Confidence interval estimation for short-term load forecasting

This paper presents a method to obtain confidence intervals (CI) for load forecast. It is based on the calculation of empirical quantiles of relative forecast error observed in the past. A classification is made between days for which the load forecast is difficult and those for which it is easier; this classification is made a priori, based on forecast error knowledge. Some CI's skills for the confidence interval are evaluated on a test period and show this method is more accurate and useful than a basic method simply based on the standard deviation of error. CIs provide a way of quantifying the uncertainty of the forecast. For TSOs, application fields are numerous. They could be used to assess as precisely as possible the operating margins. They could also be used to generate extreme demand scenarios, at a given risk level, for security analysis studies.