Experimental validation of a novel power smoothing method for on-grid photovoltaic systems using supercapacitors.

Renewable energy sources have been widely developed in grid-connected systems. However, a challenge to overcome is the random characteristic of renewable resources such as solar irradiance, photovoltaic power fluctuations caused by cloud movement could cause instability of the utility grid. To solve this drawback, several authors have proposed various power smoothing methods for photovoltaic systems using supercapacitors. Nevertheless, sizing optimization and operability of the supercapacitor has not been properly studied. Forecasting power fluctuations is an important strategy to avoid the unnecessary operation of the supercapacitor in certain cases. In this paper, a novel power smoothing method (predictor – corrector) using supercapacitors for a grid-connected photovoltaic system is proposed, the method consists of two stages, prediction and correction. The main novelty of the new method is the use a simple k-means algorithm application model in the cycle estimation stage for supercapacitors, with the aim of selecting representative data of power fluctuations and supercapacitor charge/discharge cycles. Then, for the correction stage, the novel proposed method uses ramp rate algorithms to generate the reference signal to control the state of charge of the supercapacitor. The validation of the new proposed method has been done through exhaustive laboratory experiments under different cloudiness events. The results show that the energy losses when applying the new method are lower with respect to the moving average and ramp rate methods. Furthermore, the number of technical violations is reduced, demonstrating the feasibility of the proposed method to ensure successful mitigation of PV power fluctuations. • A novel method of power smoothing (corrective predictor) based on the moving average and ramp rate techniques is presented. • The method allows to optimize the operability of the supercapacitor by forecasting PV power fluctuations. • Extensive experiments show that energy losses are reduced by 16% than with the traditional ramp rate method. • A learning model is proposed to generate representative data of the power fluctuations and charge/discharge cycles of SC. • The new method is validated by exhaustive laboratory experiments in real conditions. [ABSTRACT FROM AUTHOR]