MAXIMUM RIPPLE ELIMINATING MULTI-LEVEL BIDIRECTIONAL DC-DC CONVERTER VIA OPTIMIZED DEEP LEARNING NETWORK

In this research, a novel deep learning based Multi-level Bidirectional DC-DC converter using a modified Emperor Penguin Optimized multi-graph neural network (MEPO-MGN2) has been proposed. Grid power is converted to direct current by means of an AC-DC converter, and then a bidirectional multi-level DC-DC converter (MLDC) is used for the power transfer in both directions between the grid and PMDs. The proposed method utilizes an attention multi-graph convolution network (AMGCN) for the duty cycle (α) prediction of MLDC to regulate the flow of power between the grid and PMD, vice versa. This system employs multiple Proportional-Integral (PI) controllers to manage voltage and current for stable, optimized power transfer. Modified emperor penguin optimization (MEPO) is used to tune the parameters of MLDC, thereby achieving a well-regulated DC output voltage in the charging stations. The MEPO-MGN2 performance is evaluated, and implementation is carried out using the MATLAB platform. The efficacy of the proposed method is 98.5%, which is better than the other existing converter techniques.