Extracting the Weather Response from Long-Term Hourly Electricity Load Data in an Eastern Region of the United States

Abstract:

Understanding climate change impacts on the energy sector requires understanding how electricity consumption responds to weather conditions, such as temperature. This study applied a state-space model to 22 years (1993-2014) of publicly available hourly load data from the Pennsylvania-New Jersey-Maryland (PJM) Interconnection. Prior to our analysis, we removed long-term trends which are usually considered to be related to socio-economic and demographic factors, in the various sub-regions of the PJM interconnection to focus on the response to weather. The state-space models were comprised of weekly cycle, autoregressive-moving average components, and regressions on temperature, relative humidity, and wind speed variables. A separate model was fitted for each hour of the day. We found that the best relationship between temperature and electricity load may occur with a lag depending on the time of the day. The base temperature giving optimal mean squared residual magnitude was found to be lower than the 65 ^oF (18.3 ^oC) value traditionally used for cooling- and heating-degree days calculations. Relative humidity, wind speed, and sometimes a past temperature-variability term also increased the predicative power of the model. A few outliers existed in the hourly load dataset, which were not predicted well by the model, but the other residuals of the models were <=5% of the observed values. Removal of the outliers did not significantly impact the estimated model structure or parameters.