Analyzing the impact of temperature on PV module surface during electricity generation using machine learning models

Use of fossil fuel in industries causes Carbon emission, which is mostly responsible for global warming. Another aspect is that environment friendly energy production and sustainable development goal is highly dependent on the production of clean energy. According to the IEA solar energy has a huge potential and will contribute up to 16 % of the global electricity by 2050. Hence, prediction of solar energy production has a great deal of demand in renewable energy sector. This paper compares machine-learning algorithms to evaluate the impact of PV module back surface temperature (degC) on the generated power. Support Vector Machine for Regression (SMOreg), Multilayer Perceptron (ANN), Linear Regression, M5 Rules, k-Nearest-Neighbor (Ibk) and Random Forest methods are employed to test their performance in different ratio of training and testing data. The dataset comprises five independent parameters such as PV module back surface temperature (degC), Dry bulb temperature (degC), Relative humidity (%RH), Atmospheric pressure (mb), and Precipitation (mm). The dependent parameter is Maximum power of PV module (W). The correlation coefficient was determined by varying the percentage of training data from 60 % to 85 %. The numerical tests were done for two data sets, one dataset includes all the independent variables and another one excluded the PV module back surface temperature. Except for M5 Rules, other models exhibit consistent correlation coefficients with several of training data. All models demonstrate a dependency on the PV module back surface temperature, with Random Forest surpassing others in overall performance with a correlation coefficient of 0.9713 at 75 % of training set.