios. Its mathematical model is expressed as follows:

[0151]

[0152] in, is the acceleration of the vehicle at time t; v0 is the desired speed of the vehicle, which is usually the speed limit of the road; v i (t) is the speed of the vehicle at time t; s0 is the minimum safe stopping distance; Δv i (t) is the speed difference between the vehicle and the preceding vehicle at time t; T is the safe headway; h i (t) is the headway between the vehicle and the preceding vehicle at time t; b is the comfortable deceleration; a is the maximum acceleration parameter; l is the vehicle length; and δ is the acceleration index.

[0153] In summary, this application has the following effects:

[0154] 1. This application, based on a cloud-based control platform architecture, breaks through the limitations of perception range and communication distance of traditional vehicle-infrastructure collaborative technologies for urban predictive cruise control. The introduction of a cloud platform not only expands the system's information acquisition capabilities but also, through its powerful computing resources, overcomes performance bottlenecks in processing complex algorithms on both the vehicle and road sides, providing strong support for more efficient predictive control.

[0155] 2. This application proposes a two-layer real-time optimization architecture, innovatively applying the DDPG algorithm to address real-time energy management. DDPG trains a fast-responding real-time energy manager at the upper layer and embeds it into the speed planning for economical driving at the lower layer, serving as a key basis for evaluating the lower-layer economical driving maneuvers. This sequential optimization approach of predictive cruise control at the lower layer and energy management strategy at the upper layer ensures independent optimization of each subsystem while also forming a closed-loop evaluation mechanism through feedback from the lower layer, guaranteeing the effectiveness of the strategy.

[0156] 3. This application designs a real-time response safe and economical speed logic switching mechanism to achieve energy optimization while ensuring driving safety. Relying on a trained real-time energy management model, this mechanism can quickly adapt to dynamic traffic environments and output optimization strategies, effectively overcoming the local optimality problem of single intersection planning in traditional methods. This two-layer architecture based on real-time computing not only improves the robustness of predictive cruise control, but also realizes real-time coordinated optimization of energy management and speed planning, providing an efficient and feasible technical solution for energy-saving control of intelligent connected vehicles.

[0157] A hybrid vehicle driving optimization system is proposed according to an embodiment of the present application.

[0158] Next, a hybrid vehicle driving optimization method proposed according to an embodiment of the present application will be described with reference to the accompanying drawings.

[0159] Figure 12 4 is a flow chart of a hybrid vehicle driving optimization method according to an embodiment of the present application.

[0160] like Figure 12 As shown, the hybrid vehicle driving optimization method is applied to a cloud server and includes the following steps:

[0161] In step S101 , data on the status of the hybrid vehicle and information on the status of traffic lights and traffic flow in the current area where the hybrid vehicle is located are obtained.

[0162] Among them, the status data of the hybrid vehicle is uploaded by the hybrid vehicle, and the traffic light status and traffic flow information of the current area where the hybrid vehicle is located are obtained by the edge cloud.

[0163] In step S102, the state data, the state of the traffic lights in the current area and the traffic flow information are input into the predictive cruise algorithm, and the predictive cruise algorithm outputs the speed planning data of the hybrid vehicle.

[0164] It is understandable that the embodiment of the present application can input status data, traffic light status and traffic flow information in the current area into the predictive cruise algorithm, and the predictive cruise algorithm outputs speed planning data for the hybrid vehicle.

[0165] In step S103 , the speed planning data is sent to the hybrid vehicle, wherein the hybrid vehicle determines the engine power of the hybrid vehicle based on the speed planning data and controls the hybrid vehicle based on the engine power.

[0166] It should be noted that the aforementioned explanation of the hybrid vehicle driving optimization system embodiment is also applicable to the hybrid vehicle driving optimization method of this embodiment, and will not be repeated here.

[0167] According to the hybrid vehicle driving optimization method proposed in the embodiment of the present application, the status data of the hybrid vehicle, the status of the traffic lights in the current area where the hybrid vehicle is located, and the traffic flow information can be input into the predictive cruise algorithm deployed on the cloud server to determine the speed planning data of the hybrid vehicle. The speed planning data is sent to the hybrid vehicle, and the hybrid vehicle determines the engine power of the hybrid vehicle based on the speed planning data, and controls the hybrid vehicle based on the engine power. By deploying the speed planning problem on the cloud server, the computational complexity is reduced and the computational efficiency is improved.

[0168] Figure 13 The figure is a flow chart of a hybrid vehicle driving optimization method according to another embodiment of the present application.

[0169] like Figure 13 As shown, the hybrid vehicle driving optimization method, applied to a hybrid vehicle, comprises the following steps:

[0170] In step S201 , the status data of the hybrid vehicle is uploaded.

[0171] The status data of the hybrid vehicle can be obtained through sensors of the hybrid vehicle.

[0172] In step S202 , speed planning data of the hybrid vehicle sent from a cloud server is received.

[0173] In step S203 , the state data and the speed planning data are input into a real-time energy management model, and the real-time energy management model outputs the engine power of the hybrid vehicle, and the hybrid vehicle is controlled based on the engine power.

[0174] It can be understood that the embodiments of the present application can input status data and speed planning data into a real-time energy management model, and the real-time energy management model outputs the engine power of the hybrid vehicle, and controls the hybrid vehicle based on the engine power. By deploying the speed planning problem of the complex collaborative optimization problem in the cloud, and deploying the real-time energy management model on the vehicle side to output the engine power, the computational complexity is reduced and the computational efficiency is improved.

