NEMSIM: Agent-based Simulator for Australia's National Electricity Market

Transcription

1 NEMSIM: Agent-based Simulator for Australia's National Electricity Market George GROZEV; David BATTEN; Miles ANDERSON; Geoff LEWIS; John MO; Jack KATZFEY* CSIRO Manufacturing and Infrastructure Technology, Melbourne, Australia *CSIRO Atmospheric Research, Melbourne, Australia Abstract. NEMSIM is an agent-based simulation model under development that represents Australia's National Electricity Market (NEM) as an evolving system of complex interactions between human behaviour in markets, technical infrastructures and the natural environment. Users of NEMSIM will be able to explore various evolutionary pathways of the NEM under different assumptions about trading and investment opportunities, institutional changes and technological futures including alternative learning patterns as participants grow and change. The simulated outcomes will help the user to identify futures that are eco-efficient e.g. maximising profits in a potentially carbonconstrained or environmentally regulated future. The NEMSIM project is part of CSIRO's Energy Transformed Flagship research program, which aims to provide innovative solutions for Australia's pressing energy needs. Motivation for the project is the Flagship's mission to develop low emission energy systems and technologies. 1. INTRODUCTION Electricity industries in many countries have gone through significant changes and restructuring associated with introduction of markets, competition and new technologies. Also, there is a growing number of implementations of market-based policy instruments for environmental purposes, mainly to reduce greenhouse gas (GHG) emissions. A competitive electricity market can be a vehicle to serve the public interest, however, there are a significant number of challenges in designing a market structure able to address complexities of the electricity grid, essential infrastructure and economic performance of the market participants. One instance of these new energy markets is Australia s National Electricity Market (NEM), which commenced operations in December It is a gross pool-type market that comprises four Eastern states: Queensland, New South Wales, Victoria, South Australia and the Australian Capital Territory. Tasmania will be linked to the NEM with the undersea Basslink early in There is a growing need of better understanding and comprehensive modelling of market operations and evolution for the purposes of market participants, governments and regulators. To answer some of these questions, CSIRO has initiated a research project to develop a multi-agent-based simulation model that represents Australia s National Electricity Market (NEM) as a co-evolving system of complex interactions between participants in the NEM and other associated markets, and their effects on technical infrastructures and the natural environment. The simulation software is called NEMSIM (see Figure 1) and it aims to help in decision making in the face of complexity and uncertainty. Figure 1: NEMSIM main screen In NEMSIM, agents (representing generator companies, network service providers, retail companies and a market operator) buy and sell electricity in a simulated trading environment. The model is designed to examine scenarios having principal inputs of generator companies bidding practices, bilateral financial contracts, transmission network limitations and new investment in generating plants and transmission lines. Regional demand for electricity is based on historical demand patterns and can be varied to suit growth forecasts and exceptional weather conditions. A market operator agent balances the dispatch of supply to ensure that demand is always met within every 30 minutes (the market-clearing trading interval).

