Nash equilibrium problems in power markets and product design: Analysis and algorithms

competition;Game;Cournot;Bertrand;power markets;product deisgn;algorithms;analysis

Kannan, Aswin ; Shanbhag ; Vinayak V. ; Kim ; Harrison H.M.

关键词: competition;    Game;    Cournot;    Bertrand;    power markets;    product deisgn;    algorithms;    analysis;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/15973/Kannan_Aswin.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

The focus of this research is on the analysis and computation of equilibria in noncooperative Cournot and Bertrandgames. The application of focus for Cournot competition is power markets while that for Bertrand competitionis product design. We consider Cournot-based models for strategic behavior in power markets while Bertrandbasedmodels are employed for analyzing the behavior of price-based competition in product design. This thesis ispartitioned into three parts. Of these, therst two parts focus on power market applications while the third partfocuses on product design.Motivated by the risk of capacity shortfall faced by market participants with uncertain generation assets, the rst part considers a game where agents are assumed to be risk-averse optimizers, using a conditional value-at-risk(CVaR) measure. The resulting game-theoretic problem is a two-period risk-based stochastic Nash game with sharedstrategy sets. In general, this stochastic game has nonsmooth objectives and standard existence and uniquenessresults cannot be leveraged for this class of games, given the lack of compactness of strategy sets and the absenceof strong monotonicity in the gradient map of the objectives. However, when the risk-measure is independent ofcompetitive interactions, a subset of equilibria to the risk-averse game are shown to be characterized by a solvablemonotone single-valued variational inequality. If the risk-measures are generalized to allow for strategic interactions,then the characterization is through a multi-valued variational inequality. Both this object and its single-valuedcounterpart, arising from the smoothed game, are shown to admit solutions. The equilibrium conditions of the gamegrow linearly in size with the the sample-space, network size and the number of participatingrms. Consequently,direct schemes are inadvisable for most practical problems and instead, we present a distributed regularized primaldualand dual projection scheme where both primal and dual iterates are computed separately. Rate of convergenceestimates are provided and error bounds are developed for inexact extensions of the dual scheme. Unlike projectionschemes for deterministic problems, here the projection step requires the solution of a possibly massive stochasticprogram. By utilizing cutting plane methods, we ensure that the complexity of the projection scheme scales slowlywith the size of the sample-space. Insights regarding market design and operation are obtained after testing themodel on a 53-node electricity network.The second part extends this model by considering the grid operator to be a pro t maximizer. However theeffect of risk is neglected in this model. The resulting problem is a quasi variational inequality. An analysis of theequivalent complementarity problem (CP) allows us to claim that the game does admit an equilibrium. By observingthat the CP is monotone, we are in a position to employ a class of iterative regularization techniques namely theiterative Tikhonov and the iterative proximal algorithms. The algorithms are seen to scale well with the size of theproblem. The model is employed for examining strategic behavior on a twelve node network and several economicinsights are drawn.The third part of this thesis deals with Bertrand competition in a product design regime. With due considerationto the attribute dimension in addition to price competition, more specifically for design and consumer service industries,a game theoretic model is formulated. The logit model, in lieu of some of its tractable properties, is deployed tocapture consumer preferences and thereby the demand. Subsequently the variational formulations corresponding tothe game are analyzed for existence of solutions. The lack of convexity of objectives, analytical intractability of thevariational formulations corresponding to the game state some drawbacks of the logit model. Several projection andinterior point schemes are deployed for solving these classes of problems. Numerical results for smaller instances ofthese games are illustrated by means of a painkiller example. Suggestions on alternate revenue maximization modelsare presented.