Correlation between predictability index and error performance in Customer Baseline Load (CBL) calculation

This paper explores the correlation between the low-frequency content of customers' consumption load, measured by a proposed index called predictability index, and the accuracy of Customer Baseline Load (CBL) estimation methods. The customer's consumption signal is transformed from time-domain to frequency-domain to separate its high- and low-frequency components. After reconstructing the time-domain equivalent of both high- and low-frequency signals, the predictability index for all customers is calculated. This proposed index is employed for the purpose of clustering the customers into different bins by a k-means clustering algorithm. The CBL for customers of each bin is estimated by two methods: CAISO and Randomized Controlled Trial (RCT). The data employed in this study belongs to Australian Energy Market Operation (AEMO), and are hourly consumption of 189 customers for the time span of a year (2012). In this paper, the correlation between average predictability index of each bin and its normalized average error is calculated. It is found that there is a strong correlation between the predictability index and the error performance of the CBL estimation methods.