Soft Sensor of 4-CBA Concentration Using Deep Belief Networks with Continuous Restricted Boltzmann Machine

Abstract

In order to improve the quality of industrial products and effectively reduce the cost of raw materials, the prediction of 4-CBA concentration in industrial PTA oxidation process is very important. A soft sensor model based on the deep belief network (DBN) and BP neural network is proposed to predict the 4-CBA concentration in the PTA industrial production process. The deep belief network is composed of a continuous restricted Boltzmann machine. Unsupervised learning method is used to train the deep belief network in which the weights are updated by the error between training data and reconstruction data. The outputs of the deep belief network are the inputs of the BP neural network. The simulation results by use of real industrial data show that the soft sensor of 4-CBA concentration based on DBN-BP neural network has excellent prediction accuracy.