An Improved Bayesian-Based Wavelet Package Denoising Method for Data Reconciliation to Coking Chemical Process

Measurement data usually do not reflect actual chemical processes correctly because of inevitable errors in measurement, which is known as unbalance of measurement data. Data reconciliation and gross error detection are methods of processing measuring instruments that are inconsistent with mass and energy balances. At present, research on data reconciliation mainly focuses on the steady state model of the linear system. However, data reconciliation cannot be implemented in cases of sudden changes in production conditions, operations or instability, or discontinuous data acquisition. An improved Bayesian based on wavelet package denoising for data reconciliation is proposed. After a measurement signal is decomposed into a certain level, the optimal tree is selected with Bayesian classification. Bayesian classification was used to select the node of the optimal tree that would accelerate the calculation speed. The simulation and the coking chemical application show that the improved denoising method contributes to the effectiveness and efficiency of data reconciliation.