Deep Learning Attention Mechanisms Combined with Roll Padding: Application to Household Electric Power Consumption

• Type: Article in International Conference Proceedings Book
• Date: 2022

Authors

Goncalves, R (Author) Other; Ribeiro, VM (Author) FEP; Pereira, FL (Author) Other

Conference proceedings International

Title: TECHNOLOGIES, MARKETS AND POLICIES: BRINGING TOGETHER ECONOMICS AND ENGINEERING

Pages: 249-254

5th International Conference on Energy and Environment - bringing together Economics and Engineering (ICEE)

Univ Porto, Sch Econ, Porto, PORTUGAL, JUN 02-03, 2022

Indexing

ISI Web of Knowledge - 0 Citations

Other information

Authenticus ID: P-00X-64X

Abstract (EN): The accurate prediction of electric energy consumption in the residential sector is a desirable action to ensure the minimization of potential losses and the maximization of social welfare. This study proposes a new Deep Learning Neural Network architecture conceived for multivariate time series problems, which consists of including a special mechanism of attention taking the form of a multi-head bi-dimensional convolution and a novel padding method called roll padding into a ConvLSTM2D model. After being trained, tested and compared to several benchmark alternatives considering the Household Electric Power Consumption data set provided by the University of California at Irvine machine learning repository, results show that the proposed model exhibits the lowest forecasting error in the predictive exercise.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 6