Adaptive segmentation and machine learning based potential DR capacity analysis

Abstract

By shedding their electricity consumption or getting ready to be shed, the Demand Response (DR) program compensates the shortage of generation or acts as spinning reserve. Depending on how the end-users join in the program. the potential capacity of DR is, therefore, a key issue no matter to system operator or DR aggregator. The proposed method employs adaptive k-means approach to evaluate the potential consumer as candidate participants of DR. The consumption prediction models of controllable appliances are constructed by Gaussian Processes for Machine Learning (GPML). Through combining the candidates' data and prediction models, the potential capacity is then achieved. Case studies evaluate the accuracy and efficacy of the proposed method with practical low voltage advanced metering infrastructure (LVAMI) data achieved from Taiwan. The results show good efficiency and practicability of the proposed method.