Game Design

Broker

 * Responsible for building and operating a portfolio by offering tariffs, and for balancing the supply and demand within that portfolio by trading in the Wholesale Market and in the Balancing Market.

Distribution Utility and Accounting Service

 * The Distribution Utility (DU) models the regulated monopoly that owns and maintains the distribution infrastructure. In the context of Power TAC, it is a neutral third party, primarily responsible for balancing supply and demand in each timeslot. So if a Broker is unable to balance its portfolio, the DU does it, most likely through a balancing market mechanism. Because it controls the physical infrastructure, the DU is the party who is able to observe the actual energy use of the customers, and the actual power imports and exports through its coupling with the transmission infrastructure. Therefore, it also controls the accounting process by which customers and brokers are charged and paid.

Wholesale Market

 * Consists of a day-ahead market and a real-time (spot) market, which for the time being are treated as a single market. There are a number of market models in use, from a simple Continuous Double Auction (CDA) or Periodic Double Auction (PDA) to the more complex FERC structure used in much of North America, in which bids consist of piecewise-linear supply and demand curves. The CDA model is not workable within the Power TAC simulation, because of the time compression of the simulation and the variable latency of network-connected broker agents. Therefore, the market will operate as a PDA or Call market, clearing once per timeslot, with prices determined by the intersection of inferred supply and demand functions.

Tariff Market

 * Brokers offer tariff contracts to customers, which customers may accept or ignore. Tariffs may specify a number of conditions, including minimum contract duration, bonus or fee for subscribing, rates for production and consumption of power, etc. A fee is charged to register a tariff and distribute it to customers.

Market Data Service

 * Provides historical data on production and consumption to Brokers, aggregated in various ways. Also provides actual production and consumption data on an hourly basis, at the end of each timeslot, for a Broker's existing tariff subscriptions. The packaging and delivery of actual data will likely be bundled with the Accounting service.

Tariff Customer

 * A small-scale customer, such as a household or small-to-medium business, may be net supplier or consumer. Most (or all) tariff customers are represented by population models in order to achieve reasonable scalability.

Contract Customer

 * A relatively large (compared to an individual Tariff customer) customer who wishes to negotiate an individual contract with a Broker for electric power service. May be a net supplier or a net consumer. Examples include large commercial and industrial facilities, and government entities.


 * This feature is not currently well-specified, and has not been implemented.

Bank

 * Keeps accounts on behalf of the Brokers and the DU. Probably bundled with the Accounting service.

Simulation Timeline
The game simulates interaction over several days. The number of days is limited by the timeslot granularity (one hour), the maximum length of a simulation (about 2 hours for a tournament situation), and the minimum wall-clock time allocated to a timeslot (target value is 5 seconds). 12 timeslots/minute for two hours gives 1440 timeslots or 60 24-hour days.

The game timeline is therefore a sequence of days, with each day divided into 24 one-hour timeslots. Each timeslot will occupy approximately 5 seconds of real time. The exact number of days simulated must be deliberately undefined to discourage the use of end-game or known-horizon strategies.

Execution and Contracting
Within the simulation, the one-hour timeslot is the smallest unit of time for which prices, power production, and power consumption can vary. During a given timeslot, the makeup of a Broker's portfolio of customers remains unchanged. Brokers may offer tariffs, adjust prices for their subscribed customers, and place bids in the wholesale market at any time. Once during each timeslot, Brokers are notified of market activities, actual production and consumption for each of their tariff subscriptions, and current weather conditions.

Trading in the wholesale market may be conducted in multiple rounds during each timeslot; the number of rounds will be configurable between 1 and 4. A greater number of rounds will presumably communicate more price information, but at the risk of slightly less liquidity in each round.

Customer evaluation of new tariff offerings will be conducted just four times/day, or once every six timeslots.

Initialization
Given the relatively short duration of a simulation, brokers face a major problem of adapting to their environment fast enough to make intelligent decisions. One way to alleviate this burden to some extent would be to run the simulation for a period of time directly preceding the competition scenario, before brokers log in, and collect the basic data on customers and markets that brokers will need.

During this setup period, the default broker would offering a single tariff for each power type, all customers would be subscribe to that tariff, and the default broker would attempt to purchase enough power to serve its entire customer base, thereby making market prices for this quantity of power observable. The simulation would produce the following information during the setup period:
 * Hourly power consumption and production for each customer model.
 * Hourly production and price quotes for each of the gencos.
 * Market clearing records and orderbooks for each traded timeslot each time it is traded.
 * Hourly market position (net mWh of energy purchased/sold in the wholesale market) and prices paid for the default broker in each timeslot.
 * Net imbalance for the default broker in each timeslot. This is the difference between the broker's market position and the actual net consumption of its customers.
 * Hourly weather reports and forecasts.

This is a lot of data. If we run a 2-week setup period, that's 336 timeslots, and each timeslot trades 23 times in the market, giving 7728 orderbooks and market clearing reports. What information do brokers actually need?
 * They need to know about customers and their power usage profiles. This could be satisfied with an array of consumption/production numbers for each customer model, and for each power type, indexed by timeslot.
 * To the extent that power consumption and production is affected by weather, they need to know the weather conditions for the setup period. This could be satisfied with the set of actual weather reports for this period. Forecasts should not be necessary.
 * They need to know the cost of power in the wholesale market. Assuming the default broker's bidding strategy is simple and well-specified, this should be satisfied by the actual cost of delivered power for each timeslot. By combining this with the net consumption data of the customers, brokers could infer an approximate cost curve for wholesale power.

