Improvement of Long Short-Term Memory via CEEMDAN and Logistic Maps for the Power Consumption Forecasting

Nowadays, machine learning is an essential factor in computational intelligence that can provide results and solutions in many cases. Forecasting is a crucial case that uses historical data to predict future data trends, and machine learning has become an essential model for predictive methods because machine learning provides high forecast accuracy and reliable result. The fascinating Long Short-Term Memory (LSTM) model is widely used in many forecasting cases and gives exceptional results. However, according to various studies, the issues of LSTM can be addressed in that LSTM can cause the overfitting phenomenon when the dataset contains many noises, and the randomization of LSTM input weight can occur to the outlier sensitivity. In order to improve the forecasting performance of LSTM, this paper proposes a novel LSTM method by optimizing with Logistic Maps (LM) and handling the import dataset with the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), namely, CEEMDAN-LM-LSTM. The experimental results show that the proposed model can provide an r-squared value of up to 0.9999 when applied to the power consumption data from the Tetouan, Morocco dataset.