Complex Electricity Systems Modelling

Large-scale socio-technical systems, such as a transitioning electricity infrastructure, are increasingly becoming more complex in nature (Chappin, 2011; MIT, 2011; Bompard et al., 2012).

It would be beneficial to understand them, and hence to solve problems and in turn to inform decision making. According to Chyong (2014), an approach to the use of reactive solutions by observing the sporadic event-type problems that arise, prove to be less influential than proactive solutions based on understanding the system structure. Also, Dyner (1996) argues that an abstraction of reality can be useful, thus models and modelling are of utmost importance to the planning process of electricity systems. In fact, the use of models for understanding complex electricity systems is widely observed in the literature.

44

For example, legacy electricity systems use a method known as “traditional planning” in which a centralised coordinator is responsible for operational decisions, real-time control and monitoring of the electricity system. However, with future low-carbon electricity systems being distributed and decentralised the models for these systems should have more autonomic characteristics.

Nevertheless, centralised planning models such as Integrated Resource Planning (IRP) can still aid in various contexts and situations. However, the use of optimisation approaches for system expansion with electricity markets as imperfect, oligopolistic markets using sequential game theory are examples of situations where centralised models do not work well (Jordan, 2013).

Modelling of complex systems (such as an evolving low-carbon electricity system) can prove beneficial not just for understanding the system, but also for informing the decision-making within the system. Many different types of modelling mechanism have been utilised over the past few decades especially in the area of capacity generation expansion (i.e. increasing electrification) as a means of development and for enhanced utilisation of these systems. However, with these models, the uncertainty increases with the number of possible solution futures, and all the decisions are taken sequentially (Centeno, 2009). Jordan (2013), argues that additional features should be incorporated into these models, making the decision problem even more complex. Non-linear relationships can arise in both the objective function of the amount of capacity needed and the financial and time constraints considered. This will influence the set of possible solutions that can exist for the model. Concurrently, while it is mathematically attractive to have an optimal long-term (20 or more years) plan, it is rarely, if ever, adhered to completely due to unforeseen changes both exogenous and endogenous along the way. Additionally, the traditional methods and models are solely driven by an event-based view of the electricity system planning (see Section 2.2.1). The most suitable modelling mechanism should be able to capture the uncertainties that exist over a long-term planning period, as can be done via systems view thinking.

Nevertheless, there is widespread use of mathematical methods such as linear programming (LP), mixed-integer programming (MIP), dynamic programming (DP) and non-linear programming (NLP)

45

to solve development and utilisation problems of modern and future electricity systems. The need for additional optimisation and heuristic solutions employed to tackle the high complexity and dimensionality of these problems resonates with these mathematical methods. Some of the optimisation and heuristic methodologies are stochastic programming, simulation techniques, genetic algorithms, system dynamics (SD), agent-based modelling (ABM), Monte Carlo simulation, probabilistic simulation, decision theory, game theory, scenario analysis, multi-criteria techniques and real options (Centeno, 2009). All of these above mentioned methodologies prove to be useful for modelling different aspects of electricity systems for which they are most effective. These different aspects include capacity expansion investments and improving the decision making of the system such as grid balancing or energy policy analysis (Dyner, 1996; Lalor, 2005; Dimitrovski, Ford and Tomsovic, 2007; Ilic, Xie and Liu, 2013).

Of the methods mentioned above, stochastic programming, simulation techniques, genetic algorithms and Monte Carlo simulation are all focused on optimisation of the system for a specific variable/s requirement/s and are event-based approaches. Whilst, probabilistic simulation, decision theory, game theory, scenario analysis, multi-criteria techniques and real options are not focused on optimisation of the system variable, they nevertheless do not possess a whole systems view approach. However, agent-based modelling (Macal and North, 2006) and SD can give a good representation of the real world systems using a whole systems view approach and are highly capable of incorporating all of the necessary uncertainties that exist within the system. Agent-based modelling focuses on individual actions of all entities within the system whilst SD is about the understanding of how all entities in a system interact with each other (Harrison, Thiel and Jones, 2016), giving the system structure desired in Section 2.2.1. The agent-based method makes use of a bottom-up approach in which each individual active entity within the system is characterised by rules and allowed to interact with other entities. The global behaviour of the system then emerges as a result of interactions of the individual behaviours and not the complete system structure.

According to Rafferty (2010), agent-based modelling is computationally expensive and can ally itself

46

to either the holistic or reductionist viewpoints of the system. SD only allows for the system view required and is not a computationally expensive technique (Sterman, 2000). Hence, within this thesis, SD has a greater appeal than agent-based modelling, and more support for this is presented in Sections 2.2.4 and 3.1.

In summary, four categories of planning models are observed in this section similar to the observations of (Owlia and Dastkhan, 2012). These are econometric models such as LP that have low precision generally because of considering low details. Then there are energy equilibrium models such as genetic algorithms that also has a low level of details. Thirdly, there are optimisation models which utilise mathematical modelling techniques such as MIP. These models have a high level of details but suffer from the fact they are event-oriented and cannot investigate the dynamics of the system. Finally, there are simulation models such as system dynamics, which can include all-important features such as high level of details, precision, and flexibility and most all-importantly are able to investigate and analyse the dynamics of the system. The next section provides context to the desired shift for better modelling of present and future electricity systems.