Time series forecasting on multivariate solar radiation data using deep learning (LSTM)

Energy management is an emerging problem nowadays and utilization of renewable energy sources is an efficient solution. Solar radiation is an important source for electricity generation. For effective utilization, it is important to know precisely the amount from different sources and at different horizons: minutes, hours, and days. Depending on the horizon, two main classes of methods can be used to forecast the solar radiation: statistical time series forecasting methods for short to midterm horizons and numerical weather prediction methods for medium- to long-term horizons. Although statistical time series forecasting methods are utilized in the literature, there are a limited number of studies that utilize deep artificial neural networks. In this study, we focus on statistical time series forecasting methods for short-term horizons (1 h). The aim of this study is to discover the effect of using multivariate data on solar radiation forecasting using a deep learning approach. In this context, we propose a multivariate forecast model that uses a combination of different meteorological variables, such as temperature, humidity, and nebulosity. In the proposed model, recurrent neural network (RNN) variation, namely a long short-term memory (LSTM) unit is used. With an experimental approach, the effect of each meteorological variable is investigated. By hyperparameter tuning, optimal parameters are found in order to construct the best models that fit the global solar radiation data. We compared the results with those of previous studies and we found that the multivariate approach performed better than the previous univariate models did. In further experiments, the effect of combining the most effective parameters was investigated and, as a result, we observed that temperature and nebulosity are the most effective parameters for predicting future solar radiance.