Time series forecasting on solar energy production data using LSTM

Abstract

The fact that countries have increased the use of renewable energy resources in order to meet the increasing energy demands has brought to light the fact that the components and energy production amounts of the solar energy systems to be installed must be estimated accurately. With the benefits of developing technology, the forecasting calculations of these variable nature energy resources have become much more economical by using machine learning methods. In this context, the article proposes a deep learning-based methodology that includes LSTM-based tuned models for PV power estimation, with univariate time series estimation of the amount of power obtained from a solar energy system integrated on a factory roof. When the created models are compared, the results show that the model approaches named LSTM13 provide the most accurate prediction performance with the lowest RMSE metric value of 0.1470 among other proposed models.