A Study Of Reliability Of Data Reconciliation Methods Using Random Numbers For Generating Pseudo Experimental Data

For a correct evaluation of the performance of any process, it is important to reconcile the data gathered in the plant from the material balance point of view for all the streams flowing in and flowing out of the process unit. There are two main approaches to reconciliation of data, which involve minimization of the sum of the squares of: (i) the closure residuals and (ii) the component adjustments; both weighed with the estimated variances of the components. Though, these approaches have served us well, it has not been possible to ascertain if the reconciled values for various components and the split factors obtained using these approaches are truly the best estimates. In this paper, this aspect has been investigated by analyzing pseudo-experimental data of known error structure on size classification of particles in a hydrocyclone. Three types of basic error structures were considered for the measured component values: absolute error, relative error and chi-square error. The error structure was modified by adding a random component of error using normally distributed random numbers corresponding to zero mean and different levels of variance. By carrying out a large number of simulation experiments, it has been shown that in the presence of random errors, irrespective of the basic error structure of data, which is usually unknown, the chi-square error minimization criterion can be used to obtain most reliable results. Keywords: data reconciliation methods, chi-square error minimization