2 In the short term, NEMSIM can solve problems to help power companies improve their bidding in the very dynamic daily energy market. Retailers can use NEMSIM to inform their decisions on medium-term contracts with power generators. They can reduce their exposure to short-term price volatility or wholesale price rises by signing contracts for fixed-price bulk power allocations. NEMSIM is also a useful modelling tool for power-generation companies to explore investment in extra generation capacity or network upgrades to accommodate growing demand, or changing patterns of demand. 2. MAIN FEATURES The NEMSIM overview structure is shown in Figure 2. The jigsaw part of the diagram represents the various components of interest in a simulation scenario. The principal components are described in following subsections. Agents make decisions under different time horizons. The simulation engine interprets and executes the scenario and generates output results. A graphical user interface is used to monitor, either continuously or periodically, outcomes of agents behaviours. 2.1 Historical market data The development of NEMSIM is facilitated by six years of historical market data on demand, pricing and power dispatch for the NEM. Being a dynamic, repetitive and informative system, the NEM offers a huge amount of data. NEMSIM uses this data to extract representative patterns of regional demand (on a daily, weekly and seasonal basis), regional prices, supply and demand growth, and so on. The historical database includes an extensive time series of bidding data, an essential source of information about market participants and their trading strategies. NEMSIM can display some historical data (regional prices, regional demand, bidding data) and can configure generator agents to use real bidding data. These are important features for validating the software. 2.2 Technical infrastructure Simulated agent life in NEMSIM unfolds in three environments : firstly, a trading environment in which transactions can occur in interlinked spot and forward contract markets; secondly, a physical grid of sites, generation units, lines and inter-connectors across which electricity flows, and; thirdly, a natural environment, which provides energy resources and accumulates GHG emissions. Each environment is separate from the agents, on which the agents operate and with which they interact. The physical environment of transmission infrastructure imposes several constraints on electricity market operations. For example, the quantity of power sold from one region to another is constrained by the transmission capacity between the two regions. NEMSIM represents the diversity of objects and attributes associated with generation and transmission infrastructure in a simplified form. Time Horizons Historical Data Bidding Dispatch 30 min Dispatch Technical Infrastructure Agents & Markets Environment Generating Plants Generating Units Interconnectors Transmission Lines Companies Spot Market Market Operator Contract Market Electricity Fuel GHG Power Losses Supplied Resources Emissions Simulated Scenario Daily Weekly Monthly Yearly Longer Term Input User Data Input Simulation Engine Evolution Daily Bidding Supply Evolution Spot Contract Investment GHG Emissions Reports Graphs Tables Simulation Log Scenario Evaluation Figure 2: NEMSIM Overview Structure

3 Some key objects (with corresponding attributes in parentheses) are: generating plants (location, unit composition), generating units (maximum capacity, generation technology, ramp rates, efficiency and emission factors) and inter-connectors (transmission technology, adjacent regions, losses). 2.3 Agents Agents in NEMSIM are main classes designed to accommodate the complexity and adaptation of market operations. Perhaps one of the most advanced agentbased systems for electricity markets is EMCAS [2], developed by Argonne National Laboratory in the US. However, the adaptation of EMCAS for Australia s NEM is problematic because of different market structures and operations. Hence, CSIRO decided to develop an agent-based tool specific for Australia s NEM. The development has been facilitated by using a simulation framework from Swinburne University of Technology, Melbourne, Australia. Company agents can have different goals. As well as maximizing profits, some may wish to increase market share, diversify generation sources or work more closely with end-users. Short-term strategies (e.g. bidding tactics in the NEM) are affected by mediumterm strategies (e.g. hedged positions in the contract markets). In turn, both are affected by investment decisions and other changes over the longer term. NEMSIM treats agents as being uniquely intelligent, making operational and strategic decisions using the individual information available to them. Also, they are adaptive, learning to modify their behaviour in order better to realise their goals. Learning algorithms will allow agents to look back (learn from their historical performance), look sideways (learn from other participants strategies) and look ahead (take future plans and forecasts into account). In an adaptive market like the NEM, no single agent has control over what all the other agents are doing. Because of the way the market is structured, however, the marginal bidder can exert more influence on market outcomes. The overall outcome is not always obvious, because it depends on many factors. The aim of NEMSIM is to provide a platform where a population of simulated agents interact, constrained only by realistic rules and the physical grid system. Agents individual and collective behaviours then co-evolve from the bottom up, producing both expected and unexpected emergent outcomes at the system level. A significant focus of NEMSIM agents is on bidding strategies. Comprehensive observations of bidding patterns of generator companies in the NEM are provided in [2]. 