Self-adaptive energy management strategy construction method for hybrid electric vehicle

Through the adaptive energy management strategies of LSTM and SSA-GRU neural networks, the uneven torque distribution problem of hybrid vehicles under complex road conditions is solved, efficient energy management and low-cost driving are achieved, and the power of the vehicle is improved. Performance and adaptability.

CN119078786AActive Publication Date: 2024-12-06JIANGSU UNIV +2

Patent Information

Application Number
CN202411367200.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2024-12-06
Estimated Expiration
2044-09-29

AI Technical Summary

Technical Problem

The energy management strategies of existing hybrid vehicles are difficult to adapt effectively when facing complex road conditions and emergencies, resulting in uneven torque distribution and poor applicability of energy management, which affects the fuel economy and power performance of the vehicle.

Method used

Adaptive energy management strategy combined with LSTM neural network and SSA-GRU neural network is adopted to optimize the torque distribution of engines and motors through real-time road classification and torque compensation, and achieve adaptability and robustness to different road conditions.

Benefits of technology

It improves the operating efficiency of hybrid vehicles under different road conditions, suppresses sudden torque changes and engine shaking, enhances the dynamic and static characteristics and robustness of the energy management system, and reduces driving costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119078786A_ABST
    Figure CN119078786A_ABST
Patent Text Reader

Abstract

The invention discloses a hybrid electric vehicle adaptive energy management strategy construction method in the field of new energy intelligent management. An energy management strategy of the hybrid electric vehicle is formed by a driver module, an MPC and road section selection module, a self-adaptive strategy adjustment module, an LSTM neural network torque compensation module, an SOC tracking module and an SSA-GRU neural network road classification module, and self-adaptive energy management control is achieved for a hybrid power source module. The engine torque of the hybrid electric vehicle is compensated through an LSTM neural network torque compensation module, road classification and partial torque compensation are achieved by combining SSA-GRU neural network road classification and applying multiple neural network algorithms, and especially independent control over the engine torque, the motor torque and the vehicle speed is achieved. All control factors are mutually associated and mutually independent, an instantaneous optimal torque distribution sequence is designed, the problem of poor adaptability of an energy management strategy is effectively solved, and the prediction precision and convergence speed of long and short term driving speed information are high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of new energy intelligent management, and specifically relates to an energy intelligent management control strategy for a hybrid electric vehicle. Background Art

[0002] Hybrid vehicles have the advantages of both pure electric vehicles and traditional fuel vehicles, and have become an important direction for the development of the automobile industry. Hybrid vehicles can operate in different modes to cope with complex working conditions, so that the control of vehicle energy flow and the optimization of energy consumption have greater flexibility, and it is easier to achieve the dual goals of low fuel consumption and low emissions. As a typical multi-power source input and multi-objective control system, the hybrid system is very complex in terms of optimizing energy distribution, braking energy recovery, reasonable matching of power system parameters, and coordinated control. During the research and development process, its control strategy must not only consider the fuel economy and power of the vehicle under different working conditions, but also take into account various factors such as the torque size and the cost of the whole vehicle. The technical difficulty is high, so designing a suitable energy management construction method is the core content and difficulty of hybrid technology research.

[0003] At present, the energy management strategies of hybrid vehicles are mainly divided into two categories: rule-based and optimization-based strategies. The optimization-based strategies are mainly based on the equivalent fuel consumption minimum strategy (ECMS) and the Pontryagin minimum principle (PMP). For example, the document with patent publication number CN 114771293 proposes an energy management method based on a long short-term memory (LSTM) training model to predict road conditions, but the model training process does not consider the emergencies and torque compensation problems in the driving process. The document with patent publication number CN112706753 proposes an energy management strategy based on the Grey Wolf Optimization Algorithm (GWO) to adjust the optimal equivalent factor in ECMS in real time, which can adapt to changing conditions, but the real-time application of GWO cannot avoid compromises in accuracy. It can be seen that most of the existing strategies calculate or establish the relationship between control parameters and prediction information in a traditional way. However, in actual driving, due to different road conditions and scenarios, the applicability of previous strategies is poor, and many predicted torque distribution and energy management cannot avoid the impact of parameter fluctuations during driving. Therefore, it is of great significance to combine road information and update energy management strategies and torque compensation according to actual road conditions. Summary of the invention

[0004] The purpose of the present invention is to propose a method for constructing an adaptive energy management strategy for a hybrid vehicle in order to solve the problems of torque distribution, parameter fluctuations during driving, road adaptability, and battery state of charge (SOC) in existing hybrid vehicle controllers, thereby improving the road adaptability of the hybrid vehicle and making the driving cost as low as possible.

