Research On EMS For Fuel Cell Hybrid Heavy-Duty Truck Based On C-V2X Platform

The increasing severe environmental pollution and energy shortages have accelerated the commercialization of new energy vehicles.As the most promising clean energy vehicle,the fuel cell hybrid vehicle(FCHV)has the advantages of high efficiency,zero-emission,short hydrogenation time,and extended cruising range.It has become the primary task of researching and developing in various countries.As one of the critical technologies of the FCHV system,energy management strategy(EMS)dramatically influences the vehicle’s dynamic,economy,and adaptability.With the continuous advancement of intelligent network technology,the Energy Internet can improve energy utilization efficiency and achieve energy conservation as well as emission reduction for the hybrid vehicle.How to gather and mine valuable information of the C-V2 X platform,utilize artificial intelligence technology,and eventually improve the energysaving level and working condition adaptability of FCHVs in market segments.It will be the problems and challenges faced by the EMS of the FCHV.This paper’s research object is a fuel cell hybrid heavy-duty tractor with a46-ton load,which has been put into demonstration operation on Shanghai City Road and Expressway.Based on data mining theory,optimal control principle,machine learning algorithms,and simulation experiments,we explore the modeling of FCHV,the processing and mining method of driving data based on C-V2 X,the intelligent prediction method based on data-driven,and the intelligent EMSs.The research mainly includes the following essential contents.1)In order to improve the prediction accuracy of future driving conditions,a data-driven intelligent prediction method is established based on data processing and mining of driving conditions on the C-V2 X platform.Firstly,the vehicle speed prediction method based on the Markov chain is studied.The prediction driving condition in different time domains is obtained by solving the n-step transition probability matrix.Secondly,a closed-loop speed prediction model is proposed to be integrated with APSO-LSSVM and BPNN,combining the advantages of the support vector machine and BP neural network,to predict the synthetic driving cycle.Eventually,the prediction accuracy of the two methods in different time domains is compared.As a result,when the time domain is set to 5s,the prediction accuracy of APSO-LSSVM and BPNN is 6.74% higher than that of the Markov chain,and the training time is 28.57% shorter than that of BPNN.It indicates that the closed-loop speed prediction method has higher prediction accuracy and lays a good foundation for the design of EMSs.2)In order to improve the prediction accuracy of future driving conditions,based on data processing and mining of driving conditions on the C-V2 X platform,a closed-loop speed prediction model is proposed to be integrated with APSOLSSVM and BPNN,combining the advantages of the support vector machine and BP neural network,to predict the synthetic driving cycle.As a result,when the time domain is set to 5 s,the prediction accuracy of APSO-LSSVM and BPNN is6.74% higher than that of the Markov chain,and the training time is 28.57%shorter than that of BPNN.It indicates that the closed-loop speed prediction method has higher prediction accuracy and lays a good foundation for the design of EMSs.3)In order to improve the applicability of EMS in multi-directional information of working conditions,an EMS framework based on AECMS-MPC hierarchical prediction is put forward by integrating the information of historical and future dimensions of driving conditions.Firstly,the top-level control investigates the approximate optimal SOC reference trajectory from the perspective of global planning based on historical data.Secondly,the bottom control explores the adaptive power allocation based on the AECMS-MPC algorithm,from the perspective of local optimization based on future prediction.Eventually,the performance of the proposed strategy is verified.Compared with FL-EMS,hydrogen consumption reduces by 11.32%,and the calculation time increases by 9.5 times.Compared with the DP-EMS,hydrogen consumption increases by 0.47%,and the calculation time reduces by 72.80%;besides,the hydrogen consumption increases by 3.35%,7.56%,and 4.33% in CBDTruck,NYCC,and New York Bus respectively.It shows that the AECMS-MPC hierarchical predictive EMS is able to achieve the approximate global optimal effect in the appropriate time and has strong adaptability to conditions with similar characteristics.4)In order to improve the optimality,working condition adaptability,and real-time applicability of the EMS,a framework based on D3QN-DRL is proposed by combining the advantages of DDQN and the dueling network.The DDQN improves the high valuation phenomenon and convergence by decoupling the action value function.With the reconstruction of the network structure by the dueling network,the trade-off of multi-objective optimization indicators is realized,and the stability of the system is improved.In addition,the AAP criterion is introduced into the reward function,which effectively delays the degradation of the system life.Eventually,the performance of the strategy is verified.Compared with the FL-EMS,hydrogen consumption reduces by 17.49%,and the calculation time increases by 6.75 times.Compared with DP-EMS,the hydrogen consumption reduces by 6.52%,the calculation time reduces by 80.72%,and the hydrogen saving rate is as high as 16.18%,9.06%,and 13.72% in CBDTruck,NYCC,and New York Bus respectively.It shows that the D3 QN deep reinforcement learning EMS has high fuel economy,strong working condition adaptability,and online real-time application.5)A hardware-in-the-loop test platform is established to verify the performance and effectiveness of the proposed strategy in a real-time environment.As a result,both the AECMS-MPC and D3QN-DRL are able to effectively improve fuel economy and adaptability.