Abstract:
Sensors are being used in various fields such as smart factories, CPS, and digital twins. It is important to minimize and prevent losses by forecasting abnormal sensor signals when these sensors malfunction or when abnormal signals occur due to environmental effects. Recently, various studies have been conducted with LSTM, which is widely used as time series prediction, but there are not many studies that evaluate performance by applying this to sensor signals. In this study, an LSTM with excellent performance was designed with an optimal hyperparameter setting by applying a water distribution sensor containing 7-17% abnormal sensor data. In addition, as a result of performance comparison with ARIMA, it was found that LSTM was 94.29% superior on average in terms of precision and accuracy. CopyrightΒ© ICROS 2023.
