Model-Free Multiphase AC Signals Downscaling Method Using Orthogonal Phase-Lock Autoencoder and Circular Harmonic Decomposition

Abstract

Reference frame transformation methods are generally introduced in multiphase ac systems, so as to simplify system’s high dimensional ac signals into low dimensional dc ones. Classical methods, such as Clark/Park transformations, are used in balanced systems whose models are known, but not suitable for model-unknown systems which may contain unbalanced components. To address this issue, this article proposes a model-free method, that can downscale both balanced and unbalanced multiphase ac signals into dc ones, without any access to prior model knowledge. The proposed method follows the principles of low rank and sparsity, and consists of two components: 1) a phase-lock orthogonal autoencoder is designed, to transform multiphase ac signals into two-phase orthogonal ac signals; 2) a circular harmonic decomposition is introduced, to further simplify two-phase orthogonal ac signals into some dc constant values. The effectiveness of the proposed method is validated on a real-world model-unknown system, and the experimental results also indicate that the proposed method benefits of discovering some potential patterns of the system, which is helpful for an in-depth study of model-unknown systems. Comparative studies show that the proposed method is superior to classical methods in terms of interpretability, information integrity and scalability in model-unknown scenarios.