Abstract

Models for computing economic dispatch and prices in wholesale electricity market pools are typically deterministic multiperiod mathematical programs that are solved in a rolling horizon fashion. In convex settings with perfect foresight these optimization problems yield dispatch outcomes and locational marginal prices that solve a competitive equilibrium problem. Growing investment in renewable energy has increased the uncertainty in net demand to be met by dispatchable generation. To accommodate this, stochastic programming models formulated using scenario trees have been proposed for economic dispatch. The use of these models in practice is challenging for several reasons. Market participants need to agree on the scenarios used for uncertain parameters in the model, and realizations of these parameters will be different from those in any modelled scenario. When updated in a rolling horizon fashion, stochastic models can misprice the option value of energy storage and the value of changing current dispatch to meet future ramping constraints. This leads to uplift payments that compensate participants for the fact that the system operator forecasts the future incorrectly. We present a class of new economic dispatch models that attempt to overcome these drawbacks, based on agent decision rules (ADRs). Forecasting future outcomes or scenarios passes from the system operator to market participants who implicitly make state-dependent offers of energy through these decision rules. We show how storage and ramping can be priced correctly in convex markets and illustrate the advantages of the new model through simple examples.