Decentralized Supervisory Control of Networked Multienergy Buildings

A major challenge in the transition to a net-zero energy system is how to decarbonize energy use for heating and cooling via electrification while ensuring the security of a power system with a high penetration of renewable energy generation. One possible way to simultaneously address these disparate objectives is to use model predictive control (MPC) to manage multivector energy consumption and storage in individual buildings so that operational constraints in connecting networks are not violated. However, control of a large number of building energy assets and their multiple shared networks using standard MPC requires the solution of optimization problems that are large, nonconvex, and, therefore, intractable. In this article, a novel MPC scheme is proposed in which the overall control problem is decomposed and solved in parallel by decentralized control agents. Since the uncoordinated actions of decentralized agents could cause congestion in connecting electricity and district heating and cooling (DHC) networks, an energy flow coordinator is also introduced. This coordinator checks agent actions by solving optimal energy flow problems for each network and uses price signals to direct the search for a globally feasible solution. To improve computational efficiency when determining optimal energy flows, the coordinator utilizes a novel model reformulation of a DHC network. An exemplary case study of a multienergy district demonstrates that the control scheme ensures near-optimal economic performance when compared with an equivalent centralized benchmark—in this case, reducing the maximum computation time from over 55 min to just over 1 s. The approach is suitable for online management of buildings within a district, to both minimize costs to end users and to maintain secure, reliable operation of the connecting networks.

I. Introduction

Asignificant research challenge in the pursuit of net-zero carbon emissions is to “unlock” the potential flexibility of traditionally separate electricity, heating, and cooling supply systems, through co-optimization and control as part of an integrated multienergy system [1]. Central to this challenge is the difficulty in formulating and solving whole-system optimization problems, which enable system operators to compute globally optimal set-points for their controllable assets. In such problems, performance objectives relating to all system operators and end users should be optimized while operational-, security-, and demand-related constraints are also satisfied. Difficulties arise due to several factors: 1) the large number of decision variables in the combined systems; 2) the nonlinear equations used to describe the flow physics in connecting energy networks; and 3) the nonconvex, mixed-integer modeling formulations used to represent hybrid dynamical systems, for example, battery storage systems or multimode energy conversion devices [2], [3]. Each of these factors increases the likelihood that a global optimization problem for online multienergy system management will be intractable, that is, not solvable in the timescales required.