Because external brokers will not be connected, we assume that the time/timeslot could be substantially reduced during this setup period, but even one second/timeslot for a two-week simulated setup period would yield a 5-minute delay at the beginning of each competition game. One way to avoid this problem would be to run the setup period and collect the data offline, then distribute the relevant customer and market data to brokers at the beginning of the game, thereby giving the competition scenario a "running start".

Open question: What is the "base" time of the simulation? Is it the sim time of the start of initialization, or the sim time of the start of the "real" game? Making it be the start of the "visible" simulation would make life somewhat easier for brokers, presumably, but it could complicate things for server elements that carry state over from the initialization period.

Time-skipping to achieve longer timelines
One option is to spread the simulation timeline somewhat evenly across approximately two years. Each simulated day would represent a specific calendar date, and game time would progress by skipping days by 12, 12, 12, and 13 day intervals, using 4 of every 49 days and giving three weekdays and one weekend day in each "week".

For example, assume that the game is initialized with a start date of Jun 1, 2011. Then the sequence of days progresses as:
 * 1) Friday Jun 3, 2011 (simulation day 1)
 * 2) Wednesday Jun 15, 2011 (simulation day 2: add 12 days)
 * 3) Monday Jun 27, 2011 (simulation day 3: add 12 days)
 * 4) Saturday Jul 09, 2011 (simulation day 4: add 12 days)
 * 5) Friday Jul 22, 2011 (simulation day 5: add 13 days)
 * 6) ... and so on.

In other words, the date of the first simulation day is not known until the start of the game, but once that first date is known, the rest of the series can be computed deterministically. This is relevant because the underlying model in the game infrastructure attempts to simulate typical load and production levels for that date of a typical year and game participants may want to use that knowledge in their strategies.

A major problem with this approach is that it would require significant manipulation of weather data to provide correct forecasts and avoid discontinuities across the day boundaries.

Server start
The server is started by the Web App (locally or remotely). It is provided with the identies of the participating brokers as part of its configuration.

Local startup is provided for research and debugging purposes, remote startup for multi-server competition situations.

Initialization

 * Start the Repast simulation
 * Set up Repast model instances

Broker login

 * Brokers first login to the Web App.
 * In case a competition is open for participants, the Web App returns the competition server's URL. The broker then must download the game parameters and static dataset, and then connect to the competition server.
 * In case a competition is already running, the Web App either tells a broker to wait and check back at a later time, or the Web App sets up a new competition server. For research and debugging scenarios, brokers may be allowed to ask the Web App to create a competition based on certain parameters defined by the broker.
 * Once the broker has the server's URL and the necessary static data set, it logs into the simulation server.

Game start

 * Once all brokers defined as part of the server configuration are connected, the game starts.
 * Brokers start with no Tariffs or Tariff subscriptions. All Customers are initially subscribed to a default tariff, which is communicated to the Broker as part of the static data set for the specific simulation scenario.
 * Brokers can immediately begin offering Tariffs, and may observe prices and even trade in the Wholesale Market.

Wholesale market trading
Energy may be bought and sold in an auction-based wholesale market at any time. Energy is traded in units of kWh for a given one-hour timeslot. At any given time, there are 23 timeslots open for trading, ranging from one hour ahead to 24 hours ahead of the current timeslot.

Submitting bids
A bid specifies a quantity of energy (in kWh) and a price/kWh. Bids may be submitted at any time for any timeslot that is currently open for trading.

Transparency
At the end of each timeslot (once per simulated hour), the market clears bids for all open timeslots. Brokers are notified of the clearing price, their updated market commitments, and the price and quantity (but not the bidder identity) for all bids considered during that clearing interval.

Execution
Activities during each timeslot include:


 * Price changes are communicated to Customers.
 * Customers retrieve Physical Environment conditions.
 * Customers determine actual usage and production for the current timeslot.
 * Accounting retrieves supply, demand, and balancing capacity data from Customers. The cost of using customer balancing capacity is determined by tariff terms. This data is treated as broker "bids" in the balancing market.
 * Accounting retrieves Broker positions for the current timeslot from Wholesale Market.
 * Accounting/DU determines price function for external balancing capacity from the most recent timeslot in the wholesale market (the spot market price function).
 * Accounting/DU enters the cost of running its own spinning reserves into the balancing market.
 * Accounting/DU clears the balancing market, thereby performing intra-Broker, inter-Broker, and external balancing. We assume that the market clearing results in using customer balancing capacity as long as it's financially better for the respective brokers than the cost of increasing or reducing imports and of DU balancing. Inter-broker balancing will be charged/credited at the price determined by the balancing market, until we allow brokers to make bilateral deals for mutual balancing.
 * Accounting runs resulting Bank transactions.
 * Accounting computes actual charges and credits for customers based on tariff terms and customer usage and production, runs resulting Bank transactions.
 * Brokers are notified of actual consumption and production, and use of customer balancing capacity, broken down by tariff.
 * Brokers adjust prices for customers with variable pricing. Updated prices may go into effect immediately, or be delayed, depending on tariff specification.