Research On Stochastic Dynamic Programming Based Energy Management Strategy Of Hybrid Vehicles

Hybrid electric vehicles(HEVs)are considered as a solution for energy shortage and air pollution problems nowadays.How to improve the fuel economy of a HEV is an important research direction.Nowadays,the energy management strategy(EMS)with known global driving cycle is mature,but the EMSs for real-time application,and online application remain to be solved.To meet this demand,we selected a single-shaft parallel HEV as the research object.After the modeling of the parallel HEV for EMS designing and whole vehicle performance test,we employed a dynamic programming(DP)based energy management strategy in three typical urban driving cycle,and get the vehicle performance in fuel economy,state of charge(SOC)range,the SOC in the end of the cycle,average engine efficiency,average motoring efficiency,and average generating efficiency.To meet the limitations of the dynamic programming EMS method that requires the full driving cycle information in advance,we have employed and realized three algorithms to the EMS.A stochastic dynamic programming(SDP)based EMS is studied.Firstly,the torque demanded by the driver,discretized nonlinearly,is modeled as a one-step Markov chain to represent the unknown influence of future driving cycle.After the tuple data collection of “state-action-instant cost”,a policy iteration method is used to calculate the optimal strategy.The SDP based EMS calculates the instant SOC cost according to different section of the SOC that meet the actual use of the battery.Experimental results under three different driving cycles have shown the adaptation ability of the SDP method.The control strategy of the SDP based EMS is a table.In real application,we can obtain the optimal control action by look-up this table.The SDP based EMS has the disadvantage of high computation burden for the adoption of a strategy iteration method.To meet this demand,an orthogonal wavelet basis nueral dynamic programming is proposed in this research.Firstly,an orthogonal wavelet neural network is proposed.We also present how to determine the structure of the orthogonal wavelet basis neural network,and how to determine the activation function in the hidden layers,the weights of the network,and the parameters of the wavelet function.Subsequently,the orthogonal wavelet neural network is applied to the critic network of orthogonal wavelet basis neuro dynamic programming energy management strategy.Finally,a correction module is used to correct the control action of the output of the action network,and get the real control action.Experimental results show that this method has the potential for online application.The SDP based EMS is based on a one-step Markov chain of the driver's demanded torque,which will lead to a model error.To meet this demand,a stochastic model predictive control(SMPC)based EMS is proposed.Firstly,a prediction model of the state space for the HEV is established.And then a DP method and a multi-step temporal difference algorithm are both employed in the rolling optimization for reducing computation burden.The experimental results under three different urban driving cycles indicate that the SMPC based EMS can get better fuel economy and vehicle component efficiency for the characteristic of feedback correction and rolling optimization.The experimental results of a DP based EMS,a SDP based EMS,an OWNDP based EMS,and a SMPC based EMS indicate that the SMPC based method and the OWNDP based method can get suboptimal fuel economy and vehicle component efficiency;compared with the OWNDP based EMS and the SMPC based EMS,the SDP based EMS is least optimal for the discrete error and the model error of the Markov chain.