Anomaly Detection Technology for Cloud Manufacturing System based on Data Denoising and Feature Optimization

Abstract

Aiming at the problem that the traditional anomaly detection method based on threshold cannot effectively detect sensor numerical anomalies in cloud manufacturing system, this work proposes a new method to detect some sensor numerical anomalies form the industrial control system. It is the central part of a cloud manufacturing system. Firstly, this work constructs a Savitzky-Golay (S-G) filter to reduce data noises. Furthermore, an extreme learning machine based on genetic algorithm (GA-ELM) model is proposed to detect sensor numerical anomalies form the industrial control system. The genetic algorithm (GA) is used to reduce feature dimensions from 51 to 10 and the extreme learning machine algorithm (ELM) is used for classification to achieve the purpose of anomaly detection. Finally, using the public dataset called Secure Water Treatment (SWaT), the classification accuracy is 98.96%. It shows a better performance of the proposed method.