PERFORMANCE RATIO ESTIMATION AND PREDICTION OF SOLAR POWER PLANTS USING MACHINE LEARNING

• Main Author: Bandong, Steven
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/36589
• Institution: Institut Teknologi Bandung

Tropical countries such as Indonesia should have used solar cells as renewable energy source to meet their energy needs. The intensity of sunlight they receive is very large. However the shortcomings tendency of most renewable energy sources such as solar cells are very weather-dependent. In this case, analysis reliability of power production is needed. 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 is has already obsolete. The prediction of solar cell performance can also be an indicator of potential failure in solar cell systems. Machine learnings are used to estimate and predict the performance ratio of solar cells. PCA-SVM is applied to estimate performance ratio by using 35,227 row of weather data from 2015-2018. Grid search is also applied to find optimal SVM parameters in estimating ratio performances. The collected data are used to predict future ratio performances. Prior data decomposing of ratio performance is done to obtain data trend and eliminate noise which causes low prediction accuracy. Three machine learning methods are used, namely SVM, ARIMA and Multiple Linear Regressions to predict performances ratio using one-step and multi-step methods. Grid search is used to find optimal parameters of SVM and ARIMA. MSE, RMSE, R2 and MAPE will be used to evaluate the best way for predicting ratio performances of solar cells plants. The result by PCA-SVM has obtained the accuracy of RMSE = 0.11 and R2 = 0.44 in estimating performances ratio with optimal parameters of Gridsearch acquisition. SVM, ARIMA and Multiple Linear Regression are compared to see machine learning that produce the precise accuracy. In predicting performance ratio using the one-step method , ARIMA has obtained better predictive results with the value of RMSE 0.018, R2 0.93 and MAPE 2.65%. However, multi-step methods has shown that SVM provides better predictive results with the value of RMSE 0.089, R2 0.47 and MAPE 15.48%. Comparing the results between one-step and multi-step predictions has shown that one-step method have better predictive accuracy with RMSE 2.4 times smaller and R2 1.9 times greater than the results of multi-step method.