Anomaly Detection of Smart Grid Equipment Using Machine Learning Applications

Abstract

Many application systems in today’s smart grid network comprise a variety of middleware components, such as vibrating or spinning electrical equipment, data bases, storage, caches, and identification services, among others. Each component is a discrete bundle of physical or virtual computers that will generate a large amount of data in the form of logs and metrics, and failure of these high-vibrating machines will result in the system’s entire shutdown. As a result, the condition monitoring system for this smart grid equipment is more dependable and efficient in predicting the machine’s health ahead of time. To analyze the data and derive any inferences for future analysis and anomaly detection, we’ll need a separate system, which will take longer to handle data given by each component and will demand additional processing resources. As a result, in this chapter, machine learning methods are used to identify abnormalities or condition monitoring for smart grid equipment and machines. Here, we used two separate methods: multivariate statistical analysis by calculating Mob distance and auto encoders, which is an artificial neural network approach. Furthermore, the findings demonstrate that these apps are effective in identifying anomaly ahead of time, i.e. before a few days.