Cost of temperature history data uncertainties in short term electric load forecasting

Short-term load forecasting has received a lot of attention from both researchers and practitioners. Many techniques, such as neural networks, fuzzy logic, and time series models, are developed to improve the modeling and forecasting process with varying success. Some research has also been devoted to improving the weather forecast or relieving the impact of uncertainties in the weather forecast. Nowadays, the quality of load and weather data history becomes a bottleneck of enhancing the forecast accuracy. With the emerging Smart Grid initiatives, the utilities have the opportunities to upgrade their infrastructure, which includes the advanced metering systems. These upgrades will potentially lead to a high quality load history in the near future. On the other hand, quality of weather history has been a concern of the small and medium sized utilities. This paper presents a recent study for a medium sized utility in eastern US. Multiple linear regression, which is currently deployed in this utility, is used to generate the base forecast in this paper. A Monte Carlo based methodology is proposed to quantify the cost of the data uncertainties in the temperature history. Finally, a cost-benefit analysis is performed to help the utilities decide how many weather stations should be installed.