LSTM Prediction Model for Abrupt Occurrence of Embedded Controlled Maintenance Operation of Facility

Abstract

Factory and building equipment may have embedded controlled maintenance operation which is sudden disturbance from the perspective of future energy management systems such as real-time pricing adaptive control. In this study, we investigated machine learning for a time-series model that predicts the occurrence of maintenance operations five minutes in advance. Assuming a case of seral hour history, we implemented a model as a Long Short Term Memory (LSTM) neural network. An example experiment result showed the accuracy of 0.93, threatscore of 0.6, and real-time pricing management improvement of 27%.