PHEV energy-saving driving control method considering battery life based on ensemble learning

Abstract

The invention relates to a PHEV energy-saving driving control method considering battery life based on ensemble learning. The method comprises the following steps: S1, constructing a comprehensive driving condition data set; s2, acquiring a global optimal comprehensive energy efficiency cost and an optimal control decision data set; s3, discretely extracting the optimal energy consumption cost, the battery life cost and a current data set; s4, calculating the SOC of the battery according to the driving state of the vehicle; s5, establishing a comprehensive energy efficiency cost model considering the energy consumption cost and the battery life cost; s6, designing a layered energy-saving driving control architecture based on integrated learning; s7, the upper layer performs integrated learning-based economic vehicle speed optimization considering the battery life; and S8, the lower layer carries out power system energy management control according to the planned vehicle speed track. A neural network is utilized to establish a fusion energy efficiency cost model and a system state transition model which comprehensively consider energy consumption and battery life, a complex coupling relation between the energy consumption characteristic of the hybrid power system and the economic vehicle speed is efficiently decoupled, and the service life of the battery is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of plug-in hybrid electric vehicles, and in particular to a PHEV energy-saving driving control method based on integrated learning and considering battery life. Background Art

[0002] The electrification of vehicles provides an effective way to reduce the dependence of vehicles on fossil fuels, among which pure electric vehicles (EV) and hybrid electric vehicles (HEV) occupy a dominant position among various solutions. In addition, PHEV has been widely used because it combines the advantages of pure electric vehicles and fuel vehicles. This development trend makes the design of energy management control algorithms more complex, focusing on effectively managing the power output distribution between the internal combustion engine and the drive motor. On the other hand, the rapid development of vehicle networking and automation technology enables vehicles to obtain surrounding driving conditions in real time, thus providing new opportunities for optimizing economical vehicle speed. Optimizing vehicle speed and energy management together may provide new opportunities to further improve the energy-saving driving of PHEV.

[0003] PHEVs are more complex than pure electric vehicles and fuel vehicles because of the presence of multiple power sources in their power systems, which brings new challenges to the energy-saving driving control of vehicles. When the power system of the controlled vehicle has multiple control degrees of freedom, there is a complex coupling relationship between the energy consumption characteristics of the vehicle power system and the economic speed. When planning the economic speed, it is crucial to fully consider the impact of the power system energy consumption characteristics and battery life costs to improve the energy economy of the entire vehicle.

[0004] At present, in PHEV research, energy-saving driving of powertrain systems mainly includes two methods: step-by-step optimization and collaborative optimization. However, step-by-step optimization fails to accurately decouple the effects of speed optimization and energy management, and cannot obtain the optimal driving energy consumption. In contrast, collaborative optimization overcomes this shortcoming. For example, China's patent application CN110667565A proposes a method for collaborative optimization of energy management for intelligent networked plug-in hybrid vehicles. The patent points out that intelligent networked plug-in hybrid vehicles obtain speed limit conditions through intelligent traffic systems, and use dynamic programming to perform collaborative optimization of energy management for intelligent networked plug-in hybrid vehicles. By optimizing speed and energy management simultaneously, the energy consumption of the vehicle powertrain is minimized. However, traditional collaborative optimization has a large computational burden and is difficult to solve. In addition, in traditional optimal energy-saving driving control, battery life issues are not considered, which may lead to a significant decrease in battery life.

[0005] Therefore, an energy-saving driving control method based on ensemble learning and considering battery life is proposed, which can effectively improve the above problems and provide a comprehensive solution for the energy-saving driving control of PHEV. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a PHEV energy-saving driving control method based on ensemble learning and considering battery life.

[0007] A PHEV energy-saving driving control method based on ensemble learning and considering battery life, the specific steps are as follows:

[0008] S1. Collect multiple actual road condition data under real driving scenarios and build a comprehensive driving condition data set;

[0009] S2. Optimize the energy management strategy of plug-in hybrid electric vehicles under comprehensive driving conditions through dynamic programming algorithms to obtain the global optimal comprehensive energy efficiency cost and optimal control decision data set. The comprehensive energy efficiency cost includes energy consumption cost and battery life cost.