[0005] The technical solution adopted by the method for constructing a hybrid electric vehicle adaptive energy management strategy of the present invention comprises the following steps:

[0006] Step 1): The hybrid power source module is composed of an engine module, a motor module and a battery module. The input of the engine module is the engine required torque T E_r 、Engine power P E , Motor required torque T m_r 、Motor power P m And the battery voltage U batt , the output is the engine speed n E , motor speed n m , the actual speed of the vehicle V actual , the actual displacement of the vehicle X actual , actual torque T actual And the actual state of charge SOC of the battery actual ;

[0007] Step 2): The actual torque T actual , actual vehicle speed V actual The training sample set of the LSTM neural network is used to train the LSTM neural network using the long short-term memory algorithm to output the compensation torque T m2 ;

[0008] Step 3): The actual torque T actual , actual state of charge SOC actual And the actual displacement X actual After ideal SOC trajectory planning, the ideal state of charge SOC is obtained ideal ;

[0009] Step 4): Classify the real-time geographic location GPS information through the SSA group intelligence algorithm and the GRU neural network to obtain the classified reference locations of various types of road sections

[0010] Step 5): The vehicle speed controller, the power control module and the driving command module constitute a driver module, and the input of the driver module is the actual vehicle speed V actual , the actual torque T actual , the actual vehicle speed V actual and the required vehicle speed V req The speed error V * And the engine speed n E and the motor speed n m The output is the vehicle driving command sign and the engine power P E And the motor power P m ;

[0011] Step 6): The road section selection module, the Markov vehicle speed prediction module and the MPC module constitute the MPC and road section selection module, and the input of the MPC and road section selection module is the vehicle driving instruction sign, the engine power P E 、Motor power P m , actual vehicle speed V actual , the actual displacement of the vehicle X actual And various types of road section reference positions The output is the predicted vehicle speed V fore ;

[0012] Step 7): The fuel consumption calculation module, the constraint condition judgment module and the ECMS torque distribution module constitute an adaptive strategy adjustment module, and the input of the adaptive strategy adjustment module is the actual torque T actual , predicted vehicle speed V fore And the ideal state of charge SOC ideal , the output is the engine distribution torque T E1 and motor torque T m1 ;

[0013] Step 8): The compensation torque T m2 The torque T is distributed to the motor m1 Compensation is performed to form the engine required torque T E_r and motor torque T m_r :

[0014] Furthermore, the fuel consumption calculation module outputs the total fuel consumption m f The constraint condition judgment module will be the actual torque T actual and ideal state of charge SOC ideal Compare with the constraint value to determine whether it exceeds the maximum or minimum value, and output the constraint judgment signal Y=(T j ,SOC j ),

[0015]

[0016] T j It is the torque judgment signal, SOC j is the charge judgment signal, T max is the maximum torque, SOC h and SOC l They are the upper and lower boundary values ​​of SOC respectively;

[0017] The ECMS torque distribution module establishes a total cost function based on the Pontryagin minimum principle u* (t) = argminJ over (x(t),u(t),t), by iteratively solving the minimum value of the cost function and the engine distribution torque T at this time E1 , motor distribution torque T m1 ; x(t) and u(t) are the state and control variables in the Hamiltonian function, respectively. x(t) is the state of charge SOC, and u(t) is the motor torque T m1 , t is time, C f is the fuel consumption cost coefficient, C l is the constraint cost coefficient, ranging from 0.8 to 1.4, C s is the SOC deviation penalty factor, set to 0.1, C V is the speed deviation penalty coefficient, the value range is 0.1~1.0, f1, f2, e1, e2 are constants, the range of f1 and e1 is [0,10], f2 is between 1.0~2.0, e2 is between 0.5~1.0, m f (t) is the input fuel consumption, SOC ref is the reference SOC value, u * (t) represents the sequence of optimal solutions for the engine and motor torque distribution at a specific time, and arg min is the minimum solution function, which is used to obtain the cost function J over The minimum value at a specific moment and the optimal u(t) sequence recorded at that moment.

[0018] The beneficial effects of the present invention after adopting the above technical solution are:

[0019] 1. The present invention uses LSTM (Long Short-Term Memory) neural network torque compensation module to compensate the engine torque of hybrid vehicles, and combines SSA-GRU (Sparrow Search-Gated Recurrent) neural network road classification to achieve intelligent classification of roads and accurate compensation of torque, so that the vehicle can run efficiently under different road conditions. Through the learning ability of the neural network, the system can continuously optimize its own control strategy, adapt to different driving environments and conditions, effectively suppress the sudden change of torque and engine vibration during driving, so that the hybrid vehicle energy management system has good dynamic and static characteristics and strong robustness.

[0020] 2. The present invention designs an instantaneous optimal torque distribution sequence, thereby effectively solving the problem of poor adaptability of the energy management strategy; at the same time, an LSTM neural network is used to train the torque data, the purpose of which is to solve the problem of sudden torque changes in the engine or motor when encountering emergencies during driving.

