A smart adaptive LSTM technique for electrical load forecasting at source

Electrical load forecasting has never been more critical, given its high influence on the operational efficiency of the power supply system. Accurate prediction helps in making decisions and has a huge impact on the revenue of the suppliers. However, the prediction is not trivial and has several hurdles such as the nonlinear property of the time series or the seasonality pattern that generally inherently exists in a time series problem. Several techniques have been proposed to solve this problem such as statistical models, fuzzy systems, artificial neural networks (ANNs) and machine learning. This paper aims to present a comparison between popular techniques used in this problem domain with a few novel techniques that predict the electrical load at source. Several characteristics of the time series have been analyzed in depth. All the models proposed here have been compared rigorously through several objective functions. The results show that the models proposed here outperforms several statistical and machine learning techniques that are used conventionally.