一种增强型基于人工神经网络的多相交错DC-DC变换器纹波最小化技术

This paper introduces an optimized implementation of the Phase Current Shape Factors-Artificial Neural Network (PCSF-ANN) technique, which aims to minimize the total current ripple in multiphase DC-DC converters by dynamically adjusting the phase shift between PWM signals. The method is robust against parameter variations, non-identical designs, open-circuit faults (OCFs), voltage changes, and load variations. Two key enhancements are proposed to improve the PCSF-ANN technique. First, a refined data collection strategy uses linear approximations to model triangular phase current waveforms, significantly reducing computational complexity and processing time. Second, the ANN structure is simplified by decoupling OCF detection from the primary ANN and employing smaller, more efficient architectures. This optimization lowers memory usage, enhances execution time, and simplifies implementation. Experimental validation using a two-phase asymmetrical and a three-phase symmetrical DC-DC boost converter demonstrates that the proposed technique delivers ripple minimization performance comparable to the original PCSF-ANN technique while substantially reducing the computational time from 373 μs to 136 μs.