Model-Free Predictive Control of Power Converters with Multifrequency Extended State Observers

Abstract

The performance of model-free predictive control (MFPC), which is based on the ultra-local model, is highly dependent on the effectiveness of the extended-state observer (ESO). The standard linear ESO has a high-gain which enhances disturbance rejection attributes. Nevertheless, this ESO has two main limitations: i) it amplifies high-frequency measurement noises, and ii) it deteriorates disturbance rejection performance when the noise immunity is improved. Therefore, this article introduces a new technique to improve both the disturbance rejection and noise suppression of ESO without increasing the bandwidth: the multi-frequency-based ESO (MF-ESO). Two novel ESO structures that operate on the MF-ESO principle are analyzed: parallel and cascade ESOs. Extensive frequency domain analyses unravel the comparative capabilities of the MF-ESO structures for improved disturbance rejection and/or high-frequency noise suppression. A novel adaptive gain is proposed to improve CESO's dynamic performance. The superior features of the proposed MF-ESO are experimentally validated by the model-free predictive control of a grid-connected power converter.