SC13 Denver, CO

The International Conference for High Performance Computing, Networking, Storage and Analysis

Accurate Change Point Detection in Electricity Market Analysis.

Student: Will Gu (Lawrence Berkeley National Laboratory)
Supervisor: John Wu (Lawrence Berkeley National Laboratory)

Abstract: Electricity is a vital part of our daily life; therefore it is important to detect any irregularities such as those during the California Electricity Crisis of 2000 and 2001. Many advanced machine learning algorithms exist to detect anomalies, but they are computationally expensive and could not be applied on large data sets such as those from electricity markets. In this work, we develop a strategy to accelerate the computation of the Gaussian Process (GP) for financial time series within the framework of a Change Point Detection (CPD) process, reducing its computational complexity from O(N^5) to amortized O(N^2). We apply this fast algorithm to correlate the change points detected with the known events during the California Electricity Crisis. This calculation would have been impossible without the fastest supercomputers.

Poster: pdf
Two-page extended abstract: pdf

Poster Index