Availability and Failure Prediction for Telecommunication Equipment in Smart Grid

Telecommunication, with its recent advancement, has become an essential and crucial part of smart grids to enable reliable, efficient, and safe operations. Telecommunication predictive maintenance plays a key role in transforming the way in which the electricity grid is monitored. By predicting the availability status and failure possibility of telecommunication equipment, the operators in the smart grid control center can make proper decisions and efficient maintenance scheduling. In this work, a monitoring system for telecommunication equipment in the smart grid is proposed. The system includes three main components: availability prediction, failure prediction, and visualization. Time series analysis is performed using K-means clustering to classify the availability time series into clusters with similar behaviors. For each cluster, an ensemble boosting algorithm is applied to predict the availability value of each piece of equipment in the next hour, 12 hours, and 24 hours. Fully convolutional network (FCN) and Residual network (ResNet) are employed to predict the failures as a classification task. With availability prediction, the Extreme Gradient Boosting (XGBoost) model shows accurate prediction results with a mean absolute error between 1 - 9%. For failure prediction, deep learning outperforms the ensemble models with an F1 score of 91%. The prediction results are visualized in a Grafana dashboard to facilitate the smart grid monitoring services. Telecommunication automation coupled with appropriate visualization tools enables the advancement in the smart grid field.