[0021] 3. The present invention uses a variety of neural network algorithms to achieve road classification and partial torque compensation, especially independent control of the engine, motor torque and vehicle speed, so that each control factor is both interconnected and independent of each other, effectively solving the problem of poor adaptability of energy management strategies.

[0022] 4. The present invention only needs to use the adaptive energy management strategy to adjust the input and output data of vehicle speed, torque, position, etc. to realize the identification of the LSTM torque compensation module and the classification of road sections using the SSA-GRU neural network. These variables are easy to obtain in engineering practice, and the neural network algorithm only needs to be implemented through software programming, without the need to add other hardware equipment, low cost, and easy engineering implementation. The prediction accuracy of long-term and short-term driving speed information is high, the convergence speed is fast, and the robustness is strong, which further improves the accuracy of road allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a block diagram of the energy management strategy of hybrid electric vehicles;

[0024] Figure 2 yes Figure 1 An equivalent structural block diagram of the hybrid power source module 4;

[0025] Figure 3 yes Figure 1 The structural block diagram of the driver module 1 in FIG.

[0026] Figure 4 yes Figure 1 The structural block diagram of the vehicle speed MPC and road section selection module 2;

[0027] Figure 5 yes Figure 1 Schematic diagram of the structure of the adaptive strategy adjustment module 3.

[0028] In the figure: 1. Driver module; 2. MPC and road section selection module; 3. Adaptive strategy adjustment module; 4. Hybrid power source module; 5. LSTM neural network torque compensation module; 6. SOC tracking module; 7. Road classification module; 11. Driving speed controller; 12. Power control module; 13. Driving instruction module; 21. Road section selection module; 22. Markov vehicle speed prediction module; 23. MPC module; 31. Fuel consumption calculation module; 32. Constraint judgment module; 33. ECMS torque distribution module; 41. Engine module; 42. Motor module; 43. Battery module. DETAILED DESCRIPTION

[0029] See also Figure 1The hybrid vehicle adaptive energy management strategy of the present invention is composed of a driver module 1, an MPC (vehicle speed model predictive control) and a road section selection module 2, an adaptive strategy adjustment module 3, an LSTM neural network torque compensation module 5, an SOC tracking module 6 and an SSA-GRU neural network road classification module 7, which realizes adaptive energy management control of a hybrid power source module 4 of the hybrid vehicle.

[0030] Combination Figure 2 As shown, the hybrid power source module 4 is composed of an engine module 41, a motor module 42 and a battery module 43, which are the power sources of the entire hybrid vehicle. The input of the hybrid power source module 4 is the engine demand torque T E_r and engine power P E , Motor required torque T m_r and motor power P m And the battery voltage U batt The output of the hybrid power source module 4 is the engine speed n E , motor speed n m , the actual speed of the vehicle V actual , the actual displacement of the vehicle X actual , actual torque T actual And the actual state of charge SOC actual .

[0031] The input of the engine module 41 is the engine demand torque T E_r and engine power P E , its output is the actual engine torque T E , engine speed n E and actual vehicle speed V actual . For the actual vehicle speed V actual The actual displacement X of the vehicle can also be obtained by integration actual , the calculation formula is as follows:

[0032] T E =T E_r / η E ,

[0033] n E =60·P E / (2π·T E ),

[0034] V actual =P E ·R / (i g ·T E ),

[0035] X actual =∫V actual ,

[0036] Where, T E is the actual engine torque, η E is the mechanical efficiency, n E is the engine speed, P E is the engine power, T E_r is the engine torque requirement, V actual is the actual vehicle speed, R is the wheel radius, i g is the transmission ratio, X actual is the actual displacement of the engine, that is, the actual distance traveled.

[0037] The input of the motor module 42 is the motor required torque T obtained after distribution. m_r and motor power P m , the output is the motor speed n m and the actual motor torque T m .

[0038] The engine module 41 and the motor module 42 are based on the input engine demand torque T E_r and motor torque T m_r , combined with the working efficiency and energy loss of the actual working components, multiplied by the efficiency coefficient to obtain the actual engine torque T E The actual torque T of the motor m , and the actual engine torque T E and the actual motor torque T m The actual torque T of the entire hybrid power source actual , the calculation formula is as follows:

[0039] T actual =[T E ,T m ],

[0040] T m =T m_r / η m ,

[0041] n m =60·P m / (2π·T m ),

[0042] Where, T actual is the actual torque composition, T E 、T m are the actual torque of the engine and the actual torque of the motor, T m_r is the motor required torque, η m is the motor efficiency, P m is the motor power.

