Research On Adaptive Energy Management Strategy Of Fuel Cell Vehicle Based On Traffic Information And Driving Style

As one of the important forms of new energy vehicles,fuel cell vehicle has become the research focus of many automobile manufacturers and scientific research institutions due to its advantages of zero emission,high energy conversion efficiency,and short fueling time.However,the current poor durability and high cost of fuel cells have become the main factors restricting the rapid commercialization of fuel cell vehicle.As one of the important vehicular control strategies,the energy management strategy plays a critical role during the operation of the vehicle.It distributes the required power among the power sources according to different optimization objectives.And it has a significant impact on the fuel economy and durability of fuel cell vehicle which operating under different driving styles and road conditions.Based on the above background,this paper conducts in-depth research on the energy management strategy of fuel cell vehicle in order to optimize the fuel economy and durability of the vehicle under the premise of satisfying the power performance,and improve the self-adaptability of energy management strategy to traffic information and driving styles at the same time.This paper takes an indirect fuel cell vehicle as research object,and carried out the research on powertrain modeling,real-time optimization energy management strategy,vehicle speed prediction based on traffic information and driving style,learning-based energy management strategy and hardware-in-the-loop test.The concrete research contents are as follows:(1)A fuel cell powertrain simulation model is established.Based on the objective vehicle,the indirect fuel cell-lithium-ion battery powertrain is selected as the configuration of the powertrain model.Then,the parameters of the main components of the powertrain are matched and selected based on the vehicle parameters and dynamics performance indexes,including the driving motor,fuel cell and lithium-ion battery.Based on the matched component parameters,the powertrain simulation model is built in the MATLAB/Simulink.The dynamic performance of the powertrain model is verified under the power following strategy,and the results show that the maximum vehicle speed,the 0 to 100 km/h acceleration time and the maximum gradeability all meet the requirements of the vehicle dynamic performance indexes.(2)The real-time optimal energy management strategies are designed based on the Pontryagin’s Minimum Principle(PMP)and Model Predictive Control(MPC)theory.The first part is the design of the PMP strategy.The theoretical model of power distribution optimization problem is constructed based on the PMP,including the establishment of Hamiltonian function and objective function.The initial values of costates under different typical driving conditions are determined by the shooting method.The PMP strategy is simulated under three driving conditions of UDDS,NEDC and HWFET and compared with results of the power following strategy,which verifies the high optimization of the PMP strategy in terms of power distribution,fuel economy and fuel cell durability.The results of PMP strategy are also used as a benchmark for comparison with the following energy management strategies.The second part is the design of MPC strategy.The theoretical model of the power distribution optimization problem is constructed based on the MPC theory,including the construction of the vehicle speed prediction model,the selection of the optimization algorithm in the prediction horizon,and the establishment of the objective function.The Markov chain is used as the vehicle speed prediction model,and the predictive step length is 10 s.The globally optimal dynamic programming algorithm is selected as the optimization algorithm in the prediction horizon.In the objective function,the degradation cost of fuel cell dynamic load change and the equivalent hydrogen consumption cost corresponding to the lithium-ion battery SOC change are considered.The MPC strategy is simulated under above three driving conditions and compared with the power following strategy and the PMP strategy.The results show that the MPC strategy shows high optimal performance in terms of fuel economy,dynamic load change of fuel cell system output power and lithium-ion battery SOC maintenance.(3)The driving style classification and recognition algorithms are designed and the traffic simulation model is established.The first part is the design of driving style classification and recognition algorithms.Thirteen characteristic parameters related to vehicle speed and acceleration are selected as the basis for driving style recognition,and the NGSIM dataset is used as the data source for the driving style classification and recognition algorithms.Principal component analysis method is used to reduce the dimension of the characteristic parameters,and the first four principal components are selected as the input of the driving style classification algorithm.Then,K-means clustering algorithm is used to classify different driving data into three types of driving styles: mild,normal,and aggressive.Finally,a driving style recognition algorithm is designed based on Recurrent Neural Network(RNN),and the classified driving style data is used as the training data and test data of RNN,in which four principal components are used as input parameters,and the corresponding driving styles are used as data labels.Through multiple training and optimization,the final recognition rate of the algorithm reaches 99.10%,which verifies its effectiveness.The second part is the establishment of the traffic simulation model.Based on Vissim,a traffic simulation model which includes main elements such as roads,traffic lights,and vehicles is built.Then,three copies of driving data are extracted through multiple simulations based on this model,which are used as the basis for the design of following speed prediction algorithm and energy management strategy.The driving data includes the speed and acceleration of the current vehicle and the speed and acceleration of the preceding vehicle.(4)The vehicle speed prediction algorithm and the learning-based energy management strategy are designed.Firstly,the vehicle speed prediction algorithm is designed based on Long Short-Term Memory network(LSTM).Through analysis and comparison among LSTMs with different hidden layer node,input step length,and output layer node,the network structure with the optimum overall effect of prediction performance and computational efficiency is finally determined.Then,the driving data extracted from the traffic simulation model is used as the training and testing data of the prediction algorithm.The driving style coefficients,the speed and acceleration of the current vehicle,the speed and acceleration of the preceding vehicle are used as the input of LSTM,and the future driving speed is used as the output.The final prediction results show that the predicted vehicle speed can effectively follow the change of the real vehicle speed,and the algorithm prediction accuracy is high.The second part is the design of energy management strategy based on the reinforcement learning Deep Deterministic Policy Gradient(DDPG)algorithm.The lithium-ion battery SOC,vehicle speed and acceleration are used as the state variables of the strategy,and the fuel cell system output power is used as the control variable.In the design of the strategy objective function,four parts are considered: the instantaneous hydrogen consumption of the fuel cell system,the change and maintenance of the lithium-ion battery SOC,the efficiency of the fuel cell system and the power load change of the fuel cell system.The strategy was simulated under three typical driving conditions and three driving styles’ driving conditions.The simulation results are compared with the results of PMP strategy,which show that the DDPG strategy has obtained good effect in the suppression of fuel cell power dynamic load change,the maintenance of lithium-ion battery SOC change and the fuel economy of the vehicle,reaching the approximate optimal solution.And the results also show that the DDPG strategy can learn the characteristics of different driving styles and the changing rules of traffic information from the speed predicted based on driving styles and traffic information based on its own structure and learning ability,which means it is self-adaptive to these two factors.(5)Hardware-in-the-loop test verification of the energy management strategy.Under different driving conditions,the hardware-in-the-loop test verification of PMP strategy,MPC strategy and DDPG strategy is carried out,which mainly includes the construction of the hardware-in-the-loop test platform,the debugging and import of the powertrain simulation model,and the compilation and flash reprogramming of the energy management strategy.The strategies are verified under different driving conditions,and the results show that the above energy management strategies have achieved good power distribution performance and real-time performance,which verify their effectiveness in the actual controller environment.It can be seen from the above research work that the energy management strategies proposed in this paper can effectively improve the fuel economy and durability of the vehicle while satisfying the power performance,and have high real-time performance and self-adaptability,reflecting its application value.