Implementaton aspects of On-line grid data analytics systems

Aligned with significant attention to smart grid, timely and precise analysis of grid data that is continuously reported from a massive number of field devices such as sensors, meters, and so on, is considered a key factor to accelerate smart grid deployment. However, many utilities today have no grid data analytic system, or have grid data analytic systems that are operated in an off-line manner. As a result, today data analytics systems are mostly offline analysis tools and cannot properly and effectively handle impending contingencies such as load surf, sudden voltage drops, etc. Motivated by this need, we have been building an online grid data analytic system, DMAS (Data Mining and Analytics System). In this paper, we describe lessons learned from the DMAS implementation using a set of real utility data. The discussion is focused on trade-offs between diverse implementation choices when faced with the objective of fast processing of a sequence of time series utility data, its correlation with other relevant AMI and SCADA data, and report or alarm generation. Further, we show performance results measured over the real data collected from more than 100k meters, with a grid topology comprising of distribution transformers, feeders, and substations, along with weather data, customer demographic data, and billing data.