A Class Of Monitoring Method Based On Data Driven For Batch Processes

Batch process,as a significant production mode,has been openly utilized in the dyes,food,pharmaceutical,and other fields.At present,establishing a monitoring system for batch processes to execute anomaly detection has become a hot issue in modern industrial process control field.In this paper,we study how to monitor an ongoing batch process real-timely and precisely based on the ?-nearest neighbor rule,and the contents are as follows.(1)A real-time monitoring method using the ?-nearest neighbor rule is proposed.Firstly,cutting the measurement data into multiple time slice data along the time axis,and then building a model at each sampling time based on the time slice data,finally using these above models to monitor whether the ongoing batch process is abnormal in real time.(2)A fast ?-nearest neighbor algorithm is proposed to improve the time efficiency of the anomaly detection.The key idea is to narrow the range of near neighbors searching by E2 LSH.In detail,for each sample,we filter the samples away from it by E2 LSH firstly,and then apply the ?-nearest neighbor rule to find its k neighbors in remainder samples to execute the anomaly detection.(3)With the multiple phases characteristics of batch processes,a phase-based real-time monitoring method using the ?-nearest neighbor rule is proposed.Firstly,utilizing the random projection and K-means clustering algorithm to divide the entire manufacturing process into several sub periods,and then establishing the sub-phase models to monitor the batch process real-timely.