Data-Driven Approach for Accelerating Selective Harmonic Elimination Algorithm in Parallel Power Converters

TL;DR

This paper introduces a neural network-based phase-shifting technique for parallel power converters that quickly identifies optimal switching phases to minimize current ripple, enhancing efficiency and lifespan.

Contribution

It presents a novel deep learning approach with a new dataset generation method to optimize phase shifts in parallel converters in real time.

Findings

Real-time identification of optimal phase shifts achieved.

Significant reduction in implementation time demonstrated.

Maintains performance comparable to classic optimization methods.

Abstract

Current ripple minimization is one of the challenges in parallel converters to increase the capacitor lifetime in various applications. In this paper, a deep neural network-based phase-shifting (PS) technique is proposed for parallel-connected buck converters to minimize the amplitude of a selective harmonic component and facilitate a classic optimum PS at the same time. The proposed method identifies the global optimum point in real time, without the need for complicated computations. The common-link current, common-link voltage, and the duty ratios are selected as the inputs of the neural network to provide the proper phase shifts for the switching signals. To accumulate the required dataset, a Different Start-Same Step (DSSS) technique is also introduced to generate the training data and test/validation data in a separate way. The effect of the number of hidden layers on the network output error is investigated, and a proper number of hidden layers is designed based on a compromise between accuracy and computation efficiency (and execution time). Experimental results prove that the proposed artificial neural network-based PS method preserves the performance of classic optimum PS and minimizing the implementation time significantly.