Photovoltaic Power Forecasting Using Recurrent Neural Network Based On Bayesian Regularization Algorithm

Abstract

Photovoltaic (PV) is a generator that utilizes solar energy into electrical energy. On-grid photovoltaic is the system can reduce electricity bills, besides that the electricity produced is environmentally friendly and free emission. PV power output intermittent depending on weather conditions. Therefore, this research will predict the power output PV one day ahead using Recurrent Neural Network (RNN) method with Bayesian Regularization Algorithm because it can solve problems regarding prediction, classification, and energy management. The measure of accuracy error from the simulation result in this study is calculated using Mean Absolute Percentage Error (MAPE). The PV power forecasting accuracy using RNN method is compared with actual data. The amount of load power that PV cannot fulfill will later be back up by the grid. The prediction of PV power using RNN method with 4 neuron hidden layers and learning rate 0.01 resulted in the best MAPE value of 2,2784 %. Based on the results, PV power forecasting output using the RNN method with historical data can be applied to determine the amount of PV power for day ahead.