Model predictive controller-based Convolutional Neural Network controller for optimal frequency tracking of resonant converter-based EV charger

Abstract

The rapid establishment of e-transportation is decided by two factors, namely the range and charging time of the Electric Vehicle (EV). The smart and fast charging topology are the main concerns to increase the range and reduce the charging time. The controllers are the heart of the smart EV chargers which decide the charger's optimized performance and reliability. Since the AI based smart charger performs well for fast and dynamic charging of battery, in this article Model predictive controller (MPC) based Convolutional Neural network (CNN) is used as the controller. The EV charger with a Totem pole PFC circuit followed by an LLC resonant DC-DC converter is the proven topology for the efficient and high-power handling capacity of the charger. Initially the PI controller is used for controlling the frequency of the LLC converter and the MPC is developed, finally the CNN controller is structured, trained, validated, and tested to emulate MPC with a similar performance to reduce the computational burden and complexity of the real time implementation. A 7.2 kW LLC-based EV charger with <5 % current ripple and <12 % THD is designed, and the performance is analysed under various load conditions for constant current and constant voltage charging. The detailed analysis of the EV charger for PI, MPC and DNN controllers is done, and the measurement of the performance parameters ensures the robustness.

Keywords

• Deep neural network
• EV charger
• LLC resonant converter
• Model predictive controller