摘要

In order to make competitive electricity markets effective, bidding generation companies (GENCOs) need to estimate market demand models according to information available to each of them. However, many stochastic factors (e.g. weather, demand side features) make it very hard for GENCOs to accurately capture the actual market demand in a model. Each GENCO might hold an estimated model deviating, from the real market model as well as from its peers'. Little work has been done in discussing the impacts of model deviations towards the design of GENCO's bidding strategies.
In this paper, the effects of model deviations upon the equilibrium-oriented bidding methods (EOBMs), more specifically conjectural variation (CV) based methods, are studied. We relax the strong assumptions that one uniform and accurate market demand model is employed by all GENCOs in the basic CV-based learning bidding algorithm (CVBA). In this work, the market demand model utilized for bidding by each GENCO is different from each other and from the actual market model as well. The impacts of such model deviations are analyzed from both theoretical and simulation perspective. Theoretical analyses point out that as a consequence of the model deviations it is possible that the basic CVBA algorithm will bring the bidding process into an unstable state. In order to eliminate the effects from inaccurate modeling, a CV-based learning bidding method with data filtering capabilities is proposed. Several sets of simulations have been done to test the impact of the model deviations. The simulation results confirm the theoretical analyses. The feasibility and effectiveness of the proposed bidding methods are also verified. The proposed algorithm can bring systems into stable state even when model deviations exist.

全文