2.4 Time horizons for decision making The key purpose of the NEM is to match wholesale supplies of electricity to demand, itself a complex, short-term task. Different operations such as forecasting, bidding, electricity generation and system stabilising are also important. In the medium-term, companies want acceptable returns on capital and an acceptable maintenance plan. In the long-term, demand for low cost, reliable electricity will continue to increase but, more than likely, in the context of carbon constraints. The timing of such changes is unknown. Thus, new investments required in the long-term face substantial uncertainty with respect to environmental regulation. The short, medium and long-term timescales cannot be treated independently, since what happens on one time-scale will affect what happens on another. To permit decision-making and planning for different time horizons, NEMSIM allows various events to be scheduled at different times (e.g. hourly/half-hourly, daily, weekly, monthly, annually and over several years). In particular, it will include feedback loops between the spot (NEM) and contract markets, so that the hedging decisions of generators and retailers can be represented. These behaviours influence incremental investment and the closure decisions of agents. Finally, it can be adapted to explore further external changes, such as the introduction of an emissions trading scheme (see Section 4). 2.5 Simulated outcomes The purpose of NEMSIM is not to predict the future, but rather to identify and understand the various alternative futures that could unfold under different conditions i.e. to analyse what-if scenarios. The simulator can also be used to show the possible evolutionary trajectories of a given scenario under given conditions. For example, the introduction of more distributed generation (see Section 3) into the marketplace involves a transition from the current paradigm of the centrally dispatched electricity grid to new more decentralised ones. This may involve new markets, new brokers, new technology and new grid structures. NEMSIM is a generative tool that can identify the transition states needed to reach specific final states such as these. 3. DISTRIBUTED GENERATION Distributed generation (DG) is a set of small-scale power generation technologies (up to 10,000 kw) located close to where electricity is used (e.g., a home or business) to provide an alternative to or an enhancement of the traditional electric power system with central generation. Small generators such as backup generators and on-site power systems have been used for a long time. On the other hand, new DG technologies (e.g., microturbines, fuel cells, photovoltaic and wind systems) have been developed recently. Distributed generators are connected to the distribution part (11-33 kv) of the electricity networks, at or near the end user. Potential benefits from DG are lower cost, higher service reliability, high power quality and increased

4 energy efficiency. DG is a promising solution for the security of electricity supply, providing distributed and diverse energy source infrastructure. The use of renewable distributed generation can also provide a significant environmental benefit in terms of reducing GHG emissions. DG can be beneficial to both electricity consumers and energy utilities (for example reducing the cost of transmission and distribution system upgrades). Utility deregulation is also one of the reasons for the high level of interest in DG. Because DG helps to manage peak load demands, it could reduce price volatility and companies market power. There are significant technical difficulties in implementing the connection of DG, for example most distribution network infrastructure allows power to flow only in one direction. NEMSIM will aid in understanding the advantages and disadvantages of DG and the impacts of growth in the uptake of DG technologies. For example, Figure 3 shows the demand and price graphs for NSW on 1 st of December 2004, which was a very hot day. Around midday, demand peaked, causing the supply system to be stressed, resulting in the market regional pool price jumping to the maximum value of 10,000 $/MWh. If significant DG was available, it could have been engaged mid-morning (perhaps when the price exceeded a threshold value of 320 $/MWh), thereby reducing demand and hence preventing price spikes. The dotted lines in Figure 3 represent the simulated influence of DG on demand and price curves. Regional demand [MW] NSW 1/12/2004 Regional Regional Price No. of five-minutes intervals Figure 3: DG perturbed demand GREENHOUSE GAS EMISSIONS Electricity generation is a substantial