When determining energy consumption baseline through regression analysis, as recommended by best practices, sensible energy managers pick a formula that relies of physics of the process, not a formula that produces high correlation (R2) between predicted and measured data.
In vast majority of cases relation between energy consumed and it’s driver is linear. Here are several examples of drivers of linear consumption: production volume, degree-days, amount of heat rejected in refrigeration, volume of steam produced, volume of compressed air.
Higher power polynomial always produces higher correlation, it does mean it better describes process. Reliance on such polynomial to assess savings may bring such ridiculous results as savings exceeding total consumption.