Net-load forecasting of renewable energy systems using multi-input LSTM fuzzy and discrete wavelet transform

This paper represents a new framework to forecast electricity power net-load in renewable energy systems. Estimating electricity power net-load with high accuracy affects economic well-being, stability, and security of power networks. Despite this, large nonlinear variations make its estimation difficult and complicated. In this paper, a new, simple, robust, and straightforward method is proposed to predict signals with volatile characteristics to satisfy this need. This framework combines deep learning of the multi-input LSTM network type with fuzzy system and discrete wavelet transforms. Wavelet-based transforms provide insight into hidden details and aid in forecasting points with high chaos. Furthermore, technical indicators provide a way to determine the trend and momentum of data and to select the optimal time frame for estimation. Finally, the real case of Austrian data is selected, and the electricity power net-load is estimated. According to the results, the proposed framework can forecast the electricity power net-load with 97.7% accuracy. Furthermore, the forecast accuracy is improved to 99.5% by using wavelet transforms and fuzzy system simultaneously in the forecasting process.