Prediction of Energy Consumption Based on Discrete Fourier Transform and BiLSTM

Effective management of energy consumption is essential for reducing overall energy use, addressing global climate change, and advancing toward a low-carbon, sustainable society. Since it serves as the foundation for energy usage optimization, accurate prediction of energy consumption is essential to the management of infrastructure and energy systems. This study presents a new method for accurate energy consumption prediction. Using spectrum analysis and the discrete Fourier transform (DFT), the energy consumption time series is first broken down into trend, seasonal, and residual components. Using a long short-term memory (LSTM) architecture, an artificial neural network (ANN) predicts each of the three elements independently in the next phase. The projected components are then combined to get the final energy consumption estimate. Ten real-world datasets were used in the validation process, and trials were carried out for both short-term (daily) and long-term (weekly) predictions. The results indicate that the proposed method consistently outperforms the baseline across all datasets, demonstrating its effectiveness in enhancing energy consumption prediction accuracy.