Self-learning Online Energy Management Strategy For Plug-in Hybrid Electric Bus

Plug-in hybrid electric vehicles can incearse the efficiency of the propulsion system and reduce the use-cost of vehicles.To make full use of advantages offered by plug-in hybrids,the design of energy management strategy is crucial.The ultimate goal of energy management strategy is to implement online while achieving close to optimal results.Therefore,it is necessary to summarize and analyze the global optimal results.This dissertation systematically studies and applies the knowledge remained in the optimal results to further improve the decision-making effect,according to the different usage scenarios of buses.For a determined route,learn from the optimal engine output power sequence.The reason for finding new neural network input features is first described using a single driving cycle.The constructed concept of length ratio and the vehicle acceleration are then qualitatively added to the neural network inputs,which verifies that neural network energy management strategy is feasible to learn from engine power sequence.To quantitative determine the best input features of the neural network,a methodology using the classification accuracy of random forests for selecting input features is proposed.The four-feature neural network energy management strategy is developed,which achieves 1%-3% deviation from the global optimum on the trained and untrained driving cycles.The condition when to retrain the neural network is given,and the reason why the strategy is not suitable for variable bus routes is explained.For variable bus routes,learn from the optimal SOC curves.Energy planning methods suitable for small samples and large samples are designed based on partial trip information such as mileage,average speed,high/low-speed ratio.A SOC reference curve is generated when the trip starts,and the energy management problem is converted to the SOC following problem.By analyzing the characteristics of the optimal engine output power sequence,the fuzzy rules and elaborate decision rules of the engine output power are then formulated,and the adaptive fuzzy logic energy management strategy and data-driven hierarchical energy management strategy are designed.The data-driven hierarchical energy management strategy is recommended because the elaborate decision rules ensure that both the engine and the drive motor work in the high-efficiency region,which achieves 1%-3% deviation from the global optimum on the trained and untrained driving cycles.Finally,different strategy models are built in the Simulink environment,and the C code of each strategy is obtained using automatic code generation.After security verification of the C code and code modification,each energy management strategy can be embedded to the Infineon Aurix TC275.The execution time is measured under the same microcontroller configurations.The single-pass decision-making time of each approach is less than 0.1 milliseconds,which verifies that the energy management strategies proposed in this paper fully satisfy the need for vehicle control.