[0010] S3, using the results of global optimization to discretely extract the optimal energy consumption cost, battery life cost, and current data set;

[0011] S4. Using a neural network, an adaptive SOC trajectory model is constructed to calculate the battery SOC according to the vehicle driving status;

[0012] S5. Establish a comprehensive energy efficiency cost model that considers energy consumption cost and battery life cost based on an integrated learning algorithm;

[0013] S6. Using the established comprehensive energy efficiency cost model and state transition model, design a hierarchical energy-saving driving control architecture based on ensemble learning;

[0014] S7, the upper layer performs economic vehicle speed optimization based on ensemble learning and considering battery life according to system constraints in specific driving scenarios;

[0015] S8. The lower layer performs power system energy management control according to the planned vehicle speed trajectory.

[0016] Furthermore, the actual road conditions of step S1 include the driving conditions of the vehicle in different road environments such as congestion, urban areas and highways.

[0017] Furthermore, the step S2 needs to be based on the study of existing PHEV energy management strategies, and the key variables that affect the energy consumption characteristics of the power system include vehicle speed, acceleration, mileage and SOC, and then energy management optimization is performed.

[0018] Furthermore, the step S3 obtains the optimal cumulative cost matrix and the optimal control decision matrix through DP algorithm optimization.

[0019] Furthermore, the step S4 estimates the state transition of the PHEV power system based on the vehicle driving state to establish a model of the power system adaptive SOC trajectory.

[0020] Furthermore, the step S5 estimates the comprehensive energy efficiency cost of the PHEV power system based on the vehicle driving state, and establishes a comprehensive energy efficiency cost model having two base learners and an integrated network. The base learner includes a BP-NN1 neural network for calculating the energy consumption cost and a BP-NN2 network for calculating the battery life cost, and the integrated network is an RBF-NN neural network, which takes the energy consumption cost and the battery life cost as input, integrates the results of the first two base learners, and calculates the comprehensive energy efficiency cost as output.

[0021] Furthermore, the DP optimization result in step S6 is used to extract a data set for training a data-driven model, and the online module is divided into two parts: 1) economic vehicle speed optimization based on ensemble learning; 2) energy management of the power system.

[0022] Furthermore, the step S7 calculates the comprehensive energy efficiency cost of the power system considering energy consumption and battery life through a comprehensive energy efficiency model according to the system state vector and the control decision vector. The model is an evaluation model, and the adaptive SOC trajectory model is a system model. The SOC is calculated according to the vehicle driving state.

[0023] The beneficial effects of the present invention are as follows: the present invention extracts a data set of power system energy consumption, battery life and current state changes based on the optimization results of the DP algorithm, and uses a neural network to establish a comprehensive energy efficiency cost model that comprehensively considers energy consumption and battery life and an adaptive SOC trajectory system state transition model, which are used to express the coupling relationship between vehicle motion state, power system energy consumption, battery life and state changes, and efficiently decouples the complex coupling relationship between the energy consumption characteristics of the hybrid power system and the economic vehicle speed, while extending the battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0025] Figure 1 is a flow chart of the present invention;

[0026] Figure 2 The adaptive SOC trajectory model of the power system of the present invention;

[0027] Figure 3 The comprehensive energy efficiency model structure of the present invention that integrates energy consumption and battery life;

[0028] Figure 4 It is the dual-layer energy-saving driving optimal control architecture of the present invention;

[0029] Figure 5 It is the NN-based DP optimization algorithm of the present invention;

[0030] Figure 6 This is the economical vehicle speed trajectory planning result diagram of the present invention. DETAILED DESCRIPTION

[0031] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further described below.

