Performance Ratio Estimation and Prediction of Solar Power Plants Using Machine Learning to Improve Energy Reliability

Abstract

One of the reliability indicators of a solar cell system is the ratio of solar cell performance. The performance of the solar cell itself is obliged to be predicted so that energy providers can carry out certain plans related to the maintenance or replacement of solar cell systems if the efficiency of solar cells has been already obsolete. The prediction of solar cell performance can also be an indicator of potential failure in solar cell systems. Machine learning is used to estimate and to predict the performance ratio of solar cells. Principal Component Analysis - Support Vector Machine (PCA-SVM) is applied to estimate the performance ratio by using 35,227 rows of three years weather data from 2015-2018. Grid search is then applied to find optimal SVM parameters in estimating performance ratio. The collected data are also used to predict future performance ratio. Prior data decomposing of performance ratio is done to obtain data trend and eliminate noise which causes low prediction accuracy. SVM and Multiple Linear Regression are used to predict performances ratio using one-step time series method. MSE, RMSE, R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> and MAPE will be used to evaluate the best way for predicting performances ratio of solar cells plants. The PCA-SVM leads to the accuracy of RMSE = 0.11 and R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.44. For prediction, SVM and Multiple Linear Regression are compared to see machine learning that produces the best accuracy. Using the one-step method, SVM results in better prediction results compared to Multiple Linear Regression. Two weeks timestep prediction produces the smallest RMSE and MAPE and the highest R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> both by using SVM or Multiple Linear Regression.