The EMS we have developed in this chapter is crucial for enabling house- holds to efficiently operate DETs, and through them modify their pattern of consumption. Creating this flexibility in household consumption is what is needed to help networks with balancing renewable energy and keeping the network within its operating limits.
To the best of our knowledge, this is the first work to produce a scalable and accurate solution in the presence of uncertainty about future prices, occupant behaviour and environmental conditions. Using models represent- ative of physical devices and random processes, we have shown the monetary and comfort cost savings that can be achieved by using online stochastic al- gorithms over reactive control, and the comparison of performance between a 2-stage approach and acting on expectations. Studies such as the one provided in this chapter are important for rallying industry and customers towards more effective energy management schemes.
Further research is needed to investigate how closely reality can be mod- elled with random processes, and if in turn they are suitable for online learning. We also need to further investigate how time step sizes influence performance, and to conduct more experiments for different months of the year to get a broader understanding of the value of such technologies. The experimental setup we have developed can be used to experiment with and compare different pricing schemes, for example: TOU pricing, RTP and prices that change depending on whether the house is buying or selling elec- tricity.
Chapter 4
Network-Aware
Coordination
4.1
Introduction
The EMS presented in the previous chapter successfully reduces costs for a home exposed to dynamic electricity prices by automatically controlling local DETs. From the point of view of a utility, dynamic pricing can be used to coordinate the combined actions of many thousands of such homes; however, if not done carefully, this can cause undesirable herding or other unexpected impacts. In this chapter we present a method of calculating these prices, and hence for coordinating multiple EMSs, so that the combined result is the most efficient for the overall system.
Conventional networks use centralised markets to solve the similar prob- lem of dispatching generators (see section 2.1.1). These markets, to one degree or another, seek to achieve an optimal power flow (OPF) [Glover et al., 2011, chapter 12], which is a traditional power systems optimisation problem that is concerned with minimising operational costs. The goal is to dispatch generators in a way that minimises costs, so that all loads are met and without overloading the network. Such markets and the traditional OPF problem were not designed to operate in the new world of prosumers where every customer is potentially an active participant. This massive increase in scale, the time-coupled behaviour of DETs and the unique preferences of prosumers means that a different approach is needed.
We envisage a future where network operators provide a competitive electricity market that anyone can participate in, and where the distinction between generators, loads and prosumers is removed. The overall operation of this market is managed through distributed algorithms and with each EMS calculating their own small local part of the overall problem. This will be of particular importance for the operation of microgrids, which require more finesse to ensure that demand and supply are balanced and that the
network is in a safe operating state in each instance.
Several works have applied distributed solving techniques to the prob- lem of coordinating many participants [Kraning et al., 2014, Gatsis and Giannakis, 2012, Mohsenian-Rad et al., 2010]. These distributed algorithms greatly parallelise the problem and help to preserve the privacy of parti- cipants. As a by-product, they provide a natural market mechanism for allocating payments between participants. Theoretically, these algorithms require the problem to be convex in order to guarantee convergence to a globally optimal solution. However, the behaviour of many household loads are discrete in nature [Ramchurn et al., 2011], and the equations that govern how power physically flows on the network are non-convex.
We show that these theoretical problems can in practice be dealt with, specifically in the context of microgrids where the problem of balancing sup- ply and demand is more challenging because individual participants have more influence. We show that for a distributed algorithm in a microgrid, exact non-convex power flow models perform well compared to inexact con- vex models, which makes them a valuable candidate in practice. Secondly, we find that the non-convex nature of discrete house loads to be a non-issue, and that in practice simple approaches to handling these discrete loads are effective at the microgrid level. By solving these problems, we show that the use of distributed algorithms for managing the balance of power in a microgrid is in practice not only possible, but also highly effective.
We formulate our EMS coordination problem as a multi-period OPF problem to account for multiple time steps over a day, which can be used as part of a day-ahead pricing scheme or, as we propose, a receding horizon control algorithm. We solve the multi-period OPF problems in a distrib- uted manner by adapting the alternating direction method of multipliers (ADMM) approach presented in [Kraning et al., 2014]. We experiment with a range of power flow models of varying degrees of accuracy, to compare their relative behaviour in a distributed algorithm. We then introduce and compare several approaches layered on top of ADMM which manage the introduction of discrete variables into the problem. Technically, our contri- butions can be summarised as:
• A comprehensive experimental comparison of the convergence of five commonly used power flow models when used for distributed OPF in a microgrid context.
• The identification that the exact non-convex power flow model in prac- tice not only converges in this context, but also finds near-optimal solutions in a timely fashion relative to other models.
• The introduction and comparison of three simple but effective ap- proaches to managing the discrete shiftable loads that are typically found within houses.
4.2. POWER FLOWS 53