[0043] The input of the battery module 43 is the connected battery pack voltage U batt , the current I is obtained by calculation batt , and further calculate the actual state of charge SOC actual As output:

[0044]

[0045] SOC actual =SOC0-∫I batt / Q batt ,

[0046] In the formula, I batt is the battery current, U batt is the battery pack voltage, P batt is the battery pack power, R int is the battery internal resistance, SOC actual is the actual state of charge of the battery pack, SOC0 is the initial SOC value, Q batt is the battery capacitor.

[0047] The LSTM neural network torque compensation module 5 is formed by using an LSTM neural network with 2 input nodes and 1 output node. The actual torque T at the output end of the hybrid power source 4 is actual , actual vehicle speed V actual The signal is normalized to form a training sample set of the LSTM neural network. The LSTM neural network is trained using the long short-term memory algorithm. After about 600 trainings, the mean square error of the neural network output is less than 0.2, which meets the requirements. Determine the weight coefficients of the neural network to obtain the compensation torque T output by the LSTM neural network torque compensation module 5 m2 . Compensation torque T m2 The motor distribution torque T output by the adaptive strategy adjustment module 3 m1 Compensation is performed to form the engine required torque T E_r and motor torque T m_r , reduce the engine torque mutation:

[0048]

[0049] Where T E_r is the engine demand torque, T E1 is the engine torque distribution, T m_r is the motor required torque, T m1 is the motor distribution torque, T m2 is the compensation torque.

[0050] The input of the SOC tracking module 6 is the actual torque T provided by the hybrid power source 4 actual, actual state of charge SOC actual And the actual displacement X actual , after ideal SOC trajectory planning, combined with the actual displacement X actual The road section location and the actual motor torque T m Get the ideal state of charge SOC ideal And as output, it is input into the adaptive strategy adjustment module 3:

[0051]

[0052] Where SOC ideal is the ideal state of charge, SOC actual is the actual state of charge, T m is the actual torque T actual The motor torque part, X actual is the actual displacement, X total is the total stroke length, η m is the motor efficiency, U is the battery rated voltage, Q batt is the battery capacitance value.

[0053] The SSA-GRU neural network road classification module 7 takes the actual geographic location GPS information as input, and randomly divides 70,000 groups of driving data into 70% training set, 15% validation set, and 15% test set ratios for training and testing through the SSA swarm intelligence algorithm and the GRU neural network, and builds a GRU neural network model to process time series data. Then, the hyperparameters or weights of the model are optimized through SSA, and the goal is to minimize the loss function on the validation set, and iteratively search for the optimal solution. Finally, the entire driving section is divided into 5 categories: urban street sections, highway sections, mountain slope sections, congested sections, and bridge sections, and the classified sections of various types are referenced to the location. Input to MPC and road segment selection module 2.

[0054] Combination Figure 3 As shown, the vehicle speed controller 11, the power control module 12 and the driving instruction module 13 together constitute the driver module 1, and the input of the driver module 1 is the actual vehicle speed V actual , actual torque T actual , actual vehicle speed V actual and the required vehicle speed V req The speed error V * , that is, V * =V actual -V req , and engine speed n E and motor speed n m The output is the driving command sign and the engine power P E and motor power Pm The purpose of the driving instruction sign is to give the control module clear driving instructions.

[0055] The vehicle speed controller 11 receives the actual vehicle speed V actual and the required vehicle speed V req The speed error V * , vehicle speed error V * Then, the acceleration or braking PID signal is obtained through conventional PID control and output to the driving instruction module 13 .

[0056] The input to the power control module 12 is the actual torque T actual and engine speed n E and motor speed n m The power control module 12 calculates the power P corresponding to the engine and the motor at the current vehicle speed and rotation speed according to the following formula: E , P m :

[0057]

[0058] Where P E is the engine power of the hybrid vehicle, T actual is the actual torque, n E is the engine speed, P m is the motor power of the hybrid vehicle, n m is the motor speed.

[0059] The input of the driving instruction module 13 is the PID signal of acceleration or braking and the actual vehicle speed V actual In the driving instruction module 13, the optimal working area table of the driving instruction sign of the vehicle obtained according to the engine efficiency map is preset, which is also the relationship table between the throttle opening and the vehicle speed. It is an optimal working area table built by a series of throttle opening and vehicle speed combinations designed in advance according to the efficiency characteristics under different speeds and loads in the known engine efficiency map, ensuring that the engine can operate in a higher efficiency area, thereby improving fuel economy and reducing emissions. Based on the mapping relationship of the optimal working area table of the driving instruction sign of the vehicle, the throttle opening can be obtained, that is, the driving instruction sign of the hybrid vehicle. Therefore, this mapping relationship is uniquely determined by looking up the table:

