Predictive modeling based on Deep Machine Learning models coupled to Discrete Wavelet Transform applied to short-term solar radiation forecasting

Resumo

Forecasting solar radiation is important for analyzing the feasibility of power generation and optimizing
the operation. In this context, approaches using simple Machine Learning models, such as Artificial Neural
Network (ANN) and Support Vector Machine (SVM) were made, but could not satisfy the performance
requirements in complex scenarios. Therefore, it was necessary to use hybrid models with the Discrete Wavelet

Transform (DWT). This paper intended to implement solar radiation forecasting using ANN, SVM, Long-Short-
Term-Memory Network (LSTM) and their hybrids coupled with DWT as a preprocessing technique using Python.

The results obtained were compared to the Autoregressive Integrated Moving Average model (ARIMA). The data
of solar radiation was collected by the automatic weather Salinas station of Nova Friburgo (RJ), the Holdout
validation was used to evaluate the prognosis performances using Root-Mean-Square error (RMSE) and
Coefficient of Determination (R2). The results revealed that the simple models had worse results than ARIMA,
and that hybrid models had superior performance, especially the DWT-LSTM which had an R2 of 0.994 and an
RMSE of 0.516. In addition, Python showed to be powerful as an Open Source tool for implementation of robust
models that are useful for applications in science and engineering.