Optimal Number of Turns Design of IPT for Maximum Power Efficiency base on Reinforcement Learning with DQN

This paper proposes the method to find the optimal number of turns in IPT for a maximum power efficiency by using deep Q-learning network (DQN) based on reinforcement learning (RL) algorithm. Since it is barely possible to get an optimal number of turns for a transmitter (Tx) and a receiver (Rx), satisfying to operate at a maximum power efficiency, the most of Tx & Rx turns are normally wound until the coils are occupied over the cores. Moreover, it requires extremely long time to iteratively proceed to simulate all existing combinations of the Tx & Rx coil windings to derive the maximum power efficiency. To minimize the time to compute the number of coil-turns to get the maximum power efficiency, the proposed DQN algorithm can select the optimal coil turns having high Q-value through ε-greedy turn selection process. After a few episodes of the neural network system, the expected power efficiency by the proposed RL can reach to the maximum power efficiency by less than 20% of all possible combinations. The proposed RL algorithm has been evaluated by FEM simulation analysis, showing that the optimal number of turns for various WPT cases having different loads can be easily found.