[0032] like Figures 1 to 6 As shown, a PHEV energy-saving driving control method based on ensemble learning considering battery life, the specific steps are as follows:

[0033] 1) Collecting multiple actual road condition data in real driving scenarios to construct a comprehensive driving condition data set; in this embodiment, the driving condition includes the driving status of the vehicle in different road environments such as congestion, urban areas and highways;

[0034] 2) Optimize the energy management strategy of plug-in hybrid electric vehicles under comprehensive driving conditions through dynamic programming algorithms to obtain the global optimal comprehensive energy efficiency cost and optimal control decision data set. The comprehensive energy efficiency cost includes energy consumption cost and battery life cost;

[0035] 2-1) According to the research on existing PHEV energy management strategies, the key variables that affect the energy consumption characteristics of the power system include vehicle speed, acceleration, mileage and SOC; the equivalent distance coefficient of the impact of mileage and SOC on energy consumption is calculated according to the following formula:

[0036]

[0037] In the formula, SOC r Indicates the remaining SOC; S r Remaining mileage, coefficient L s Indicates the ratio of the maximum distance to the maximum SOC of full charge; when S e When it is greater than or equal to 1, it can be inferred that the power is sufficient to maintain the remaining driving distance, S e The upper limit is set to 1; when S e When the range is within [0, 1], it indicates that the remaining energy in the battery cannot sustain the remaining journey, and the internal combustion engine needs to participate in driving the vehicle to the destination;

[0038] 2-2) Energy management optimization part: The economic speed trajectory is obtained according to dynamic programming. The power system energy consumption cost and battery life cost are used as the objective function, the power system control decision is used as the control variable, and the SOC is used as the state variable to construct the energy management optimization problem:

[0039]

[0040] u(s)=P m (s)

[0041] x(s)=SOC(s)

[0042] In the formula, J ene is the objective function of energy management optimization, u and x are the control variable vector and state variable of energy management optimization respectively;

[0043] 2-3) This optimization problem needs to meet the relevant constraints of the vehicle power system. The instantaneous SOC is greater than the minimum SOC and less than the maximum SOC; the instantaneous battery current I bat (s) is greater than the minimum charging current and less than the minimum discharging current; instantaneous engine torque T e (s) is greater than the minimum engine torque and less than the maximum engine torque; instantaneous engine speed n e (s), power P e (s) respectively satisfies the requirement of being greater than its minimum value and less than its maximum value; the instantaneous speed of the motor n mot (s), torque T mot (s), power P e (m), respectively greater than its minimum value and less than its maximum value;

[0044] 3) The optimal cumulative cost matrix and optimal control decision matrix obtained by DP algorithm optimization cannot be directly used for the training of the comprehensive energy efficiency model and state transition model of the power system. The results of global optimization are converted into the optimal energy consumption cost, the optimal battery life cost and the current. The present invention proposes a method for extracting the power system energy consumption and state data set based on the DP optimal solution; the driving condition and the PHEV driving state are discretized, and the equivalent distance coefficient matrix under different battery SOCs in each stage can be calculated according to the following formula:

[0045]

[0046] 4) Relying on the vehicle driving state to estimate the state transition of the PHEV power system, a model of the adaptive SOC trajectory of the power system is established. The model includes a back propagation neural network (BP-NN) with a three-layer network to construct the state transition model of the adaptive SOC trajectory of the PHEV power system. The Levenberg-Marquardt (LM) algorithm is used to search for the optimal parameters of the BP-NN in each iteration. This algorithm is an iterative algorithm that uses gradients to find extreme values ​​and has the advantages of both the gradient method and the Newton method. BP-NN is a feedforward neural network whose activation function is the hyperbolic tangent sigmoid function, which can realize any nonlinear mapping between input and output. The learning process of BP-NN is mainly divided into two stages: the first stage is to input the learning sample, set the initial weight and threshold of the network, and calculate the output of each neuron in the first layer of the network. In the second stage, the weight threshold is modified, and the influence of each weight and threshold on the gradient (total error) is calculated from the last layer, so as to adjust each weight and threshold accordingly. The activation function is shown in the formula:

[0047]

[0048] Among them, a 1 and a 0 are the neuron outputs of the current layer and the previous layer respectively; n, W and b represent the cumulative output, weight and bias respectively;

[0049] Establish an adaptive SOC trajectory model:

[0050]

[0051] In the formula, f BP-NN represents the nonlinear mapping function describing the relationship between the vehicle driving state and the optimal SOC trajectory, such as Figure 2 As shown;

[0052] The proposed DP based on neural network (NN) is used to solve the optimal driving speed curve for energy-saving driving. The power system model is integrated into the economic speed optimization in the form of a data-driven model. There is no need to use traditional methods to interpolate and look up tables and derive the power system model. This can significantly reduce the computational load of energy-saving driving control considering the energy consumption characteristics of the power system, and achieve efficient economic speed optimization considering the characteristics of the power system.