[0060] sign=L(n E ,l(V actual ,a)),

[0061] Where, sign is the vehicle driving command signal obtained by mapping, L is the table lookup function. In the engine characteristic diagram, the optimal performance parameter operating point of the engine under different speed and load conditions is related to the throttle opening and can be uniquely determined by the table lookup method. l(V actual , a) l is the working load calculated by speed and acceleration, V actual is the actual vehicle speed, and a is the specific acceleration stripped from the acceleration or braking signal. actual ,a) is determined by the following formula:

[0062]

[0063] In the formula, The engine power requirement is is the maximum output power of the engine, C r is the road rolling resistance coefficient, m is the vehicle mass, g is the gravitational acceleration constant, C d is the air resistance coefficient, A o is the frontal area of ​​the vehicle body, ρ is the air density, and θ is the road slope.

[0064] Combination Figure 4 As shown, the MPC and road section selection module 2 is composed of a road section selection module 21, a Markov vehicle speed prediction module 22 and an MPC module 23. The input of the entire MPC and road section selection module 2 is the vehicle driving instruction sign, the power P corresponding to the motor and the engine E , P m , actual vehicle speed V actual , actual section position actual displacement X actual And various types of road section reference positions The output is the predicted vehicle speed V obtained by solving fore , and is used for the next stage of calculation.

[0065] The input of the road section selection module 21 is the power P corresponding to the engine and the motor. E , P m , the actual displacement X of the hybrid vehicle actual , various types of road section reference positions And the actual vehicle speed V actual The road section selection module 21 determines the road section category corresponding to the current vehicle road according to the above input and outputs the category information R kIn order to predict the vehicle speed in the next step. The implementation of this process is divided into two steps: the first step is to pre-process the vehicle driving data of a specific road section into an electronic map to establish the relationship between the geographical coordinates of the hybrid vehicle and the real road. Based on this information, the k-means clustering algorithm can be used to obtain the characteristic parameters of different roads. This step must be completed before online simulation. The second step is to determine the road type information R by comparing the engine, motor power, road section location, actual vehicle speed and the similarity of different road types. k And output:

[0066]

[0067] Where j is the pre-processed road type, usually j = 1, 2, ..., 5 to represent five different road types, d is the distance metric function, and c is the data point, which contains P E , P m 、V actual , X actual Four characteristic parameters, c j is the centroid position of the j-th type of road, which is also determined by the above four characteristic parameters P E , P m 、V actual , X actual The argmin function is used to determine the value of j when d is minimum, so R k is the road type represented by the cluster closest to data point c. Note that is the centroid characteristic parameter value of the j-th type of road predetermined by the k-means clustering algorithm, and is the reference position of the jth type of road In k-means clustering, the centroid c j It is determined by iteratively updating the following formula until convergence:

[0068]

[0069] In the formula, is the center of mass position after iterative update, and when the convergence condition is met C j is the set of all points belonging to the jth cluster.

[0070] The input of the Markov speed prediction module 22 is the actual vehicle speed V actual and driving command sign, the output is the selected vehicle speed state transfer matrix P e According to the historical vehicle speed information, the probability of any state transferring to all other states in one step is obtained, and the speed state transfer matrix is ​​established:

[0071]

[0072] Where P e is the state transition probability, S is the set of all values ​​of the parameter corresponding to its subscript at the superscript time, V actual is the actual vehicle speed, sign is the driving instruction, k is the minimum step length defined in the transfer range, generally 0.1 or 0.01s, and is all possible states in the next moment, Indicates the status by Transfer to The number of times, Indicates from the state The total number of possible departures from all transfers.

[0073] The MPC module 23 converts the road type information R output by the road section selection module 21 and the Markov vehicle speed prediction module 22 into k and vehicle speed state transfer matrix P e As input, the output is the predicted vehicle speed V fore :

[0074] V fore ={...,V actual (k-1),V actual (k), V(k+1), V(k+2), ...},

[0075]

[0076] Where V fore The output predicted vehicle speed information includes the actual vehicle speed information and the predicted future vehicle speed information V(k+1), V(k+2), ..., V actual is the actual vehicle speed information, V next is the predicted vehicle speed at the next moment, P e is the transition probability, R k is the road type information, and k is the prediction step length.

[0077] Combination Figure 5 As shown in FIG. 1 , the fuel consumption calculation module 31, the constraint condition judgment module 32 and the ECMS (equivalent fuel consumption minimum strategy) torque distribution module 33 together constitute the adaptive strategy adjustment module 3. The adaptive strategy adjustment module 3 is the most important part of the entire hybrid vehicle energy management strategy construction method. Its input is the actual torque T actual , predicted vehicle speed V fore And track and solve the ideal SOC ideal , SOC is the battery charge state (remaining power), and output is the engine torque T obtained after reasonable distribution E1 and motor torque Tm1 .

