Energy Management Strategies For Hybrid Electric Bus Based On Stochastic Power Demand Prediction

With increasing car ownership, exacerbating environment pollution and energy crisis, developing hybrid electric vehicle becomes more and more important. A hybrid electric vehicle has a combustion engine and one or more electric motors as power sources. It combines many advantages of conventional vehicle and electric vehicle, such as high fuel economy, low emissions, long mileage. Thus, it is suitable to use hybrid electric vehicles as city bus.Nowadays, hybrid electric bus is becoming more and more popular in China. However, the fuel economy of massive running hybrid electric buses isn’t as good as expectation, because of poor energy management strategies. In this dissertation, three different energy management strategies is presented for hybrid electric buses in order to improve fuel efficiency and reduce emissions. The main contributions of the dissertation are summarized as follows:First, utilizing the fact that a bus runs on a fixed route, a position dependent energy management strategy is proposed. With the history data of a bus, a position dependent non-homogeneous Markov peed transition matrix is acquired. Combining vehicle system model and the transition matrix, the energy management problem on hybrid electric bus can be transferred to a Markov decision process (MDP). The solution can be solved offline by dynamic programming and saved in a look-up table for real-time implement. The simulation results suggest that the MDP based strategy can obviously improve fuel economy and reduce emission.Second, because the MDP based strategy can’t be solved online and can’t automatically adapt to the specific bus traffic condition, a self-learned energy management for hybrid electric bus is proposed. Action-value function approximation and Q-learning algorithm are used to achieve the energy management strategy. The simulation results indicate that the Q-learning based strategy can achieve good fuel economy, and a comparison between Q-learning based strategy and MDP based strategy is also made.Third, a driver behavior model is tried to be involved in designing energy management strategy. Multi-model approach is used to describe driver behaviors, and related model parameters are acquired with least square method. Some simulations are provided to verify the proposed driver behavior model. At last, simulation results show that driver behavior based strategy can achieve better fuel economy than rule-based strategy.