Improved principal component analysis with double-weighted local outlier factor and its application in process monitoring for roasting furnace

Aiming at the monitoring problems of roasting furnace production process where the abnormal furnace states are difficult to detect, this paper proposes a modified PCA algorithm based on the combination of double weighted strategy and local outlier factor (DWLOF-PCA). The proposed method mainly solves two problems. Firstly, the LOF algorithm is introduced to construct the new monitoring indicators. Secondly, in order to obtain better online monitoring results, combining double weighting strategy and LOF algorithm to improve the new statistics. In the first-level weighting strategy, different weights are assigned to different components to highlight abnormal information. In the second-level weighting strategy, a moving weighted window is designed to weight the previous monitoring results to reduce the drastic fluctuation of the current results. Finally, the offline test is performed using the roaster field data of a smelter in Hunan.Compared with the existing PCA process monitoring algorithm, this method can effectively detect abnormal furnace conditions,the monitoring performance is greatly improved.