Learning-based Multivariate Real-Time Data Pruning for Smart PMU Communication

This paper proposes a novel machine learning-based multivariate real-time data pruning and prediction framework for smart PMU (phasor measurement unit) communication. In an Internet-of-Things (IoT) enabled smart grid monitoring application, the proposed data-driven pruning technique exploits cross- and auto-correlation in multiple attributes sensed by a PMU (IoT node). The attributes are classified into base and nonbase groups based on their ability to aid prediction of the remaining attributes. The idea of transmitting only base attributes reduces the data dimensionality significantly. A reconstruction algorithm is designed for the edge node (local Phasor Data Concentrator) for efficient data reconstruction. The performance of the proposed framework is evaluated on large-scale real-time data from the PMUs. Comparison of the proposed technique with the closest state-of-the-art multi-threaded uni-variate data pruning algorithm in literature demonstrates around 40% more bandwidth saving and ∼42% reduction in retraining count.