[0053] 5) Based on ensemble learning, a comprehensive energy efficiency cost model considering energy consumption and battery life is established. The overall framework is as follows: Figure 3, the computational cost module contains two base learners and an integrated network. The base learner contains a BP-NN1 neural network for calculating the energy consumption cost and a BP-NN2 network for calculating the battery life cost. The integrated network is an RBF-NN neural network that integrates the results of the first two base learners. The integrated network integrates the energy consumption and battery life costs, reduces the overall computational complexity, and improves the accuracy of the overall energy consumption assessment. RBF-NN is a feedforward network widely used in time series prediction because it can approximate any continuous function with arbitrary precision; using radial basis function (RBF) as the "basis" of the hidden layer elements to construct the hidden layer, so that the input low-dimensional pattern data can be converted to a high-dimensional space, and by transforming the input vector, the low-dimensional linear inseparable problem becomes linearly separable in the high-dimensional space; by updating the weights, it can be trained to learn highly nonlinear input / output relationships; compared with other neural network models, it has faster convergence speed and lower computational complexity; generally speaking, Gaussian function is used as RBF in the hidden layer to activate neurons, and the approximation function of RBF-NN is expressed as:

[0054]

[0055] Where x is the input vector and the weight vector is given by ω j Indicates that its dimension is K, j is the Gaussian center at x j The j-th variance on ;

[0056] 5-1) Establish energy consumption NN1 cost model

[0057] The input vector x consists of v, a and S e The output is the estimated powertrain energy cost The model can be expressed as:

[0058] x=[v,a,S e ] T

[0059]

[0060] In the formula, f BP-NN A nonlinear mapping function that describes the coupling relationship between the vehicle's driving state and the optimal energy consumption of the power system;

[0061] 5-2) Establishing the battery life NN2 cost model:

[0062] The input vector is composed of I c , SOC, I, the output is the estimated battery life cost cost bat , then the battery life cost model can be expressed as:

[0063]

[0064] 5-3) Integrate energy consumption cost and battery life cost to establish a comprehensive energy efficiency cost model, with the input vector being cos t bat , The output is the estimated comprehensive energy efficiency cost The comprehensive energy efficiency cost model can be expressed as:

[0065]

[0066] 6) Design a hierarchical energy-saving driving control architecture based on ensemble learning for hybrid powertrain vehicles. The proposed energy-saving driving control architecture is as follows: Figure 4 As shown, in the offline module, the DP optimization results are used to extract the data set for training the integrated learning model. The online module is divided into two parts:

[0067] 6-1) Economic speed optimization based on ensemble learning;

[0068] 6-2) Energy management of power systems;

[0069] 7) Establishing the hierarchical control architecture for energy-saving driving. Upper layer of online module: Economic speed optimization based on ensemble learning. The standard DP algorithm cannot solve the optimal economic speed problem in the proposed control architecture, because the traditional power system model is not applied to the economic speed optimization. A DP algorithm based on ensemble learning is established to solve the optimal driving speed curve for energy-saving driving. The comprehensive energy efficiency model of the power system energy consumption cost and battery life is defined as the evaluation model, and the power system energy consumption is calculated according to the system state vector and the control decision vector. The adaptive SOC trajectory model is the system model, and the SOC is calculated according to the vehicle driving state, such as Figure 5 As shown, the SOC state transfer is calculated according to the following formula:

[0070]

[0071] In this way, there is no need to use traditional methods to interpolate and look up tables and derive the power system model, which can significantly reduce the computational load of energy-saving driving control that takes into account the energy consumption characteristics of the power system;

[0072] Based on the optimization results of the DP (Dynamic Programming, DP) algorithm, a data set of power system energy consumption and battery life state changes was extracted. A fusion energy efficiency cost model and a system state transition model that comprehensively consider energy consumption and battery life were established using a neural network to express the coupling relationship between the vehicle's motion state and the power system energy consumption and state changes. The complex coupling relationship between the energy consumption characteristics of the hybrid system and the economic speed was effectively decoupled, while extending the battery life.

