Analysis of the Performance of Different Machine Learning Techniques for the Definition of Rule-based Control Strategies in a Parallel HEV

Two different machine-learning techniques have been assessed and applied to define rule-based control strategies for a parallel hybrid midsize sport utility vehicle equipped with a diesel engine. Both methods include two phases: a clustering algorithm and a rule definition. In the first method, a homemade clustering algorithm is preliminarily run to generate the set of clusters, while the rules are identified by minimizing an objective function. In the second method, a genetic algorithm provides the optimal size of the clusters, while the associated rules are extracted from the results obtained with a benchmark optimizer. The controllers were tested over NEDC, 1015, AMDC and WLTP.