Prediction of Alarm of Insulation Monitoring System on Customer Facility using Random Forest

Abstract

In order to cope with the shortage of electrical technicians, it is expected to the improvement of work efficiency by the introduction of advanced technology, such as the Internet of Things (IoT) and Artificial Intelligence (AI). In the safety inspection of electrical facilities, insulation monitoring is expected to systemize correspondence judgment based on data such as measured leakage current.

In this study, from the data of the security business core system such as leakage current value measured at the periodic inspection and weather data, we created some models to predict the leakage current measured when the abnormality warning was issued and per customer and the presence or absence of alarms on the next day. The combination of the best explanatory variables makes the model more accurate. Variable importance analysis using Random Forest (RF) was performed to find variables that are important for each objective variable. This analysis shows that the accuracy of the prediction model of the leakage current is the highest when the explanation variable is the data of the security business core system, the weather data, the presence / absence of alarms on the previous day. Other predictive models need further verification. As a result of variable importance analysis, We found out that the leakage current value at the periodic inspection and the time of alarm are important for all purpose variables.