Regression analysis for performance prediction of geothermal power plant according to system parameters

In this study, the performances of different regression analysis methods for predicting the net power generation using the measured data of a three-flash+binary cycle geothermal power plant located in Western Anatolia are compared. The aim of these regression models is to statistically estimate the effect of the parameters affecting the performance of the geothermal power plant on the net electricity generation of the power plant. A total of 1460 data consisting of geothermal fluid temperature, fluid flow rate, air temperature and net power average values were obtained from the power plant. First, Skewness-Kurtosis normality test was performed on these data and normality condition was met by removing some data. Then, multiple linear, polynomial, decision tree, random forest and gradient boosting regression models were created. Since overfitting was detected in some models, 5 and 10 fold cross validation methods were applied to these models. According to the error measurement parameters R2, MAE, MSE and RMSE values, the most successful method was 10 fold cross validation and the most successful model was the gradient boosting model.