[0078] The actual torque T actual As the input of the fuel consumption calculation module 31, its output is the fuel consumption m f , based on the instantaneous fuel consumption Through the internal integration module, we get:

[0079]

[0080] Where: n E is the engine speed, T actual is the actual torque, b represents the instantaneous fuel consumption rate, r f and g represent the fuel density and the acceleration due to gravity respectively.

[0081] The constraint condition judgment module 32 converts the input actual torque T actual and ideal SOC ideal Compare with the constraint value of each functional component to determine whether it exceeds the maximum or minimum tolerance value, and output the constraint judgment signal Y:

[0082] Y=(T j ,SOC j ),

[0083]

[0084] In the formula, Y is the output constraint judgment signal, which represents the degree of compliance with the global constraint conditions, T j It is the torque judgment signal, SOC j is the charge judgment signal, T actual is the actual torque, T max is the maximum torque, SOC ideal is the ideal SOC of the input, SOC h and SOC l They are the upper and lower SOC boundaries set for the complete trip.

[0085] Therefore, we must first calculate the actual torque T actual and the maximum torque T max Get the torque judgment signal T j , according to the ideal SOC ideal 、SOC upper limit value SOC h and SOC lower limit SOC l Get the charge judgment signal SOC j , and then based on the torque judgment signal T j And charge judgment signal SOC j Obtain constraint judgment signal Y.

[0086] Fuel consumption mf , condition judgment signal Y, ideal SOC ideal , predicted vehicle speed V fore and the actual vehicle speed V actual The input is sent to the ECMS torque distribution module 33, which establishes the total cost function J of the hybrid vehicle based on the input variables and parameters and with reference to the existing Pontryagin Minimum Principle (PMP). over The entire ECMS torque distribution module 33 iteratively solves the minimum value of the above cost function and the engine distribution torque T at this time. E1 , motor distribution torque T m1 And output to subsequent modules.

[0087]

[0088] u * (t) = argminJ over (x(t),u(t),t),

[0089] In the formula, J over is the total cost function, x(t) and u(t) are the state and control variables in the Hamiltonian function, respectively. In essence, x(t) is the state of charge SOC, and u(t) is the motor torque T m1 , t is time, C f is the fuel consumption cost coefficient, which is determined by the fuel price, C l is the constraint cost coefficient, ranging from 0.8 to 1.4, C s is the SOC deviation penalty factor, which is generally set to 0.1, C V is the speed deviation penalty coefficient, ranging from 0.1 to 1.0, f1, f2, e1, and e2 are constants, where f1 and e1 range from [0,10], f2 ranges from 1.0 to 2.0, and e2 ranges from 0.5 to 1.0. f (t) is the input fuel consumption, Y is the input constraint judgment signal, V fore is the input predicted vehicle speed information, V actual is the actual vehicle speed, SOC ideal is the ideal SOC of the input, SOC ref is the reference SOC value, u * (t) represents the sequence of optimal solutions for the engine and motor torque distribution at a specific time, and arg min is the minimum solution function, which is used to obtain the cost function J over The minimum value at a specific moment and the optimal u(t) sequence at this time can be recorded. The optimal engine torque T allocated in real time can be further determined through this sequence. E1 and motor torque T m1 :

[0090]

[0091] Where, T m1 (t) is the allocated motor torque, u(t) is the optimal control sequence, T E1 (t) is the distributed engine torque, T actual (t) is the actual torque.

[0092] like Figure 1-5 The energy management strategy of the hybrid vehicle is controlled by the driver module 1, the MPC and road section selection module 2, the adaptive strategy adjustment module 3, the LSTM neural network torque compensation module 5, the SOC tracking module 6 and the SSA-GRU neural network road classification module 7, which effectively improves the torque distribution of the hybrid power source of the hybrid vehicle and reduces the torque mutation between the various power sources. At the same time, without affecting the driving function of the hybrid vehicle, the driving cost is reduced and the economy is improved.

[0093] When the hybrid vehicle energy management strategy constructed by the present invention is controlled, the driver module 1 receives the actual vehicle speed V of the hybrid vehicle. actual and the required vehicle speed V req After being adjusted by the vehicle speed controller 11, the power controller module 12 and the driving command module 13, the vehicle driving command sign, the power P corresponding to the engine and the motor are output. E , P m When the signal enters the vehicle speed MPC and road section selection module 2, it is predicted based on the location information and road classification results, so as to effectively predict the future vehicle speed and determine the road type, thereby improving the adaptability of the energy management strategy. In the adaptive strategy adjustment module 3, according to the principle of minimum equivalent fuel consumption, combined with the input ideal charge state SOC ideal , predicted vehicle speed V fore And the feedback torque is used to distribute the torque at the next moment. Before the torque is input to the power source, it is combined with the compensation torque obtained by the LSTM neural network training in the LSTM neural network torque compensation module 5, and finally input to the power source of the hybrid vehicle, realizing the effective suppression of torque mutation. The SSA-GRU neural network road classification module 7 uses data processing methods such as singular spectrum analysis and principal component analysis to clean and reduce the data due to the characteristics of driving data such as disorder, strong volatility, strong randomness, and non-periodicity, and can effectively classify road types, and the results are significant.