source of GHG emissions. According to the Australian Greenhouse Office [3] it contributed 182 Mt of carbon dioxide equivalent emissions (Mt CO 2 -e) in 2002 or 33% of net national emissions (550 Mt CO 2 -e) in Australia. The net emissions are calculated across all sectors under the accounting provisions of Kyoto protocol for Australia. NEMSIM is capable of calculating the GHG emissions associated with electricity generation of a given simulation scenario. The method is bottom-up type aggregation. The advantages of NEMSIM are that its Wholesale regional price [$/MWh] simulation framework allows quite precise modelling of the emissions up to the level of each generating unit, accommodating a variety of changes in the operational, technological, company, market and regulatory values and parameters. We believe that the main advantage in this sense is the agent-based framework that allows slow changes to accumulate over long periods and sudden changes to be introduced due to the emerging events based on the decision making and interactions of the participating agents. Calculating the GHG emissions due to electricity generation is implemented in NEMSIM on a fossil fuel consumption basis using fuel (generation technology) specific emission factors. A set of generation technologies are modelled: conventional black coal - pulverised fuel, conventional brown coal - pulverised fuel, natural gas simple cycle, etc. New generation technologies and fuel types can easily be added. The main attributes of each generation technology are the emission GHG factor (f) and the net energy efficiency (e) how much of the embodied energy of the fossil fuel is transformed into electricity energy. Each generating unit is assigned one of the defined generation technologies. The net energy efficiency can also be defined for a selected generating unit (in which case it overrides the same attributes of the assigned generation technology) to allow flexibility in the long term, when due to different reasons the efficiency may change. GHG emissions for a given simulation period or simulation scenario are estimated by using the following formula: 3.6 g f q =, where 3 10 e q is the amount of CO 2 equivalent emissions expressed in t; g is the amount of the electricity generation expressed in MWh; f is the emission factor for a given generation technology expressed in kt CO 2 -e/pj; e is the net energy efficiency of the generation technology or generating unit (dimensionless); and is a conversion factor from kt/pj to t/mwh. The direct values of emission factors are used, which means that emissions associated with extraction and production of the fossil fuels are not considered. This approach allows easier comparison between different plants, companies and regions, however, its accuracy is inferior as it does not include indirect emissions that are usually several percent from direct and they show moderate variability by region, company and technology. An example of a NEMSIM output window for GHG emissions is shown in Figure 4. It displays a day s simulated GHG emissions, in t CO 2 -e, for four coal based units within one generation plant. The values are for a hypothetical scenario and are for illustrative purposes only.

5 Energy: Stationary sources and fugitive emissions, Commonwealth of Australia, Canberra, ubs/01.pdf. 4. CSIRO s Energy Transformed National Research Flagship, ytransformed Figure 4: Example of GHG emissions display SUMMARY AND FUTURE WORK In this paper we have presented NEMSIM, an agentbased simulation system under development for Australia s NEM. The multi-agent structure of the system promises to be an effective way of representing the complex interactions between market players and to facilitate modelling of their decision making. It should prove useful to a variety of companies and other parties as they explore future investment opportunities, risk management strategies and changing rules and market structures, for example, under an increasingly carbonconstrained future, which may demand increased DG and reduced GHG emissions. Possible future extensions include enhanced agents learning capabilities, for example, by using genetic algorithms, integration with power engineering tools to account for grid constraints and including the impact of additional markets such as GHG emissions trading schemes. ACKNOWLEDGMENTS NEMSIM has been developed under a research project that is a component of CSIRO s Energy Transformed Flagship research program [4]. The main mission of the Flagship is to develop new, efficient and low-emission energy technologies for Australia. The authors of the paper are grateful to Mr. Paul Graham, Energy Futures theme leader within the Flagship, for his overall support and contributions towards the project development and specifically for the GHG emission modelling. NEMSIM is built using a simulation framework developed by Swinburne University of Technology, Melbourne, Australia. REFERENCES 1. Electricity Markets Complex Adaptive Systems (EMCAS), (last visited 10 April 2005), 2. Hu, X., Grozev, G., Batten D. (2005), Empirical observation of bidding patterns in Australia s National Electricity Market, Energy Policy, vol. 33, no.16, pp Australian Greenhouse Office (2004), National Greenhouse Gas Inventory Factsheet 1: