Energy load forecast using S2S deep neural networks with k-Shape clustering

Ensuring sustainability demands more precise energy management to minimize energy wastage. With the deployment of smart grids that provide a huge amount of data, new methods of machine learning come to light to ensure more precise predictions. With the huge amount of data, deep learning methods which require a huge amount of data, are starting to show promise with increasing accuracy. In this paper, we present a methodology for predicting time series which uses Deep Neural Networks, specifically Long Short-Term Memory (LSTM) algorithm with a Sequence to Sequence (S2S) architecture. We improve the prediction accuracy, when predicting the aggregated load of the network, by using the k-Shape clustering algorithm to create groups of consumers that are easier to predict. The methods described in this paper were tested on real world data of electricity consumption of Slovak enterprises.