Load forecasting and weather normalization

Several utilities have noticed that since the Great Recession of 2008-2009, load forecasting models have been less accurate. New patterns of customer behavior plus increasing effects of global climate change may be causing this. Brattle has developed tests and alternative specifications for load modeling to deal with these changes. For example, for PJM Interconnection, the largest regional transmission organization (RTO) in North America, Brattle experts conducted a comprehensive review of its models for forecasting demand that had significantly under-estimated recent peak loads. We re-estimated new models with specifications that fit the historical data equally well and corrected for serial correlation in the error term. We found that the forecast error was entirely due to an inability to accurately forecast weather and recommended more detailed modeling of forecast uncertainty. More recently, Brattle recommended that a gas distribution utility in the Mid-Atlantic switch from a 30-year weather normalization to a 10-year normalization for similar reasons.