Advanced forecasting and disturbance modelling for model predictive control of smart energy systems. (15th June 2021)

• Record Type: Journal Article
• Authors: Thilker, Christian Ankerstjerne; Madsen, Henrik; Jørgensen, John Bagterp
• Is Part Of: Applied energy. Volume 292(2021)
• Publication Date: 2021-06-15
• Subjects: Model predictive control -- Stochastic differential equations -- Disturbance models -- Smart energy systems; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2021.116889
• Languages: English
• ISSNs: 0306-2619

Abstract

Abstract: We describe a method for embedding advanced weather disturbance models in model predictive control (MPC) of energy consumption and climate management in buildings. The performance of certainty-equivalent controllers such as conventional MPC for smart energy systems depends critically on accurate disturbance forecasts. Commonly, meteorological forecasts are used to supply weather predictions. However, these are generally not well suited for short-term forecasts. We show that an advanced physical and statistical description of the disturbances can provide useful short-term disturbance forecasts. We investigate the case of controlling the indoor air temperature of a simulated building using stochastic differential equations (SDEs) and certainty-equivalent MPC using the novel short-term forecasting method. A Lamperti transformation of the data and the models is an important contribution in making this SDE-based approach work. Simulation-based studies suggest that significant improvements are available for the performance of certainty-equivalent MPC based on short-term forecasts generated by the advanced disturbance model: Electricity savings of 5%–10% while at the same time improving the indoor climate by reducing comfort violations by up to over 90%. Graphical abstract: Highlights: An advanced short-term weather forecasting model for buildings based on stochastic differential equations. The formulated models are based on grey-box principles which enable usage of combined information from physics and data. Results suggest improved performance for model predictive control when using the advanced forecasts. The smart building controller using advanced forecasts perform almost as good as perfect forecasts.