Influence of Temperature and Load Forecast Uncertainty on Estimates of Power Generation Production Costs *J. Valenzuela *M. Mazumdar **A. Kapoor *Industrial Engineering Department University of Pittsburgh Pittsburgh, PA 15261 **GPU Energy Reading, PA 19612 Abstract- The paper considers the use of production costing models for short term operations planning. In this period an accurate forecast of the hourly ambient temperature during the study horizon and the knowledge of the initial operating states of each generating unit are assumed to be available. Uncertainties in the production cost and marginal cost resulting from the generator availabilities and hourly loads are considered. Measures of these uncertainties are obtained from an analysis of two years’ actual hourly load data. Monte Carlo simulation is used to estimate the contributions of uncertainties resulting from load and generator availabilities to the variance of production costs. It is found that ignoring the ambient temperature forecasts and correlation among hourly loads results in inaccurate prediction of costs, and that load uncertainty accounts for a significant portion of the variance of production costs. The initial states of the generating units also have an important effect on the costs. I. INTRODUCTION Production costing models are widely used in the electric power industry for the purpose of forecasting the cost of electricity. This use covers short, medium and long-term electric utility operations and planning. For example, in the very short-term, which may range from the next hour to the next 24 hours, the models are used to forecast electricity prices and guide decisions regarding unit operations. In the moderate term horizon which may last anywhere from one month to one year, these models are used to write supply contracts with major users and vendors, and for the long-term spanning a period of one year and beyond, these models are used to guide decisions regarding acquisition and disposal of generation assets. For the short-term use of these models, the current operating states of the generation units and the prevailing weather should play an important role whereas in long-term planning, steady state conditions can be assumed. There has been an increased emphasis during the recent years on predicting the short-term marginal costs. Marginal cost for a power generation system at a given hour is defined here to be the operating cost ($/MWH) of the last unit used to meet the load prevailing at this hour. In a deregulated environment, marginal costs are expected to play an important role. These costs will be used to determine the optimal tariffs based on market clearing prices as well as the operating profit of each individual generator. In this paper we consider the use of the production costing model for the short term operations planning and examine the effect of temperature, initial states of the operating units and load on the estimated production and marginal costs. The production cost over a given time interval is a random variable because it is dependent on the uncertainty associated with the availability of the generating units as well as the load. For longer time horizons, the uncertainty associated with fuel prices is also a contributing factor. The well-known Baleriaux formula [1] accounts only for the uncertainty due to the availability of generating units and provides a formula for the expected value of production costs. The variance of the production cost [9] has only recently been studied, although this was deemed by Wollenberg [12] to be a much needed topic for research in his discussion on the paper by Sager, Ringlee and Wood [10]. The variance measure provides information on the fluctuation of these costs and is essential in risk evaluation. The increasingly competitive markets for electric power and the impending deregulation of the electric power industry call for an accurate treatment of the variance of production and marginal costs. Baleriaux’s formulation represents the hourly load in the form of the load duration curve (LDC) and the availability of the generating units by the forced outage rate (F.O.R.) index. It has been pointed out in the recent literature [9] that the LDC and the F.O.R. indexes do not provide sufficient chronological information for computing the variance of production costs. For this, not only is a chronological load model necessary, but the frequency and duration of generating unit outages also need to be taken into account. The papers by Ryan and Mazumdar[9], and Lee,