Reinforcement-Learning-Based Coordinated Control of a Bidirectional Multiport Converter for Sustainable Electric Transportation Systems

Abstract

The electrified transportation sector is rapidly becoming more reliant on highly efficient charging infrastructure that can interact with the utility grid and coordinate renewable generation and energy storage as well as a varying load from vehicles. Existing multiport converters often suffer from problems such as direct current (DC)-link voltage excursions, poor charging current tracking, high total harmonic distortion (THD), and source-sharing inefficiency in intermittent operating conditions. This work presents a smart intelligent bidirectional multiport power converter topology connecting photovoltaic (PV) generation, wind generation, a battery energy storage unit, an electric vehicle (EV) charger and an alternating current (AC) grid by means of an 800 V DC link. Reinforcement Learning (RL) controls converter duty ratios and grid-side modulation via voltage error, renewable power, battery state of charge (SoC), electric vehicle EV current and components of the grid-current vector as well as harmonic distortion. PV extraction, wind conversion; battery polarization; Constant Current–Constant Voltage (CC–CV) charging over the entire powertrain, including synchronous-reference-frame grid control in addition to component-level efficiencies and THD, are modeled quantitatively using systematically derived equations. In terms of simulation-platform evaluation, Simulation results demonstrate with each unit providing a settling time of 0.22 s, a maximum DC-link deviation of $0.75\%$, charging-current error, efficiency, THD power-sharing error and SoC-regulation error all being, respectively, equal to ${1. 1 0 \%, 9 7. 2 0 \%, 2. 2 0 \%, 1. 4 0 \%}$, and $0.90\%$.