QMC MIMT PROGRAM for Windows GROSS ERROR DETECTION Raw data often contain large systematic errors (gross errors) resulting from instrument miscalibration, malfunction or failure. If these errors are not identified and removed from the measured data, the reconciled data will present a distorted picture of actual process operation. Since the reconciliation process tends to spread gross errors among the variables to satisfy the constraint equations, the reconciled data may well be further from the true values than the raw data. The purpose of a gross error detection algorithm is to identify and eliminate gross errors from a data set, and to estimate the correct values of the discarded measurements. The MIMT algorithm is based on a statistical test for outliers applied to the residuals from the linear regression problem. The residuals are simply the differences between the measured and reconciled values of the variables. In the MIMT method, the regression problem is solved recursively. At each stage, the residuals are tested for outliers, and the measurement corresponding to the most significant residual is deleted from the set of measurements. Iterations are continued until none of the residuals is statistically significant. It is important to eliminate measurements one at a time since otherwise many valid measurements may be discarded due to the spreading of gross errors noted above. The MIMT algorithm also permits a measurement to be discarded only if so doing results in a feasible solution to the data reconciliation problem. A solution is considered feasible if the values of all variables fall between specified upper and lower bounds. For mass flow systems, lower bounds would typically be set at zero, while upper bounds might be set a twice the measured values. Thus, solutions that contained negative flow rates, for example, would be considered infeasible. System Requirements Available for: Win95/98/NT
|
|
* * * * * * * * * * * * * * * * * * * * * * * * *
* * * * * * * * * * * * |