An LSTM-based predictive monitoring method for data with time-varying variability
提出一种结合LSTM预测区间和神经网络变异性预测的在线监测方法，用于检测具有时变变异性的过程数据中的结构性变化，仿真和案例验证其优于其他基于神经网络的均值偏移检测方法。
The recurrent neural network and its variants have shown great success in processing sequences in recent years. However, these neural networks have not aroused much attention in the process monitoring literature. Whereas the flexibility and outstanding performance of these models enable them to be solutions to many problem settings within process monitoring. One such problem setting is process data with inherent time-varying variability, an aspect of data that is often ignored. Therefore, this paper proposes a prospective monitoring method for detecting structural changes in data with inherent time-varying variability. We integrate long short-term memory (LSTM) prediction intervals with a neural network (NN) for variability prediction to detect structural changes in data online. We compare our proposed method to benchmarks in a simulation study, showing that the proposed model outperforms other NN-based predictive monitoring methods for mean shift detection. The proposed method is also applied in two case studies to sensor data, which confirms that the proposed method is an effective technique for detecting shifts.