Fault Detection Based on Sliding Window and Multiblock Convolutional Autoencoders
In order to further improve the fault detection performance and fully mine the timing and hidden feature information,a fault detection method based on convolutional auto encoder is proposed.On the basis of modeling the original information set,the modeling of cumulative information and rate of change information is added to enhance the mining of implicit information;The three reconstructed information sets are sampled by sliding windows,and time series feature extraction and modeling are performed based on convolutional auto encoders.Bayesian fusion of the decision results of the convolutional auto encoder is performed to obtain the statistics,and the control limit is determined by the method of kernel density estimation for fault detection.The method is numerically simulated and applied in TE process,and the simulation results confirm the effectiveness and detection performance.
万方数据
维普
