关键词: 深度学习, 深度置信网络, 神经网络, 软测量

Abstract: In industrial PTA production process, 4-CBA concentration is the important basis of PTA product quality evaluation. This paper combining the deep belief networks and BP neural networks proposes a soft sensor model of 4-CBA based on deep belief networks. Deep belief network is one kind of typical deep learning algorithm. The algorithm has remarkable superiority in feature learning. According to experimental results, a soft sensor model based on deep belief networks can predict 4-CBA concentration well. Compared with BP neural network model, the model based on deep belief networks has higher prediction precision.

Key words: deep learning, deep belief networks, neural networks, soft sensor