A Soft-Sensing Model for Water Content Based on PCA and MNN

Abstract

Pneumatic drying process of ore concentrate is a typically complex industrial process, which involves the theory of gas and solid flow, heat and mass transfer and drying kinetics. There are a large number of factors affecting the pneumatic drying process. As a key parameter of the drying process, the water content of the dried ore concentrate has effect on the stable operation for the smelting process directly. The manual measurement of the water content has serious time-delay, which influences the operation optimization of the drying process. It is significant to research on the soft-sensing method for water content in drying process. The neural network model for soft sensing with many inputs and very complicated model architecture is very time-consuming to train and easy to over-fit the data with low predication accuracy and bad robustness. A soft-sensing model for water content based on PCA (principal component analysis) and multiple neural networks (MNN) is proposed to solve this problem in this paper. Firstly, PCA method was used to decrease the number of input variables, and then classified the data by k-means algorithms based on evolutionary strategies. Each kind of data after clustering is used to train a neural network sub-model. Finally, these sub-model are combined using principal components regression (PCR) method to obtain a soft-sensing model. Simulation results of the data from the practical production process show that the model can effectively sense the water content in the pneumatic drying process.