Fig. 4 and Fig. 5 show the laboratory measurements, as well as the estimated alumina concentration from the MRFbased soft sensor and the model based on back propagation (BP) neural network. It can be seen that the red line is closer to the blue line than the green line, that is, the MRF-based soft sensor can more accurately track the measured values. In order to make the comparison result clearer, Fig. 5 shows a detailed look at the overall accuracy of the soft sensor for the region [50, 80] samples. It can be seen that the tracking of MRF-based soft sensor is very good, with only a few differences. The model based on BP neural network obviously has greater deviation.