Optimization of online system for COD measurement in water pollution through bias correction and uncertainty evaluation over an entire range of materials

Abstract

Chemical oxygen demand (COD) is an item of water pollutant that must be measured by the State. When undertaking COD measurement tasks, compared with the reference operation (as the Y-method) in the laboratory, the online system (as X-method), a process stream analyzer of automatic monitoring in the domestic field, is much superior. However, we believe that X-method is a non-standard system, and its measurement reliability deserves attention. In this paper under top-down measurement condition, we propose to use the variable error fitting of Deming regression, establish a bias correction function between the two methods, based on the matrix sample size at different levels by our power curve estimate in chronological order, and carry out a more rigorous check on the reliability of X-method. To ensure that the residuals derived from the bias correction function is in an independence and identical distribution (i.i.d) condition, we strongly recommend a more rigorous and robust Anderson Darling (AD) hypothesis test technique for confirmation. If the AD null hypothesis is held, we can believe that the reliability of X-method is validated, and incorporate all cumulative effects, with the minimized complex correlation, brought in the bias correction function fitting process, into the uncertainty estimate over an entire range. The practices we advocate in this paper are very helpful in raising the quality level of the online system.