Machine Learning Based Controlled Filtering for Solar PV Variability Reduction with BESS

Abstract

The intermittent nature of solar power prevents the large-scale penetration of Photovoltaic (PV) systems in the utility grid as it causes various irregularities such as voltage fluctuations, frequency deviations, and reduced overall output power quality. This paper introduces a novel smoothing control methodology for firming of PV power fluctuations. Battery Energy Storage System (BESS) is coupled with solar panel arrangements and included into the grid for solar power smoothing and to stabilize the above-mentioned irregular behaviors. Additionally, smoothing filters such as Low Pass Filters (LPFs) are integrated along with the BESS for optimal functioning and cost reduction. It has been established that the time constant of a LPF directly impacts the degree of solar PV smoothing. Thus, the proposed methodology utilizes the concepts of machine learning and model predictive control to design a control system that intelligently controls the LPF time constant to efficiently rid the PV profile from fluctuations while operating under practical constraints. A high accuracy prediction system is also developed using neural networks. The proposed controller can flatten solar power variations by utilizing the inputs from our prediction system. In addition to the smoothing performance of our controller, the effect on the battery ramp rate and state of charge is also observed. The proposed firming concept has been described theoretically and simulation results have also been demonstrated.