Value of Categorical and Probabilistic Temperature Forecasts for Scheduling of Power Generation

Abstract

Bayesian decision models are formulated for the use and evaluation of categorical and probabilistic forecasts of continuous variables. The models are applied to the problem of short-term scheduling of power generation in an electric system on the basis of a single-period temperature forecast. Likelihood functions are constructed using results of experiments conducted at the National Weather Service. The probabilistic forecasting scheme is of the type wherein the forecaster quantifies his degree of uncertainty in terms of variable-width, fixed-probability credible intervals. Each forecasting scheme, categorical and probabilistic, is evaluated in a coupling with two decision procedures: (1) an optimal (Bayesian) procedure which accounts for forecast uncertainty, and (2) a more conventional, non-optimal, procedure which disregards forecast uncertainty, but which would be optimal if the forecasts were perfect. Numerical examples are presented to illustrate the economic values of both types of forecasts, gains from probabilistic forecasts, and expected opportunity losses to be incurred by decision makers who ignore forecast uncertainty. Key words; Bayesian decision theory, value of information, meteorologic forecasts, probabilistic forecasts, electric power generation, power load forecasting.