Deep learning techniques for energy forecasting and condition monitoring in the manufacturing sector

Mawson, Victoria Jayne and Hughes, Ben Richard (2020) Deep learning techniques for energy forecasting and condition monitoring in the manufacturing sector. Energy and Buildings, 217. 109966. ISSN 0378-7788

[img] Text (Mawson-Hughes-EB-2020-Deep-learning-techniques-for-energy-forecasting-and-condition-monitoring)
Mawson_Hughes_EB_2020_Deep_learning_techniques_for_energy_forecasting_and_condition_monitoring.pdf
Accepted Author Manuscript
Restricted to Repository staff only until 20 March 2021.
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo

Download (1MB) | Request a copy from the Strathclyde author

    Abstract

    The industrial and building sector demands the largest proportion of global energy, therefore adopting energy efficiency related strategies, optimization techniques and management is an important step towards global energy reduction. The use of machine learning techniques in energy forecasting is gaining popularity due to their ability to solve complex non-linear problems, however this is predominately seen in the residential and commercial sector. This study proposes and compares the use of two deep neural networks, feed forward and recurrent, to forecast manufacturing facility energy consumption and workshop conditions based on production schedules, climatic conditions, thermal properties of the facility building, along with building behaviour and use. The feed forward model was able to predict building energy, workshop air temperatures and humidity to an accuracy of 92.4%, 99.5% and 64.8% respectively when the model was provided with a new dataset, with the recurrent model predicting these variables to accuracies of 96.82%, 99.40% and 57.60%. The neural networks were trained with data obtained from the simulation of a medium sized manufacturing facility in the UK. Coupling simulation techniques with machine learning algorithms allows for a low cost, non-intrusive methodology of predicting and optimising building energy consumption in the manufacturing sector. Furthermore, the use of neural networks provided forecasted building energy profiles for the identification of spikes in energy consumption; an undesirable and considerable cost in the manufacturing sector, as well as the predication of manufacturing environmental conditions for condition monitoring of condition sensitive production environments.