Stochastic Charge Planning with Dynamic Programming

Abstract

The development of charge planning algorithms which extend further than those considering uncorrelated disturbance models and produce robust policies is an important subject. The freight sector is moving towards battery electric trucks where uncertainties can have a major impact on missions due to state of charge constraints. Therefore, this thesis investigates dynamic programming algorithms for use in charge planning. Disturbances are modeled as Gaussian processes, which for certain structures admits an equivalent transformation to an LTI SDE system. Using this transformation, the distribution along state trajectories are estimated using an unscented Kalman filter. The UKF showed good performance for the modeled disturbances, with a largest mean bias of 1.106% in a worst-case scenario. The proposed approximate dynamic based charge planning algorithm became robust under stochastic external uncertainties from wind and traffic by implementing chance constraints. The proposed planning algorithms achieved better performance than both a simpler deterministic dynamic programming algorithm and a simple heuristic planner. Computational complexity remains a key concern for real time implementations and is a crucial challenge when designing stochastic charge planning algorithms.