Distributed Fog Computing Architecture for Real-Time Anomaly Detection in Smart Meter Data
Rituka Jaiswal, Antorweep Chakravorty, Chunming Rong
Abstract
The use of Fog Computing for real-time Big Data monitoring of power consumption is gaining popularity. In traditional systems, Cloud servers receive sensor Big Data, perform predictions and detect anomalies or any threat patterns and then raise the alarms. With exponentially increasing sensor data, Cloud servers are becoming impractical to process this data because of the issues of volume, velocity, variety, network bandwidth, real-time support and security issues. Fog Computing is introduced as a Distributed Computing paradigm that uses intermediate Computing infrastructure for processing to overcome the limitations of Cloud Computing. In this paper, we propose a hierarchically Distributed Fog Computing architecture to deploy machine learning based anomaly detection models for generating insights from the collected Smart meter sensor data from the household. The anomaly detection is divided into two steps: model training and anomaly detection. We perform detailed analysis and evaluation of the models using standard open datasets obtained from UCI machine learning repository. The results confirm the efficacy of our proposed architecture. We used open source framework and software for our experiments.