Solar-Assisted Emergency Portable Battery Charger for Electric Vehicles: A Model-Free Nonlinear Integral Backstepping Controller Optimised by Deep Deterministic Policy Gradient

The design, stability analysis and real-time validation of a Solar-Assisted Emergency Portable Battery Charger (SA-EPBC) for use with Electric Vehicles (EVs) is presented in this research. This system uses solar photovoltaic (PV) energy, harvested using a perturb-and-observe MPPT algorithm and a unidirectional dual-active-bridge (DAB) DC-DC converter to provide regulated electricity to charge an EV that has become stranded and is unable to be charged. The output of the converter is controlled by a model-free nonlinear integral backstepping controller (MF-NIBC) which is designed to operate without requiring an explicit model of the dynamic characteristics of the plant, and will therefore remain inherently robust to changes in battery parameters and load transients. The MF-NIBC adapts three scalar controller gains online using a deep deterministic policy gradient (DDPG) reinforcement learning agent, and the critic aspect of this architecture is treated as a continuous-action Markov decision process. A composite Lyapunov function has been used to establish the asymptotic solidity of this closed-loop controller mathematically. Testing of this MF-NIBC controller was completed using hardware-in-the-loop (HIL) experiments on the OPAL-RT OP4510 hardware platform under conditions simulating three different types of EV battery operating voltages (400/400V, 400/300V and 400/500V), and showed that the proposed controller resulted in a voltage settling time of less than 18 ms, maximum overshoot of less than 3.2%, and steady-state error of better than +/-0.5V, each of these variables improved by 62% and 44%, respectively, compared to the baseline proportional-integral controller and model predictive controller. In addition, all test cases resulted in an end-to-end conversion efficiency exceeding 93.1%.