Hour-ahead Forecasting of Photovoltaic Power Output based on Hidden Markov Model and Genetic Algorithm

It is well known that the variability in PV power output is primarily owing to fluctuations in radiation received by the solar panels. Forecasting in the short-teen horizon particularly is very crucial to power quality and power schedules such as load drop or gain, and power dispatch planning. This study details an innovative method based on ordinary model (Hidden Markov Model, HMM) and HMM optimized with Genetic Algorithm (GA) for hour-ahead forecasting of the power output (P-o) of a 1.2 kW PV system. Solar irradiance, module temperature acquired by mathematical modelling and wind speed were used as initial forecast data. The model testing and validation was built on the computation of normalized Root Mean Square Error (nRMSE). As the results, GA-optimized HMM is able to forecast P-o an hour-ahead with low nRMSE than HMM under clear sky day (CSD) condition. However, the abnormalities of the forecasting model resulting from instantaneous fluctuations in solar irradiance under cloudy day (CD) condition were decreased with correction factor (xi). It was deduced that if the average change in the absolute value of solar irradiance (vertical bar(Delta Gs) over bar vertical bar) is more than 128% and 90% in the morning and evening times respectively, the GA-optimized forecasting model with or without xi presents average nRMSE of 2.33%. Therefore, HMM+GA gives more accurate P-o forecast for CSDs whereas HMM+GA+ presents the best P-o for CDs, supporting the consideration of the proposed forecast model as a good technique for hour-ahead power output forecasting of PV system.