Limited-Knowledge Economic Dispatch Prediction using Bayesian Averaging of Single-Node Models

A great deal of literature examines economic dispatch from the perspective of a grid operator, assuming full knowledge of generating unit costs and constraints, network topology, line capacities, etc. But others without access to the complete, data-intensive model may also wish to predict unit dispatch under various scenarios. This paper develops a Bayesian approach to predicting future economic dispatch, which relies only on historical dispatch observations, as well as general assumptions of operating costs. The approach uses a Markov-chain Monte Carlo method to create an ensemble of network-free models, which capture fundamental properties of economic dispatch (like merit order and marginal generator behavior) under particular “regimes” of grid operation (a particular range of load, set of congested lines, set of committed units, etc.). It then uses Bayesian averaging over many of these simple models to create an ensemble model, which approximates economic dispatch under general operating conditions. A case study using data from New York is used to verify the ensemble model.