Learning and Predictive Energy Consumption Model based on LSTM recursive neural networks

摘要

This paper presents a new model for learning and predicting energy consumption based on recurrent neural networks. Specifically, the Long Short Time Memory (LSTM) networks. In this model, we first calculate the moving average of the energy consumption according to a window, well-chosen in accordance with the nature of the data, in order to build an approximate output of the model. Then we use a deep neural network model that combines a multitude of different types of layers to learn how to predict energy consumption in any context. To implement this model, we used the TensorFlowJS Framework in web, mobile or embedded application context. By comparing the prediction results with those obtained by the moving average, we conclude that our model has learned perfectly well how to make good predictions and we can trust it in a different context.