Demand response estimation based on aggregate data from heterogeneous users

The article presents methodologies for the prediction of the aggregate consumption of a large number of heterogeneous energy users with deferrable loads. It is assumed that the users respond to an identical coordinating signal (e.g. a price) broadcast by the system operator. The main contributions of the article are two: (i) the presence of heterogeneous users is explicitely addressed by considering the possibility of several 'classes' of consumers; (ii) the described methodologies provide predictions and estimates that have the advantage of making use of aggregate data only. When a dynamic response mechanism is implemented, the presence of consumers with deferrable loads introduces memory in the system in the form of the backlogged demand that has to be satisfied in the near future. From the perspective of the system operator, an accurate estimate of the current backlogged demand is of paramount importance, since it allows one to better predict what the aggregate consumption will be for a given coordinating signal. Thus, mathematical models need to be derived in order to predict future aggregate consumption using the accessible information. Unfortunately, consumers tends to be highly heterogeneous, depending on several factors including individual preferences, the interrupt-ability or the criticality of the tasks and the presence of hard deadlines. These factors can not be measured directly. Indeed, both privacy issues and practical constraints on processing and communicating data are likely to prevent utility companies and system operators from collecting and using measurements at the individual user level. Thus, methodologies for predicting aggregate demand response from aggregate data, such as the ones suggested in the article, can help overcome all these challenges.