Extreme Learning Machine-Based Power Forecasting in Photovoltaic Systems

Abstract

With the consuming of fossil fuels, sustainable alternative sources are needed for energy production. Renewable energy is an important source in the fight against climate change and provides economic benefits by increasing energy security. Solar energy is one of the popular renewable energy sources. The photovoltaic system is a technology that produces electrical energy from solar radiation. These systems, which are clean, renewable, and environmentally friendly, are important in terms of meeting our future energy needs. Photovoltaic system power forecasting is an important tool for both energy planning and energy management, as it increases the predictability of bulk power system. In addition, power estimates also allow improving the system efficiency and planning the maintenance date. In this study, due to its high accuracy, fast computation, and easy applicability, Extreme Learning Machine (ELM) method was used to estimate the daily average energy production of a solar power plant located in Elazig/Turkey. To improve the performance of the model, ELM's hyperparameters were optimized using the Modified Golden Sine Algorithm (GoldSA-II). To show the power generation change that occurs in parallel with the seasonal weather change, graphs of the real power of different months from the four seasons and the power estimated by ELM are presented in the study. The radiation, temperature, wind speed, real power, and ELM predicted power values for the ten-month period in which the operating data were obtained are given in the table. By examining this table, the effect of weather conditions on production can be observed more clearly. The R and RMSE values of the ELM model, which are calculated separately for each month, are presented in the form of a radar chart. The average daily R and RMSE values calculated for the ten-month period were calculated as 0.95 and 0.0716, respectively, and the performance of the ELM model used in the study was proven by the calculated R and RMSE values.