LSTM Algorithm For Forecasting Events In Changing Electric Consumption

The industrial sections have always been thirsty to predict events and be aware of future sudden changes. Where using time-series forecasting based on deep learning techniques can have a significant effect on addressing this issue. In this paper, we present Stacked Long Short-Term Memory (LSTM) networks to predict significant electric consumption changes when a group of devices is activated or deactivated randomly. Three Power quadratic phases are used as a dataset. The goal is to predict multi-steps in the future minutes. The supervised learning technique is used to transform series data to supervised data for training and prediction, and then the Walk-forward cross-validation is implemented to evaluate subsequent time steps prediction and output is a vector of prediction. Finally, the root-mean-square error (RMSE) method is employed as a metric for each lead time on an output vector. Furthermore, we compared our proposed model with the Vanilla and Bidirectional LSTMs architecture, in which the performance of our method shows that it was successful. Our experimental results support the view that enough data is crucial to event detection in time series data and suggest that Stacked LSTM is a good architecture with which to address it.