Two-Level Predictive Control of a Heating Substation Using an Edge Flexibility Agent

Introduction. Next-generation district heating relies on demand flexibility, typically induced by dynamic heat prices. Replacing the automation of existing heating substations is costly, whereas a software agent on an edge device can be added above a standard PLC with weather-compensated control, writing only the supply temperature setpoint. This paper determines which measurements and model structure such an agent needs for correct estimates and forecasts, and what its flexibility gives to the consumer and to the network. Materials and Methods. An indirectly connected heating substation was modelled component by component with hydraulically and thermally coupled elements and a single-capacity building model. The baseline is a well-tuned weather-compensated curve. The agent combines a state observer, using return-water and room temperatures, with model predictive control over a 36-h horizon. Costs were decomposed using the building heat balance. Peak-to-minimum price ratios of 1-6 and four weather realizations were simulated, as well as 12-h outages of the agent. Results. With only the return-water temperature, the room temperature estimate is biased by about 2.3 °C and converges no faster than the envelope time constant. Correct forecasts also require as many model disturbances as measurements; the agent's 18-h consumption plan is then more accurate than naive forecasts. At equal comfort the agent uses the same energy as the weather curve, but shifts purchases in time. At a price ratio of 4, the consumer's cost falls by 1.9 ± 0.6 %, and peak-hour purchases by 7.9 ± 2.2 %. A 1-h PLC watchdog removes the comfort impact of an agent outage. Conclusions. An edge agent adds flexibility to energy-efficient weather-compensated control without replacing equipment, provided the room temperature is measured, the model is consistent with the measurements and the PLC remains a fallback. Its value is mainly at the network level.