Data Reconciliation Method Of Multiphase Batch Process Measurement Data With Incomplete Measurements

Batch processes are one of the important industrial production methods and have the advantages of customization,flexible production and high added value of products.It has been widely used in fine chemicals,bio-pharmaceuticals and food processing.With the rapid development and application of sensing and measurement technology,the production of batch processes provides a wealth of process measurement data,which offers a basis and guarantee for data-driven online monitoring,production control,optimization,and improves the economic benefits of the enterprise.However,due to the influence of factors such as production on-site environmental fluctuations,makes the process measurement data become abnormal and reduces the quality of the process measurement data and the industrial application effect of the batch process monitoring and control technology.The existing detection method of the batch process measurement data utilizes multivariate statistical analysis or machine learning theory which does not fully consider the multiphase characteristics of the batch process and the characteristics of the process measurement data.This leads to the reduced of fault detection rate for the process measurement data and lack of the data reconciliation of the abnormal process measurement data at the same time.There are some incomplete measurements in the actual batch production process,which make the process measurement data abnormal.Therefore,considering the multiphase characteristics of the batch process and the characteristics of the process measurement data,study the method of process measurement data reconciliation under incomplete measurements and reconcile the process data can promote industrial applications batch process monitoring and control technology.By analyzing the multiphase characteristics of batch processes and process measurement data characteristics,this paper studies the data reconciliation method and proposes a static-timing constrained fuzzy clustering(SCFC)partitioning method of the multiphase batch process measurement data under incomplete measurements.The partitioning method is implemented by iteratively comparing the initial fuzzy membership and the timing fuzzy membership.Combined with the support vector data description(SVDD)algorithm,a fault detection method of the multiphase batch process abnormal measurement data based on SCFC-SVDD is proposed.This fault detection method finds the abnormal data by comparing the online process measurement data spherical center distance and the multiphase period hypersphere radius and constructing a fault diagnosis strategy based on the process variable spherical center weight,which traces to the source of abnormal process measurement data.Based on just-in-time learning(JITL)and relevance vector machine(RVM),a JITL-RVM data reconciliation method of multiphase batch process data is established,which realizes the data reconciliation of the batch process measurement data with incomplete measurement.The experimental results show that the proposed SCFC partitioning method of the multiphase batch process measurement data can accurately divide the time phases of the batch process.The proposed fault detection method based on SCFC-SVDD of multiphase batch process measurement data can accurately detect abnormal measurement data.Compared with the traditional multiphase SVDD fault detection method,the proposed method has higher abnormal data detection accuracy.The established data reconciliation method based on JITL-RVM can reconcile the abnormal process measurement data with incomplete measurements.