Application of Robust Extracting Outliers Algorithm in Fermentation Process Data Preprocessing
When the process monitoring model based on PCA is established, outliers would destroy covariance structure of model, so monitoring model deviations will affect the monitoring results. Because the CDCm outlier detection algorithm based on robust scaling underestimates the mean and standard deviation of the normal process data, some outliers will be not detected. The CDCm outlier detection algorithm based on modified scaling achieves an accurate estimate for the original data and can overcome the lack of robust scaling by using the maximum variable value to calculate the distance, using CDCm algorithm to obtain the shortest normal point between sampling values and center distance, which will obtain valid data, computing the first Markov distance of iterative algorithm, and by selecting the samplings corresponding the smaller distance value for iteration to obtain normal data. The robust scaling and modified scaling detection algorithm is applied to fermentation process and the test results have shown that modified scaling CDCm detection algorithm is more accurate and effective.
X. R. Zhang and B. Chang, "Application of Robust Extracting Outliers Algorithm in Fermentation Process Data Preprocessing", Applied Mechanics and Materials, Vols. 130-134, pp. 463-466, 2012