Study on Current Control of Single-Phase Inverter Using Deep Q-Network

Generally, power electronics circuit is controlled by proportional integral (PI) controller for regulating voltage and currents, which include inverters for connecting to power grid. Output voltage or current of inverters is controlled to sinusoidal wave with PI control. In PI control, the ideal gain is determined according to the circuit parameters. Strong nonlinearities such as magnetic saturation of inductors and dead times of the power switches can make it difficult to determine the gains. This paper presents a control using Deep Q-Network for the inverter. Deep Q-Network is combination of Q-learning and Deep Network, which can act along with changing parameters via learning. The learning is carried out and gets a neural network for Q-values in simulation. Controller based on the neural network is tested in experiments, which proves that the controller performs well.