PREDICTION OF ENERGY PRODUCTION IN PHOTOVOLTAIC POWER PLANT BASED ON BIVARIATE TIME SERIES USING MACHINE LEARNING ALGORITHM

• Main Author: Pinantun Hamonangan, Tesla
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42943
• Institution: Institut Teknologi Bandung

Tropical regions such as Indonesia are highly suitable for the utilization of solar panel as renewable energy because of the abundant energy resources available. The power consumption keeps increasing and urging various industries to develop energy alternative technologies, one of which is from the sun. One technology that has been developed is photovoltaic power plant, that convert solar power into electrical energy. The forecast of energy produced by photovoltaic power plant is necessary so that not only system maintenance plan could be made especially when efficiency has dropped but also to measure the amount of energy produced. The forecast uses historical data. Gridsearch is used to identify the most optimal parameters in predicting energy production. The independent variables that affect solar panels are irradiation and ambient temperature. Using linear regression algorithm can produce prediction with 79% accuracy depends on data patterns, fluctuations, and quality. Therefore, methods to optimize the predicted results are needed, which are MVP and ARIMA. Both techniques have parameters that affect the accuracy of the prediction, so Gridsearch is applied to help in finding the high parameter scores to provide the best parameters in data prediction. MVP parameters compriseC, ?, ? and ARIMAparameters consist of p, d, q, each of which will be combined to find the best value. The best parameter valuesin MVP and ARIMA produce prediction with anaverage accuracy above 85% and low Mape or error value. The highest level of accuracy of the prediction is 92% obtained from ARIMAand 91% gotfrom MVP. Using data from 2016, both methods provide the best results in predicting energy production for 2 months. In addition to accuracy, the Mape and RMSE values also indicate deviation between the actual value and the predicted value, and show how well the method is in producing data patterns.