In the second step, we model the de-trended electric load Zt−Ẑtlt as a function of all of the covariates which are available in our dataset that typically drive the electric load at a daily level. We then estimate a local model on the residual signal Zt−Ẑtlt−Ẑtmt ...
dataset was split (the 25% test and 75% validation parti- tion), this may have allowed users to come very close to knowing the true future temperatures. I therefore used a flexible but simple method for fore- casting temperatures that could be easily tuned. Fig. 2 ...