Validation of Genetic Algorithm Optimized Hidden Markov Model for Short-term Photovoltaic Power Prediction

Abstract

Significantly, incentives and energy guidelines are expanding the deployment of renewable energy (RE) systems in developing countries. A substantial amount of RE-based electrical power is generated over the last ten years, due to global warming issues. Solar photovoltaic (PV) is being incredibly utilized because of its boundless quality. However, the inherent intermittency of PV power production at high penetration level to the grid leads to complications related grid reliability, stability and transportable unit of electric power. A viable approach to addressing this problem is to develop a reliable power forecast model for the short-term horizon related to scheduling and transmission. Based on an existing forecast model built on genetic algorithm (GA)-optimized hidden Markov model (HMM), this paper implements the model validation process using more recent input dataset. Model evaluation is based on the computation of normalized root mean square error (nRMSE). As the validation result, HMM+GA is sufficient to accurately forecast PV P o under clear sky day (CSD) condition. Contrariwise, for cloudy days (CDs) presenting instantaneous changes in solar irradiance ( G s ) between some hours of the day, HMM+GA adapted with a correction factor (x); articulated as HMM+GA+x; is adequate to forecast the P o more precisely when the average change in the absolute value of G s () in the morning () is greater than 128% and/or when in the evening () exceeds 90%. Particularly, the average nRMSE of 2.63% showed that HMM+GA with or without x are suitable techniques for forecasting PV P o on an hourly basis. Therefore, the validation results are in harmony with those of the baseline models. https://dorl.net/dor/20.1001.1.13090127.2021.11.2.28.7