Claims

1. A method for constructing an adaptive energy management strategy for a hybrid electric vehicle, characterized in that The following steps are involved: Step 1): The hybrid power source module (4) is composed of an engine module (41), a motor module (42) and a battery module (43). The input of the engine module (41) is the engine required torque T E_r 、Engine power P E , Motor required torque T m_r 、Motor power P m And the battery voltage U batt , the output is the engine speed n E , motor speed n m , the actual speed of the vehicle V actual , the actual displacement of the vehicle X actual , actual torque T actual And the actual state of charge SOC of the battery actual ; Step 2): The actual torque T actual , actual vehicle speed V actual The training sample set of the LSTM neural network is used to train the LSTM neural network using the long short-term memory algorithm to output the compensation torque T m2 ; Step 3): The actual torque T actual , actual state of charge SOC actual And the actual displacement X actual After ideal SOC trajectory planning, the ideal state of charge SOC is obtained ideal ; Step 4): Classify the real-time geographic location GPS information through the SSA group intelligence algorithm and the GRU neural network to obtain the classified reference locations of various types of road sections Step 5): The vehicle speed controller (11), the power control module (12) and the driving instruction module (13) form a driver module (1), and the input of the driver module (1) is the actual vehicle speed V actual , the actual torque T actual , the actual vehicle speed V actual and the required vehicle speed V req The speed error V * And the engine speed n E and the motor speed n m The output is the vehicle driving command sign and the engine power P E And the motor power P m ; Step 6): The road section selection module (21), the Markov vehicle speed prediction module (22) and the MPC module (23) constitute the MPC and road section selection module (2), and the input of the MPC and road section selection module (2) is the vehicle driving instruction sign, the engine power P E 、Motor power P m , actual vehicle speed V actual , the actual displacement of the vehicle X actual And various types of road section reference positions The output is the predicted vehicle speed V fore ; Step 7): The fuel consumption calculation module (31), the constraint condition judgment module (32) and the ECMS torque distribution module (33) constitute an adaptive strategy adjustment module (3), and the input of the adaptive strategy adjustment module (3) is the actual torque T actual , predicted vehicle speed V fore And the ideal state of charge SOC ideal , the output is the engine distribution torque T E1 and motor torque T m1 ; Step 8): The compensation torque T m2 The torque T is distributed to the motor m1 Compensation is performed to form the engine required torque T E_r and motor torque T m_r :

2. A hybrid electric vehicle adaptive energy management strategy construction method according to claim 1, characterized in that: The fuel consumption calculation module (31) outputs the total fuel consumption m f ; The constraint condition judgment module (32) converts the actual torque T actual and ideal state of charge SOC ideal Compare with the constraint value to determine whether it exceeds the maximum or minimum value, and output the constraint judgment signal Y=(T j ,SOC j ), T j It is the torque judgment signal, SOC j is the charge judgment signal, T max is the maximum torque, SOC h and SOC l They are the upper and lower boundary values ​​of SOC respectively; The ECMS torque distribution module (33) establishes a total cost function according to the Pontryagin minimum principle. u * (t) = argminJ over (x(t),u(t),t), by iteratively solving the minimum value of the cost function and the engine distribution torque T at this time E1 , motor distribution torque T m1 ; x(t) and u(t) are the state and control variables in the Hamiltonian function, respectively. x(t) is the state of charge SOC, and u(t) is the motor torque T m1 , t is time, C f is the fuel consumption cost coefficient, C l is the constraint cost coefficient, ranging from 0.8 to 1.4, C s is the SOC deviation penalty factor, set to 0.1, C V is the speed deviation penalty coefficient, the value range is 0.1~1.0, f1, f2, e1, e2 are constants, the range of f1 and e1 is [0,10], f2 is between 1.0~2.0, e2 is between 0.5~1.0, m f (t) is the input fuel consumption, SOC ref is the reference SOC value, u * (t) represents the sequence of optimal solutions for the engine and motor torque distribution at a specific time, and arg min is the minimum solution function, which is used to obtain the cost function J over The minimum value at a specific moment and the optimal u(t) sequence recorded at that moment.

3. A hybrid electric vehicle adaptive energy management strategy construction method according to claim 2, characterized in that: The total fuel consumption of the driving b is the instantaneous fuel consumption rate, ρ f and g are the fuel density and the acceleration due to gravity respectively.

