Defining virtual control group to improve customer baseline load calculation of residential demand response

Calculation of the customer baseline load has remained a critical challenge in demand response and especially in residential sector due to random and significant variance in the household daily electricity load. Unlike commercial and industrial demand response customers, the load pattern is highly dynamic for residential customers and additional relevant information may not be available. Up to date, significant efforts have been made to develop enhancement methods for customer base loads. To ensure such success, setting up of an independent group has been widely accepted as a better enhancement method. However, it requires a careful selection process and exclusion of selected customers which may be expensive in the long run.

Customer baseline load models, control group approach have been investigated for a reduction amount estimation. Regarding the merits and demerits of these approaches, a virtual control group approach has been proposed to combine the advantages and eliminate these disadvantages. In the study presented by the authors: Eunjung Lee (PhD candidate), Kyungeun Lee, and led by Professor Wonjong Rhee from the Seoul National University in collaboration with Dr. Hyoseop Lee from Encored Technologies and Euncheol Kim from Korean Power Exchange, a virtual control group customer baseline load (V-CBL) model combining simple customer baseline load (CBL) models and virtual control groups was developed to provide the benefits of the control groups with less burdens. The work is published in the research journal, Applied Energy.

To overcome the inherent selection bias problem, pre-collected participation information was used to form a virtual control group for each demand response event which performed well when used with the difference-in-differences method. For instance, the customer’s differential effects in time was compared with the virtual control group’s differential effect to calculate the final load reduction amount of a demand response customer.

As proof of the concept, it was necessary to evaluate the virtual control group customer baseline load robustness based on real-world data obtained from a pilot residential demand response program. In particular, the selection bias and performance were evaluated against the traditional models in terms of the mean error and mean absolute error. On the other hand, an analysis of the actual demand response event days was provided beside the analysis based on the non-event days.

Unlike the previous works, the present study is innovative in many ways. First, it defines and shows the robustness and usefulness of a virtual control group. Secondly, instead of using a difference-in-differences based calculation method, the focus here is on explicit accuracy comparison of the traditional and proposed customer baseline load methods. The virtual control group customer baseline load reported the best performance for both program-wide impaction estimation and individual-level settlement estimation. While the mean error was significantly enhanced, a limited improvement was reported for mean absolute error. This was attributed to the daily variations in each household’s electric load.

In summary, virtual control group customer baseline load provided the best accuracy as compared to simple customer baseline load models and the control group approach. The fact that it did not require the overhead of setting up a control group made it even more appropriate. Based on its improvements, Professor Wonjong Rhee in a statement to Advances in Engineering highlighted that the virtual control group customer baseline load approach is an effective and promising approach for solving the residential demand response related challenges.