Optimal Electricity Supply Bidding

by Markov Decision Process

Haili Song, Chen-Ching Liu, Fellow, IEEE, Jacques Lawarrée, and Robert W.
Dahlgren, Member, IEEE
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 2, MAY 2000

Presented by:
Sukhmaan Singh Rathore
Sukhmaan Singh
MS – Computer Science
Intern at Walmart, AR
Introduction
 Power industry is competitive environment.
 Primary objectives
 economics and
 profitability

 Decisions have to be made on transactions, e.g., contract types and

parameters.
 Electricity is a commodity
 The spot market usually operates as a pool, i.e.,
 the market participants submit bids to a market
 that determines the transactions based on rules agreed upon by the
participants.
 The problem of bidding decision-making is studied from the viewpoint of a
generation company
Definitions and terms

 Electricity market : An electricity market is a system enabling purchases,

through bids to buy
 Spot Price: Spot, is a contract of buying or selling a commodity, The
settlement price (or rate) is called spot price (or spot rate).
 Production Limit
 Load Forecast
 Load Demand
Problem Formulation
 The decision-making problem is formulated as a Markov Decision
Process - a discrete stochastic optimization method
 Uncertainties - price and load etc.
 To develop a tractable model, a Markov process is assumed.
 The MDP here is of the discrete-state and discrete-time type.
 MDP solves multiple stage probabilistic decision-making problems.
 The stochastic process evolves in a sequence of time stages.
At stage t-
the market can be in any state1 to N.
choose one decision ‘a’ from a set of feasible decision options.

Corresponding to a
decision a,
transition probability
from a state i to
another state j is given
by Pr(i, j, a)).

one strategy (policy)-

The decisions from the first stage to the end

Decision maker receives a reward from each transition

r( I, j , a)
 BIDS – Bidding Decision Support Tool

Overall purpose – Optimize expected reward

Algorithm (CALCULATION OF TRANSITION PROBABILITIES AND REWARDS)
Ojective1 - Transition
probability ?

Object 2 – REWARD ?
CALCULATION OF TRANSITION PROBABILITIES AND
REWARDS
 For simplicity price and quantity from each supplier’s bidding parameters,
i.e.,
 For the suppliers and load periods, a scenario s is:

 The probability of this scenario, Pr(i,s), is:

 Supplier n1 has options K(i,n1)and supplier n2 has K(i,n2)options. possible

scenarios can be enumerated and the total number of scenarios is
For Transition Probability
 What we have:
 Scenario,
 Possible Scenarios
 Total scenario
 What we need :
 Spot price for each scenario
The suppliers’ bids are ranked from the cheapest to the most expensive for each load
period
most expensive bid price - spot price
For a scenario s, SP( i , s , t )
match Load forecast LF(I, t) with the bid
For this we need Decision maker’s bid decision which is given by

This is sum of production limit – bid quantity.

Now , the SP is calculated as follows
 Finally giving the Transition Probability as :

Reward = revenue – cost

*Cost is deterministic
VALUE ITERATION

 Value iteration is similar to backward dynamic programming

 It is total expected reward in τ +1remainiing stages from state i
 Last stage, τ = 0

 Calculates optimal decision from last stage to first

 Weighted by the transition probability
IMPLEMENTATION AND CASE STUDIES

 3-supplier-5-generator System
 GenCoA – 1 unit
 GenCoB – 2 unit
 GenCoC – 2 unit
 Planning Horizon – 7 days
 GenCoA makes its decision
 Unaware of GenCoB and GenCoC bids
 The transition probabilities and rewards are calculated according to the
algorithm
Example
 No of states – 210
 10 grid numbers between 0 and Production limit(PL)
 PL = 960MWh

 Suppliers bid parameters and probabilities are specified

 GenCob bids –
 Unit1: bid price 15$/MWh for the first 20MW block,
16$/MWh for the next 30MW,
maximum output 50MW for both peak & off-peak load period;

 The probability of this bid is : 0.25

 Based on some decision option set let one option be –
 Option 3 – Save Capacity
 Option 4 – sell more electricity
 Option 6 – selling nothing
 If we choose option 6 then according to the algorithm we have
 Pr(2,1,6) = 0.17188
 R(2,1,6) = 0
Conclusion
 The spot market bidding decision-making problem is studied
 An algorithm is developed for
 Transition Probability and Reward

 Simulation and Case Study

 Some power system constrains are ignored
 Decision maker is risk neutral
 If risk-averse - Add profit variance
 Maximize benefits
 MDP optimize the decision making over a planning horizon
References

 Song, Haili, et al. "Optimal electricity supply bidding by Markov decision

process." IEEE Transactions on Power Systems 15.2 (2000): 618-624.
 B. F. Hobbs and K. A. Kelly, “Using game theory to analyze electric
transmission pricing policies in the United States,” European Journal of
Operational Research, vol. 56, no. 2, pp. 154–171, Jan. 1992.
 F. H. Griffs, “Winning over key competitors,” J. Construction Engineering and
Management, vol. 118, no. 1, pp. 151–165, 1992.
Questions ?

Thanks !