Deep Learning for Short-term Instant Energy Consumption Forecasting in the Manufacturing Sector

TL;DR

This paper evaluates deep learning models, including LSTM, CNN, CNN-LSTM, and TCN, for short-term energy consumption forecasting in manufacturing, finding TCN to be the most effective.

Contribution

It compares multiple deep learning architectures for short-term energy forecasting in manufacturing, highlighting TCN's superior performance.

Findings

TCN outperforms other models in accuracy.

Deep learning models improve short-term energy prediction.

Results support data-driven approaches for energy management.

Abstract

Electricity is a volatile power source that requires great planning and resource management for both short and long term. More specifically, in the short-term, accurate instant energy consumption forecasting contributes greatly to improve the efficiency of buildings, opening new avenues for the adoption of renewable energy. In that regard, data-driven approaches, namely the ones based on machine learning, are begin to be preferred over more traditional ones since they provide not only more simplified ways of deployment but also state of the art results. In that sense, this work applies and compares the performance of several deep learning algorithms, LSTM, CNN, mixed CNN-LSTM and TCN, in a real testbed within the manufacturing sector. The experimental results suggest that the TCN is the most reliable method for predicting instant energy consumption in the short-term.