Hourly Global Solar Radiation Estimation Based on Machine Learning Methods in Eskisehir

Abstract

Due to the increasing importance of knowing the amount of global solar radiation (GSR) that is incident on solar panels; short term data, such as hourly global solar radiation (HGSR), is essentially required to obtain more accurate and reliable power generation prediction. Nowadays, Machine Learning (ML) methods are becoming a huge trend for data forecasting. Therefore, in this paper, a comparison between Collares-Pereira & Rabl empirical model modified by Gueymard (CPRG) and ML methods for HGSR estimation in Eskişehir city in Turkey is conducted. Artificial Neural Network (ANN), Regression Tree (RT), and Support Vector Regression (SVR) are ML methods that are used to predict HGSR. Besides, hourly metrological and geographical parameters for the year 2014 are taken as inputs in the training models. The inputs are solar time, solar hour angle, Julian day number, daily GSR, longitude, latitude, hourly average humidity, hourly temperature, and hourly pressure. To demonstrate these techniques, a comparison is implemented using MATLAB software with the help of existing toolboxes. Finally, this study proves that ML methods outperform the CPRG model, not to mention they have far more accurate results.Although almost all ML models gave similar results, SVR was the best among them with a correlation coefficient of 0.979532 for the training set and 0.978244 for the testing set. In a nutshell, ML are very efficient methods in that should be taken into consideration to perfectly estimate HGSR.

Maximization of Solar Radiation for Fixed and Tracking Surfaces in Antalya Province of Türkiye

“…Most of the studies to determine the solar radiation potential of a certain location were focused on the determination of total radiation falling on a horizontal surface using measured data [4][5][6][7]. Several solar radiation calculation models are performed in the literature by using Artificial Neural Networks [8,9] and Machine Learning [10,11]. The studies were carried out optimization of tilt angles and maximization of solar radiation falling on the tilted surface for the different locations are contributed to Türkiye in this field [12][13][14][15][16].…”

Estimation of the hourly solar radiation

“…Alsafadi and Başaran Filik used a design algorithm like machine learning mechanism for hourly solar radiation. Estimated the quantity of radiation on an hourly basis by applying machine learning (ML-machine learning) to empirical models for Eskişehir [4] . Serttaş proposed a brand-new pattern scan-based approach to radiation forecasting [5] .…”

“…A mathematical model prediction calculates the amount of radiation for Egypt on an hourly basis, using meteorological data [13] . Within the studies examined, general and conceptual models were formed by fitting and measurements made in numerous parts of the world and at different times into mathematical models [1][2][3][4][5][6][7][8][9][10][11][12][13] . Among these models, mathematical models and experimental coefficients, which are briefly called empirical, were employed within the studies created by Maleki et al [6] , Chandel and Aggarwal [7] , Gueymard [10] .…”

Türkiye Kısa Dönem Elektrik Yük Talep Tahmininde Makine Öğrenmesi Yöntemlerinin Karşılaştırılması

“…XGBoost modelinin MAE (Mean Absolute Error) sonucu 101,3 olarak bulunmuş ve performansı ARIMA modeline göre daha başarılı kabul edilmiştir [7]. Alsafadi ve Filik'in bir diğer çalışmasında ise, saatlik küresel güneş radyasyonu tahmini için makine öğrenmesi yöntemini kullanılmıştır [8]. Yürek ve arkadaşları, makine öğrenimi algoritmalarını kullanarak rüzgâr enerjisi ile elektrik üretimini tahmin etmişlerdir.…”