Estimating Generated Power of Photovoltaic Systems During Cloudy Days Using Gene Expression Programming

Abstract

Short-term irradiance variability because of the passing clouds of unknown size, direction, and speed is a key issue for power grid planners because of the unexpected fluctuation in the generated power of photovoltaic (PV) systems. In order to handle this issue, several models have been presented in the literature to estimate the variability of the PV systems output power during cloudy and partial shading events. However, the estimation error of all presented models in the literature is relatively high. To comply with the utility guidelines of limiting the PV systems generated power variability to a level less than 10% of the PV systems capacity per minute, a more accurate power estimation model is essential to precisely calculate the required energy storage backup. In this article, a new model to estimate the output power of a group of rooftop PV systems during cloudy events using only one sensor is proposed. In this regard, two new strategies are adopted. First, day time is divided into three-time segments in which each segment exhibits almost the same weather conditions. Second, the data of each time series are analyzed using the wavelet transform to divide them into high and low-frequency modes. The frequency modes are then used to train a gene expression programming model to estimate the entire generated power of the PV systems. The proposed model features higher accuracy than other models presented in the literature. A sensitivity analysis is performed to quantify the effect of various parameters on the accuracy of the proposed model whose robustness is validated through practical data collected from a group of distributed rooftop PV systems in Brisbane, QLD, Australia. Results reveal that the maximum average mean absolute error of the proposed model is 7.49%, whereas it is 12.12% for the existing models in the literature.