Hybrid Deep Learning (CNN-BiLSTM-LSTM) Model for Prediction of Short Power Consumption

The increase in electricity-based equipment caused by technological advances results in inordinate energy consumption and request for day-to-day power. Furthermore, energy forecasting for individual residential customers acts increasingly as an important factor in operating and designing future grids. A variety of methodologies has been used in order to project energy usage at distinct layers of distribution and transferring systems. In the current study, a new hybrid deep learning approach is presented to estimate power usage in smart homes. First, our model processes the data, excludes missing values, scales the data, and transforms them to assure the model performance. The pre-processed data is subsequently introduced to the proposed joint model as input, which uses CNN, LSTM, Bi-LSTM to choose hybrid discriminative features. Eventually, the fully connected layers are employed. The experiments were carried out to assess and compare the forecasting performance of the proposed method with recent models using the IHEPC dataset, including several variants. The outcomes show that the suggested network has the lowest mean MSE, RMSE and MAPE, compared to other cutting-edge models.