A gross error detection method based on 3MAD-GRW-MMMD

Due to various reasons, there are inevitably gross errors in the data collected in the field. Gross errors need to be detected for modeling data to keep soft-sensing model accurate. Because different variables have different degrees of influence on modeling, gross error detection method is proposed, which is based on 3MAD-GRW-MMMD weighted clustering analysis. Combined with 3MAD method, this new method avoids the affect of large errors which are in a single variable before detecting gross errors of multivariate data. Meanwhile, based on the classic clustering algorithm-MMMD method, the new method adds the appropriate weights to the calculation of variables using gray relational analysis (GRA). Then the gross errors of the dominant variables in the multivariate data sets are detected. In addition, this new method is used to process the data sets of relevant parameters in penicillin fermentation and the smelting of LF refining furnace. Experiments and simulation results show that gross error detection method based on 3MAD-GRW-MMMD weighted clustering analysis is real-time, accurate, economical and reliable.