Very Short-Term Solar Power Forecasting Using Genetic Algorithm Based Deep Neural Network

This paper presents a method for finding optimal parameters of a deep learning model by Genetic Algorithm (GA). The model is employed to forecast output of a solar farm 4 hours in advance. Its inputs are from both forecasted weather data and data obtained from weather monitoring instrument. Performance of four NN (Neural Network) types: DNN (Deep Neural Network), LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and CuDNNGRU (Cuda Deep Neural Network Gated Recurrent Unit), is compared. Feature engineering by Exponential Moving Average (EMA) finds that a time-series of irradiance helps to improve the model performance. GA is exploited to find the most appropriate the number of lookback (or the window size), and the number of neurons in each and all three hidden layers. The GRU model yields the least RMSE at 7.83%. However, if the training time is to be considered, the CuDNNGRU model yields slightly higher RMSE at 7.87%, but its training time is less than half of that of the GRU.