Matrix Games and Power Markets
Research Abstract
With the advent of e-commerce, the contemporary marketplace has evolved significantly toward
competition-based trading of goods and services. Many such market scenarios can be modeled as matrix games. This paper presents a computational algorithm to obtain Nash
equilibria of n-player matrix games. The algorithm uses a stochastic approximation based
reinforcement learning (RL) approach and has the potential to solve n-player matrix games
with large action spaces. The proposed RL algorithm uses a value iteration approach, which
is well established in the MDP/SMDP literature. Results from the recent studies establishing
the existence of equivalent matrix games for stochastic games with both discounted and
average rewards are outlined. These results emphasize the fact that the solution of matrix
games is also critical to solving a large class of stochastic games. The RL algorithm is tested
on a set of sixteen matrix games with up to four players and sixty four action choices. Also
studied in detail is an example of a matrix game representing strategic bidding in an electric
power network with two players and 625 action choices each. The algorithm presented here
can benefit from the availability of tera/peta scale cluster computing in overcoming the computational
barriers arising from large state-action spaces, and thus impact a broad range of
decision making problems.
Publications
Conference Presentations
-
Nanduri, V. and Das, T. K. A Three-Tier Game Theoretic Model for
Multi-Period Multi-Generator Capacity Expansion in Restructured
Markets. INFORMS Annual Meeting 2007, Seattle, WA.
-
Nanduri, V. and Das, T. K. A Reinforcement Learning Model to Assess Market Power under Auction-Based Energy Pricing. INFORMS Annual Meeting 2005, San Francisco, CA.
|
