A Neural Network-Based Control Algorithm for a Hydraulic Hybrid Powertrain System

Abstract

Significant research efforts are invested in the quest for solutions that will increase the fuel economy and reduce the environmental impacts of ICE-powered vehicles. The main objective of the study presented in this paper has been to analyze and assess the performance of a control methodology for a parallel hydraulic hybrid powertrain system of a transit bus. A simulation model of the vehicle has been calibrated by analyzing data obtained during an experiment conducted in real-world traffic conditions aboard a Belgrade transit bus. A Dynamic Programming optimization procedure has been applied on the calibrated powertrain model and an optimal configuration that minimizes the fuel consumption has been selected. A Neural Network-based, implementable control algorithm has then been formed through a machine learning process involving data from the optimal, nonimplementable Dynamic Programming-based control. Several Neural Network configurations have been tested to obtain the best fuel economy for the range of conditions encountered during normal transit bus operation. It has been shown that a considerable fuel consumption reduction on the order of 30% could be achieved by implementing such a system and calibration method.