Towards explainable AI-assisted operations in District Heating Systems
Milan Zdravković, Ivan Ćirić, Marko Ignjatović
Abstract
In this paper, we propose the system for AI-assisted control of District Heating Systems (DHS) for efficient distribution of heat to residential and commercial buildings in urban areas. Currently, DHS operation is controlled semi-automatically by SCADA systems, where the operation decisions are made by the DHS plant operator. Operation is considered reactive as control decisions are made based on the real-time meteorological data. This proposal is aiming at facilitating proactive DHS control, by using predictive modelling techniques based on modern Artificial Intelligence (AI) architectures, namely the Deep Learning (DL)-based multivariate time-series forecasting. The initial results of forecasting heat demand and local explanations are provided for the case of small 8MW DHS, as a proof of concept.