District Heating in the Age of AI

摘要

An artificial intelligence (AI) based control method developed by Leanheat uses temperature and humidity data from apartments when adjusting the temperature of water flowing into central heating. The idea is to set a target value for indoor temperature and to control the heating valve in the building control room so that the desired indoor temperature is reached and kept throughout changing weather conditions and building usage patterns (figure 1). During a two week teaching period directly after installation Leanheat's machine learning software creates a unique mathematical model for each building. The model takes into account the building's capacity to charge and discharge heating energy in different conditions. The model is dynamic, constantly measuring all relevant values and reacting automatically to changes of building's thermodynamics. Using the model, the building heating can be optimized so that indoor climate is always good and at the same time energy saved. Also variances in building actual behaviour versus modelled behaviour provides a good base for predictive analytics. Based on data from 40,000 apartments, introducing a dynamic control mechanism instead of a stable control curve diminishes the need of heating energy by 7% on average.