Electricity Consumption Time Series Forecasting Using Temporal Convolutional Networks

Abstract

Forecasting electricity demand is crucial for the management of smart grids to ensure a secure, reliable and sustainable supply. Recently, a variant of convolutional neural networks, called temporal convolutional networks, has emerged for data sequence, competing directly with deep recurrent neural networks in terms of execution time and memory requirements. In this work, we propose a deep temporal convolutional network to predict time series, namely, the electricity consumption with a 4-h forecast horizon. Results using nine and a half years of Spanish electricity load, with a 10-min sampling rate, are reported and discussed. In addition, the performance of the proposed model is compared with linear regression, decision trees, gradient boosted trees, random forests, deep feed forward neural networks that use different techniques to find the optimal hyper-parameters and a deep Long Short-Term Memory network. The proposed model reaches competitive results in terms of accuracy, with the smallest error verging on 1%.