Distributed MPC Applied to a Network of Households With Micro-CHP and Heat Storage

Abstract

This paper considers heat and power production from micro Combined Heat and Power systems and heat storage in a network of households. The goal is to balance the local heat demand and supply in combination with balancing the power supply and demand in the network. The on-off decisions of the local generators are done completely distributed based on local information and information exchange with a few neighbors in the network. This is achieved by using an information sharing model with a distributed model predictive control method based on dual-decomposition and sub-gradient iterations. Because of the binary nature of the decisions, a mixed integer quadratic problem is solved at each agent. The approach is tested with simulation, using realistic heat and power demand patterns. We conclude that the distributed control approach is suitable for embedding the distributed generation at a household level.Index Terms-distributed model predictive control, smart energy management, distributed generation This work is carried out as part of the Flexines project, financed by Koers Noord.

Distributed supply–demand balancing and the physics of smart energy systems

“…We briefly summarize the results and approach of [32][33][34]30]. We consider heat and power production from micro-Combined Heat Power systems (μ-CHP's) and heat storage in a network of households.…”

“…The μ-CHP consists of a prime mover whose power and heat output is coupled, and an auxiliary burner which only has a heat output. For the characteristics of the μ-CHP, see [34]. The auxiliary burner produces heat like a conventional boiler, and also stores the heat in the hot water storage.…”

“…We also have introduced two new binary variables to indicate whether the primary mover is still in the start-up phase, or is shut down. See [34] for the details.…”

“…We may encounter that in the gradient iteration switching between binary values at each time step will occur. We approach this problem by fixing the binary value once a non-converging sequence is detected, see [34].…”

“…For simulation results, we refer to [34], where a number of cases of 10 and 25 households in a network are compared. The distributed method can be implemented in a large network, since the number of optimization variables per node is constant.…”