Power generation forecast for a solar plant with a deep-learning method

Abstract

In this study, we perform power generation forecast of a solar farm using deep learning. A long short-term memory (LSTM) network is applied to forecast time series data of the overall power production from a solar farm. An LSTM network can be considered as a recurrent neural network (RNN) looping with input data (e.g., measured power data) over time steps to update the network information. The network information also has records over all previous time steps. One can use an LSTM network to predict subsequent values of a time series (denoted as open loop forecasting) or sequence using previous time steps as input (denoted as closed loop forecasting). Both forecasting methods are built in the LSTM network. Preliminary results show that closed loop forecasting can allow to have predictions of solar power in more time steps, but less accurate than the other method.