Neural network based power management of hydraulic hybrid vehicles

Abstract

Effective power management is key to maximizing the performance and efficiency of hydraulic hybrid powertrains. However, the strong influence of future driving events on the optimal control policy limits the effectiveness of many approaches investigated to date. To address this issue the authors have proposed and investigated a novel power management controller that aims to predict online the accumulator’s near optimal state trajectory. It is demonstrated in this paper that if the optimal accumulator state trajectory is known, then an implementable control scheme can achieve near globally optimal fuel efficiency. Controller development began by optimally controlling a series hybrid over a representative drive cycle using Dynamic Programming (DP). A Neural Network (NN) was then trained to reproduce the DP optimal accumulator pressure trajectory based on the vehicle’s velocity over the previous thirty seconds. In this way the NN generalized the relationship between vehicle velocity and accumulator pressure. The NN power management controller’s performance was then evaluated on a hardware-in-the-loop transmission dynamometer using untrained drive cycles to demonstrate the generality of the proposed approach. During these untrained evaluation cycles the NN controller was able to decrease average fuel consumption by 25.8% when compared to a baseline constant pressure control strategy.