[0175] It should be noted that the aforementioned explanation of the hybrid vehicle driving optimization system embodiment is also applicable to the hybrid vehicle driving optimization method of this embodiment, and will not be repeated here.

[0176] According to the hybrid vehicle driving optimization method proposed in the embodiment of the present application, the status data of the hybrid vehicle wave and the speed planning data sent by the cloud server can be input into the real-time energy management model, and the real-time energy management model outputs the engine power of the hybrid vehicle. The hybrid vehicle is controlled based on the engine power. By deploying the speed planning problem of the complex collaborative optimization problem on the cloud and deploying the real-time energy management model on the vehicle side to output the engine power, the calculation complexity is reduced and the calculation efficiency is improved.

[0177] An embodiment of the present application also provides a cloud server, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the hybrid vehicle driving optimization method as described above.

[0178] An embodiment of the present application also provides a hybrid vehicle, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the hybrid vehicle driving optimization method as described above.

[0179] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0180] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0181] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0182] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, it can be implemented using any one or a combination of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0183] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

Claims

1. A hybrid vehicle driving optimization system, characterized in that: include: a cloud server, wherein the cloud server is deployed with a predictive cruise algorithm, inputting status data of the hybrid vehicle, a signal light status and traffic flow information in a current area of the hybrid vehicle into the predictive cruise algorithm, the predictive cruise algorithm outputting speed planning data of the hybrid vehicle, and transmitting the speed planning data to the hybrid vehicle; A hybrid vehicle is deployed with a real-time energy management model, the state data and the speed planning data are input into the real-time energy management model, the real-time energy management model outputs the engine power of the hybrid vehicle, and the hybrid vehicle is controlled based on the engine power.

2. The hybrid vehicle driving optimization system according to claim 1, characterized in that: The predictive cruise algorithm includes a first action network and a first evaluation network, wherein, The first action network uses the state data of the hybrid vehicle, the state of the traffic lights in the current area of the hybrid vehicle, and the traffic flow information as the state input of the first action network, and the first action network outputs the acceleration of the hybrid vehicle at the next moment based on the current state; The first evaluation network evaluates a combined evaluation value of the current state and the acceleration at the next moment, and optimizes the first action network according to the combined evaluation value and a feedback value of an objective function.

3. The hybrid vehicle driving optimization system according to claim 2, characterized in that: The objective function is used to output a reward signal to optimize the first action network and the first evaluation network. The input of the objective function is the engine power of the hybrid vehicle at the current moment, the status data of the hybrid vehicle, and the traffic light status and traffic flow information of the current area where the hybrid vehicle is located.

4. The hybrid vehicle driving optimization system according to claim 1, characterized in that: The state data includes: SOC, vehicle speed, and the real-time energy management model includes a second action network and a second evaluation network, wherein: The second action network uses the SOC of the hybrid vehicle, the current vehicle speed, and the next acceleration as state inputs, and the second action network outputs the engine power of the hybrid vehicle based on the current state; The second evaluation network evaluates a combined evaluation value of the current state and engine power, and optimizes the second action network based on the combined evaluation value and an optimization target.

5. The hybrid vehicle driving optimization system according to claim 4, characterized in that: The real-time energy management model is trained using a deep deterministic policy gradient algorithm, where: The update strategy of the deep deterministic policy gradient algorithm is: i ′ =τθ+(1-τ)θ ′ ; Among them, θ ′ is the target network parameter, θ is the current network parameter, and τ is the soft update coefficient; The loss function of the deep deterministic policy gradient algorithm is: L=E[(r+γQ ′ Q ′ (s ′ ,μ ′ (s ′ |θμ ′ )|θQ ′ )-Q(s,a|θQ)) 2 ; Among them, L is the loss function, Q ′ is the target evaluation network, μ ′ is the target action network, E(·) is the mathematical expectation, r is the immediate reward, γ is the discount factor, Q ′ (s ′ ,μ ′ (s ′ |θμ ′ )|θQ ′ ) is the target evaluation network Q ′ For the next state s ′ and target action network μ ′ Generated action μ ′ (s ′ ), Q(s,a|θQ) is the value estimate of the current evaluation network for state s and action a.

6. The hybrid vehicle driving optimization system according to claim 4, characterized in that: The optimization goal of the real-time energy management model is to minimize the fuel consumption of the hybrid vehicle and keep the SOC of the hybrid vehicle within a target range.

7. A hybrid vehicle driving optimization method, characterized in that: Applied to cloud servers, it includes the following steps: Obtaining the state number of the hybrid vehicle and the signal light state and traffic flow information of the current area where the hybrid vehicle is located; Inputting the state data, the signal light state and traffic flow information of the current area into a predictive cruise algorithm, the predictive cruise algorithm outputting speed planning data of the hybrid vehicle; The speed planning data is sent to the hybrid vehicle, wherein the hybrid vehicle determines an engine power of the hybrid vehicle based on the speed planning data, and controls the hybrid vehicle based on the engine power.

8. A hybrid vehicle driving optimization method, characterized in that: Applied to hybrid vehicles, the method comprises the following steps: Upload the hybrid vehicle's status data, traffic light status, and traffic flow information in the current area to the cloud server; receiving speed planning data of the hybrid vehicle sent by the cloud server; The state data and the speed planning data are input into a real-time energy management model, the real-time energy management model outputs the engine power of the hybrid vehicle, and the hybrid vehicle is controlled based on the engine power.

9. A cloud server, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hybrid vehicle driving optimization method according to claim 7.

10. A hybrid vehicle having a computer program or instructions stored thereon, characterized in that: The computer program or instructions are executed by a processor to implement the hybrid vehicle driving optimization method according to claim 8 .

Images