[0073] 7-1) Establish the objective function of vehicle speed optimization:

[0074]

[0075] Where J is the vehicle speed optimization objective function for energy-saving driving control, and L1, L2, and L3 are defined by the following equations:

[0076] L1=L c (x k ,u k )

[0077] L2=L t (x k ,u k )+L(x N )

[0078]

[0079] Among them, L c Represents the energy consumption cost of each discrete stage, L t represents the travel time cost of each discrete stage, L(x N ) is the terminal cost of completing the driving task, the calculation of the effective Ah-flux of the L3 battery is the cost function to measure the battery aging, and k represents the coefficient of converting the battery life cost into monetary cost. represents the weight coefficient, σ represents the severity factor of the battery, I C Indicates the charge and discharge rate of the battery;

[0080] An integrated learning model is introduced as an evaluation model and system model for speed optimization, which comprehensively optimizes energy consumption economy and battery life, improves computing efficiency, and ultimately achieves optimal control of PHEV energy-saving driving;

[0081] 7-2) In order to further improve the computational efficiency, the travel time penalty factor β is used to eliminate the travel time from the state variables of the optimal control problem and establish the optimal control cost function:

[0082] L(x s ,u s )=L com +L t

[0083]

[0084] Where, L com Represents the comprehensive energy efficiency cost of each discrete stage, L t Represents the travel time cost of each discrete stage. The expected travel time can be achieved by adjusting the weight coefficient by dichotomy;

[0085] 7-3) The objective function of vehicle speed optimization can be transformed into:

[0086]

[0087] In the formula, L(x N ) is the terminal cost of completing the driving task, which can be determined based on the terminal constraints of vehicle speed and travel time;

[0088] 7-4) Establish control variables, state variables and state transfer equations based on ensemble learning speed optimization:

[0089] u=T

[0090] x=[a SOC] T

[0091]

[0092] SOC(s+1)=SOC(s)+f BP-NN

[0093] Where T represents the total output torque of the vehicle; M0 represents the vehicle mass; r w represents the tire radius; g represents the acceleration due to gravity; f represents the friction coefficient; C d Indicates the drag coefficient; A f Indicates the frontal area; F r Indicates driving resistance, F r The calculation formula is as follows:

[0094]

[0095] 7-5) Establish the optimal control speed, acceleration, duration, and SOC constraints. The initial SOC is equal to SOC0, the instantaneous SOC(s) satisfies the conditions of being greater than the instantaneous minimum SOC and less than the instantaneous maximum SOC; the instantaneous acceleration a(s) is greater than the minimum instantaneous acceleration and less than the maximum instantaneous acceleration; the instantaneous speed v(s) is greater than the minimum instantaneous speed and less than the maximum instantaneous speed, and the starting speed v(0) is equal to v0; the time constant t(s k ) is less than the expected time for the vehicle to reach the destination;

[0096] 8) Energy-saving driving hierarchical control architecture lower layer: Perform power system energy management control according to the planned vehicle speed trajectory, and send torque control instructions to the power system engine, motor and other components.

[0097] In order to verify the proposed energy-saving driving control method based on ensemble learning considering battery life, the present invention compares and analyzes two energy-saving driving control strategies through simulation experiments, including: a data-driven energy-saving driving control strategy without considering battery life, referred to as D-opt, and an energy-saving driving control strategy based on ensemble learning considering battery life, referred to as B-opt.

[0098] The comprehensive energy efficiency cost model established based on ensemble learning enables PHEV to cope with the complex relationship between the two in real-world environments. By finding a reasonable balance between battery life and energy consumption, it can improve the overall performance of PHEV and extend battery life.

