Data-Driven Analysis of the Correlation of Future Information and Costates for PMP-based Energy Management Strategy of Hybrid Electric Vehicle

Abstract

In control problems of hybrid electric vehicles, concepts using Pontryagin's minimum principle produce near-optimal solutions for minimizing fuel consumption. The costate in these control concepts can be interpreted as a parameter that represents the value of the electrical consumption because it is used for calculating the equivalent fuel consumption of the electric use. Therefore, it is possible to balance the state of charge (SOC) of the battery by adjusting the costate. In this study, an analysis was conducted to determine the correlation between future driving information and the costate by investigating the simulation results from different costate values. To analyze the impact of the driving conditions on the SOC balance, the driving cycles are classified into three groups, such as city, rural, and highway, using a support vector machine based on a supervised learning algorithm that categorizes the cycles with the hyperplane constructed from training data labeled in advance. Based on the analysis of the results, it is shown that the costate have different characteristics according to the classified future driving, and stochastic models for optimal costates can be obtained according to the categorized driving groups. The approximation model produced from the data-driven analysis makes it possible to design a controller that determines an appropriate costate according to upcoming future driving conditions such that a real-time controller using updated costates can be developed if the future driving conditions are provided by navigation systems and connectivity technologies.