4. A hybrid electric vehicle adaptive energy management strategy construction method according to claim 1, characterized in that: The road section selection module (21) selects the road section according to the power P corresponding to the engine and the motor. E , P m , actual displacement X actual , various types of road section reference positions And the actual vehicle speed V actual Determine the road category where the vehicle is currently located and output road category information j is the pre-processed road type, where j = 1, 2, ..., 5 represents five different road types, d is the distance metric function, and c is the data point, which contains P E , P m 、V actual , X actual Four characteristic parameters, c j is the centroid position of the j-th type of road. The argmin function is used to determine the value of j when d is minimum. is the centroid characteristic parameter value of the j-th type of road predetermined by the k-means clustering algorithm, is the reference position of the jth type of road In k-means clustering, the centroid c j By Iterate until convergence, j = 1, 2, 3, 4, 5, is the center of mass position after iterative update, and when the convergence condition is met C j is the set of all points belonging to the jth cluster.

5. A hybrid electric vehicle adaptive energy management strategy construction method according to claim 4, characterized in that: The input of the Markov vehicle speed prediction module (22) is the actual vehicle speed V actual and driving command sign, the output is the vehicle speed state transfer matrix P e is the state transition probability, S is the set of all values ​​of the parameter corresponding to its subscript at the superscript time, k is the minimum step length defined in the transition range, which is 0.1 or 0.01s, S Vactual * and S sign * is all possible states in the next moment, Indicates the status by Transfer to The number of times, Indicates from the state The total number of possible departures from all transfers.

6. A hybrid electric vehicle adaptive energy management strategy construction method according to claim 5, characterized in that: The MPC module (23) converts the road type information R k and vehicle speed state transfer matrix P e As input, the output is the predicted vehicle speed V fore ={...,V actual (k-1),V actual (k), V(k+1), V(k+2), ...}, V fore Contains actual vehicle speed information and predicted future vehicle speed information V(k+1), V(k+2), ..., V next is the predicted vehicle speed at the next moment, P e is the transition probability, R k is the road type information, and k is the prediction step length.

7. The method for constructing a hybrid electric vehicle adaptive energy management strategy according to claim 1, characterized in that: The driving instruction sign = L(n E ,l(V actual ,a)), L is the table lookup function, workload Engine power required a is the specific acceleration stripped from the acceleration or braking signal, is the maximum output power of the engine, C r is the road rolling resistance coefficient, m is the vehicle mass, g is the gravitational acceleration constant, C d is the air resistance coefficient, A o is the frontal area of ​​the vehicle body, ρ is the air density, and θ is the road slope.

8. The method for constructing a hybrid electric vehicle adaptive energy management strategy according to claim 1, characterized in that: The power control module (12) is based on the formula Calculate the engine P E and motor power P E , P m .

9. The method for constructing a hybrid electric vehicle adaptive energy management strategy according to claim 1, characterized in that: The actual engine torque T E and the actual motor torque T m The actual torque T of the entire hybrid power source actual =[T E ,T m ], the actual engine torque T E =T E_r / η E , the actual motor torque T m =T m_r / η m , engine speed n E =60·P E / (2π·T E ), actual vehicle speed V actual =P E ·R / (i g ·T E ), motor speed n m =60·P m / (2π·T m ), η E is the mechanical efficiency, n E is the engine speed, R is the wheel radius, i g is the transmission ratio, η m is the motor efficiency; The actual state of charge SOC actual =SOC0-∫I batt / Q batt , battery current I batt is the battery current, U batt is the battery pack voltage, P batt is the battery pack power, R int is the battery internal resistance SOC0 is the initial SOC value, Q batt is the battery capacitor.

10. The method for constructing a hybrid electric vehicle adaptive energy management strategy according to claim 1, characterized in that: The ideal state of charge SOC actual is the actual state of charge, T m is the actual torque T actual The motor torque part, X total is the total stroke length, η m is the motor efficiency, U is the battery rated voltage, Q batt is the battery capacitance value.

Citation Information

Patent Citations

  • Plug-in hybrid electric vehicle energy management method based on deep reinforcement learning

    CN108427985A

  • Plug-in hybrid electric vehicle energy management strategy construction method fusing traffic information

    CN113554337A

  • Intelligent driving fuel cell vehicle control method fusing road multi-dimensional information spatialization

    CN115179779A

  • Personalized vehicle operation for autonomous driving with inverse reinforcement learning

    US20230219569A1

  • Energy management for hybrid electric vehicles

    WO2008128416A1

Cited By

  • Vehicle-mounted gravity dynamic error compensation method and device based on intelligent algorithm

    CN120296332A

  • A vehicle-mounted gravity dynamic error compensation method and device based on intelligent algorithm

    CN120296332B