[0099] The economic speed trajectory planning results of the two energy-saving driving control strategies are shown in Figure 6 As shown, the comparison of energy consumption cost and battery life cost of specific UUDS working conditions is shown in Table 1.

[0100] Table 1 Comparative analysis of UUDS working condition energy consumption and battery life

[0101]

[0102] It can be seen from Table 1 that compared with the traditional D-opt strategy, the fuel consumption cost of the proposed B-opt strategy only increased by 1.75%; while the effective ampere-hour flux and battery capacity decay were reduced by 22.4% and 12.8% respectively. Compared with fuel consumption, the effective ampere-hour flux and capacity decay were reduced to a greater extent.

[0103] Although the energy-saving driving control strategy that takes battery aging into account will increase fuel consumption to a certain extent, it significantly slows down the decline of battery capacity, thereby effectively reducing the degree of battery aging.

[0104] The terms involved in the present invention are explained as follows:

[0105] Data-driven: is a decision-making and operating approach that emphasizes the use of data analysis and data insights to guide decisions, optimize processes, and develop strategies.

[0106] Dynamic programming algorithm: is an optimization method for solving complex problems, especially for problems with overlapping subproblems and optimal substructure properties. Dynamic programming improves efficiency by breaking down the problem into smaller subproblems and storing the solutions to these subproblems to avoid repeated calculations.

[0107] Neural Network: A computational model that mimics biological nervous systems, such as the brain, and is used to process various types of complex tasks and data. They play a central role in machine learning and artificial intelligence. The basic unit of a neural network is a neuron, which is similar to a nerve cell in a biological nervous system. Each neuron receives input, performs weighted calculations, and generates output.

[0108] Comprehensive Energy Efficiency Model: A framework for evaluating and optimizing system energy efficiency. It combines multiple factors and variables to comprehensively analyze the efficiency of energy use, especially in the field of electric vehicles and plug-in hybrid vehicles.

[0109] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A PHEV energy-saving driving control method based on ensemble learning and considering battery life, characterized in that: The specific steps are as follows: S1. Collect multiple actual road condition data under real driving scenarios and build a comprehensive driving condition data set; S2. Optimize the energy management strategy of plug-in hybrid electric vehicles under comprehensive driving conditions through dynamic programming algorithms to obtain the global optimal comprehensive energy efficiency cost and optimal control decision data set. The comprehensive energy efficiency cost includes energy consumption cost and battery life cost. S3, using the results of global optimization to discretely extract the optimal energy consumption cost, battery life cost, and current data set; S4. Using a neural network, an adaptive SOC trajectory model is constructed to calculate the battery SOC according to the vehicle driving status; S5. Establish a comprehensive energy efficiency cost model that considers energy consumption cost and battery life cost based on an integrated learning algorithm; S6. Using the established comprehensive energy efficiency cost model and state transition model, design a hierarchical energy-saving driving control architecture based on ensemble learning; S7, the upper layer performs economic vehicle speed optimization based on ensemble learning and considering battery life according to system constraints in specific driving scenarios; S8. The lower layer performs power system energy management control according to the planned vehicle speed trajectory.

2. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The actual road conditions in step S1 include the driving conditions of the vehicle in different road environments such as congestion, urban areas and highways.

3. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The step S2 needs to be based on the research of existing PHEV energy management strategies. The key variables that affect the energy consumption characteristics of the power system include vehicle speed, acceleration, mileage and SOC. 2-1) Calculate the equivalent distance coefficient of the impact of mileage and SOC on energy consumption; 2-2) Constructing the optimization problem of energy management: u(s)=P m (s) x(s)=SOC(s) In the formula, J ene is the objective function of energy management optimization, u and x are the control variable vector and state variable of energy management optimization respectively.

4. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The step S3 obtains the optimal cumulative cost matrix and the optimal control decision matrix through DP algorithm optimization.

5. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The step S4 estimates the energy consumption and state transition of the PHEV power system based on the vehicle driving state, and establishes a power system adaptive SOC trajectory model using a BP neural network; Construct an adaptive SOC trajectory model whose input is acceleration a, speed v and equivalent distance coefficient S e , the output is the trajectory of SOC, as shown in the following formula:

6. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The comprehensive energy efficiency cost model of step S5 includes two base learners and an integrated network. The base learner includes a BP-NN1 neural network for calculating energy consumption cost and a BP-NN2 network for calculating battery life cost. The integrated network is an RBF-NN neural network, which integrates the results of the first two base learners. 6-1) Establish the energy consumption NN1 cost model, with inputs consisting of v, a and S e The output is the estimated powertrain energy cost The model can be expressed as: In the formula, f BP-NN A nonlinear mapping function that describes the coupling relationship between the vehicle's driving state and the optimal energy consumption of the power system; 6-2) Establish the battery life NN2 cost model, the input vector is composed of I c , SOC, I, the output is the estimated battery life cost cost bat , then the battery life cost model can be expressed as: 6-3) Integrate energy consumption cost and battery life cost to establish a comprehensive energy efficiency cost model, with the input vector being cos t bat , The output is the estimated comprehensive energy efficiency cost The comprehensive energy efficiency cost model can be expressed as:

7. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The DP optimization result in step S6 is used to extract a data set for training an integrated learning model. The online module is divided into two parts: 1) economic vehicle speed optimization based on integrated learning; 2) energy management of the power system.

8. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The step S7 calculates the comprehensive energy efficiency cost of the integrated power system energy consumption cost and the battery life cost according to the system state vector and the control decision vector, the adaptive SOC trajectory model is a system model, and the SOC is calculated according to the vehicle driving state; 7-1) Establish the objective function of vehicle speed optimization: Where J is the vehicle speed optimization objective function for energy-saving driving control, and L1, L2, and L3 are defined by the following equations: L1=L c (x k ,or k ) L2=L t (x k ,u k )+L(x N ) Among them, L c Represents the energy consumption cost of each discrete stage, L t represents the travel time cost of each discrete stage, L(x N ) is the terminal cost of completing the driving task, the calculation of the effective Ah-flux of the L3 battery is the cost function to measure the battery aging, and k represents the coefficient of converting the battery life cost into monetary cost. represents the weight coefficient, σ represents the severity factor of the battery, I C Indicates the charge and discharge rate of the battery; 7-2) In order to further improve the computational efficiency, the travel time penalty factor β is used to eliminate the travel time from the state variables of the optimal control problem and establish the optimal control cost function: L(x s ,and s )=L com +L t Where, L com Represents the comprehensive energy efficiency cost of each discrete stage, L t Represents the travel time cost of each discrete stage. The expected travel time can be achieved by adjusting the weight coefficient by dichotomy; 7-3) The objective function of vehicle speed optimization can be transformed into: In the formula, L(x N ) is the terminal cost of completing the driving task, which can be determined based on the terminal constraints of vehicle speed and travel time; 7-4) Establish control variables, state variables and state transfer equations based on ensemble learning speed optimization: u=T x=[a SOC] T SOC(s+1)=SOC(s)+f BP-NN Where T represents the total output torque of the vehicle; M0 represents the vehicle mass; r w represents the tire radius; g represents the acceleration due to gravity; f represents the friction coefficient; C d Indicates the drag coefficient; A f Indicates the frontal area; F r Indicates driving resistance, F r The calculation formula is as follows: 7-5) Establish the optimal control speed, acceleration, duration, and SOC restriction conditions. The initial SOC is equal to SOC0, the instantaneous SOC(s) satisfies the conditions of being greater than the instantaneous minimum SOC and less than the instantaneous maximum SOC; the instantaneous acceleration a(s) is greater than the minimum instantaneous acceleration and less than the maximum instantaneous acceleration; the instantaneous speed v(s) is greater than the minimum instantaneous speed and less than the maximum instantaneous speed, and the starting speed v(0) is equal to v0; the time constant t(s k ) is less than the expected time for the vehicle to reach the destination.

9. The PHEV energy-saving driving control method based on ensemble learning and considering battery life according to claim 1, characterized in that: The step S8 performs power system energy management control according to the planned vehicle speed trajectory, and sends torque control instructions to the power system engine, motor and other components.

Images