Executive Summary This is the second of a set of four reports to be prepared under an AEC/ARENA supported project to study the theoretical and practical bases for a specific implementation of Frequency Deviation Pricing (FDP), known as Double-Side Causer Pays (DSCP). The Inception Report set out how the project team intends to approach the project, setting out the intended steps in detail. The current report outlines the theoretical basis for FDP and how it could be implemented as DSCP. Specifically, we also examine some critical implementation issues with a view to resolving or at least clarifying and quantifying them in the set of analytical studies to be covered in the analysis report. This report outlines a model, the Linear-Quadratic Regulator (LQR) Model of the electricity system which, when extended to support inputs from available measurements, provides a robust foundation for FDP and the DSCP analysis. Computer code has been developed which, when further enhanced with additional reporting and realistic data, will be used for the studies in the next phase of the project.

Introduction
Project and Report Objective The Australian Energy Council (AEC) with ARENA support has sponsored this project to examine a specific option for pricing and promoting a market for Primary Frequency Response (PFR). The work is motivated by a desire to see a market for PFR maintain good frequency control in normal conditions in the National Electricity Market (NEM) when the current mandatory approach to provision sunsets in 2023. In its Frequency Control Frameworks Review and subsequent discussion papers, AEMC has identified some modification of the existing causer pays system for regulation as a candidate for pricing PFR.Subsequently, CS Energy commissioned a small project of IES to demonstrate how such an approach might work; the approach was called Double sided Causer Pays (DSCP). The aim of this project is to outline the basis for Frequency Deviation Pricing (FDP) specifically applied to PFR and to study a range of implementation issues. DSCP is a specific implementation of FDP applied to PFR but potentially also covering AGC regulation as well faster services such as FFR and inertia and even slower services such as ramping. AEC sponsorship does not imply a commitment to the approach by itself or its members; only a desire to see the option fully
examined.

The Basis for FDP and a DSCP Implementation
Overview The goal of frequency control is to match supply and demand at all times within the dispatch
interval, given the initial schedule set by the energy market. The control problem is to marshal the resources, including inherent system inertia, to achieve that goal. It should balance the actions needed to achieve the frequency and time error performance standards against the cost of moving away from target trajectories and the cost of control. A typical control is ramp rate. PFR is one important resource and DSCP is one means of marshalling that resource with a financial incentive. Financial incentives where practical are likely to encourage better performance and deliver lower costs than technical rules such as the current mandatory approach to PFR.

THE BASIS FOR FDP AND A DSCP IMPLEMENTATION In the existing Causer Pays approach, if participants deviate from their dispatch trajectory in a way that makes the frequency worse (i.e., they are below target when the frequency is low, or vice versa), then a penalty is determined for the participant. The penalty is the product of the volume of the deviation and the size of the frequency error measured during that four second interval. These penalties are then accumulated over four weeks. Participants then must fund AEMO’s FCAS regulation costs in the following four weeks in proportion to the size of the penalty they previously accrued.

The chart below shows responses to a step change in frequency for different time constants. Each response would also be scaled by a weighting factor to give a component of price. In this chart the
measurement length is assumed to be one cycle and the minimum time constant is therefore 0.02
seconds (20 milliseconds) , showing an immediate step change following the frequency change.
The 3600 second time constant would be appropriate for time deviation correction. Such a large time constant has only a very small impact on frequency deviation price and frequency responses which operate in a range of less than 60 seconds. This is adequate for time error correction because time error does not tend to destabilise the system. A time constant of 3600 seconds for time error correction has been used for the current AGC control system for regulation.

For a real system, direct calculation from a model appears impractical. An alternative approach is to recognise that the physical parameters determining these time constants are likely to be the same as those driving AEMO’s definition of frequency control services, both current and future. The immediate requirement is to deal with PFR, whose time constant is likely to be in the 4-6 second range. However, in the analytical studies in the next phase of this project we will also consider the possibility of longer time constants such as the approximately 35 seconds used by AGC regulation.

The time deviation in the plots appears to wander aimlessly but it is controlled, and it is normally distributed (under the assumptions of the model) when sampled over a sufficiently long period
such as a month. However, if the centred noise assumption is not met, the distribution may be
skewed. For example, while frequency will always be zero centred, if there is a bias toward under-forecasting the system will have more negative time deviations than positive and the time
deviation distribution will be skewed, with a longer tail on the negative side. Such nuances do not
detract from the validity of the LQR model. Figure 4 shows a plot of the variables defined in our simple model, expressed as power. Note that the generator approximately follows the slow drift of the load; this is the equivalent of regulation AGC in the NEM. The much faster moving noise is mostly handled by damping (PFR) and system inertia. As represented by the orange line. Note that these do not equate exactly to the components of the swing equation. We will carry out a more detailed analysis of how this balance is achieved in studies planned for later in the project. Note also that the time deviation has little
influence on short term system dynamics due to the large time constant assumed.

In the case of regulation, a participant could choose to become enabled and get the DSCP price, i.e. get paid for both; this should attract more competition and lower costs in enablement. Enablement for regulation may be the simplest way to respond to a lagged DSCP price component. Others might prefer to stay outside the system because they have no SCADA metering or simply to remain independent. However, a cost-effective high-resolution meter that supports FDP could expand participation greatly. If operating independently according to the FDP price signal, one would focus on the different price
components. These components are easily determined from frequency and time error measurements, weighted by a market price available 5 minutes in advance. Such a system together with a control strategy could easily be implemented on a laptop. A fast-acting plant would respond to the PFR component, slower acting plant to the lagged component. Batteries and similar plant could respond to both.

Source:energycouncil