An Integrated Approach based on Supervised and Unsupervised ML algorithms for Efficient Design of Demand Response Programs

Abstract

Demand Response (DR) programs provide high potential incentive-based management solution for targeting Demand Side Management (DSM) objectives such as reducing system's peak demand. The incentives can be designed in an effective manner, by leveraging the smart meter consumption data to analyze customer's load pattern variation for a defined duration and find potential target periods and customer groups to incentivize. Clustering and load profiling is a well-known methodology in DR program design. Algorithm selection, is however based on compactness of clusters. This paper proposes a methodology to select Clustering algorithm based on the set objective. In addition, the approach- based on data stratification, supervised and unsupervised Machine Learning (ML) techniques, serves to find key potential target time periods, customer groups, and decision threshold points (in kWh) for incentive-design, in efficient, direct, customer-friendly and objective-oriented manner, which can be incorporated to plan strategic incentive-based DR programs.