Prediction of aggregated power curtailment of smart grid demand response of a large number of building air-conditioners

This paper reports on the averaging effect on prediction errors in the fast automated demand response (FastADR) power curtailment amount calculated by our autoregressive and neural network models for a large number of multi-type office building air-conditioners. Our experiments showed the standard deviation of prediction errors decreased with the number of FastADR trials according to the so-called "inverse root N rule". This result will be useful for estimating the averaging effect in the prediction on resulting amount of power reduction due to FastADR for a large number of target facility loads.