Application Research Of Deep Learning Algorithm In The Soft Sensor Modeling

With the development of computer technology, modern industry technology is becoming more and more informative and intelligent. It is very necessary to control the process variables in the industrial production process. Soft sensor technology, as one of the advanced technology, can realize online real-time prediction of the process variables which are difficult to measure in the process of industrial production. Deep belief network is a kind of typical deep learning algorithm. The algorithm has obvious advantages in feature learning. Soft sensor modeling of 4-CBA content is the research object in this paper. Combining the deep belief network with the existing shallow algorithms, it further researches the deeper soft sensor model.This paper mainly studies the following aspects. There is a serious symmetric problem of the binary deep belief network algorithm in the process of dealing with the continuous sampled data. In order to solve the problem, the paper introduces continuous restricted Boltzmann machine. It fits the nonlinear function successfully. In view of the soft sensor modeling of 4-CBA content in the actual chemical process, the soft sensor model based on deep belief network is put forward. In the model training process, the sample information is fully utilized, and the prediction accuracy is improved. In the process of parameter optimization of deep belief network algorithm, the paper deeply studied the influence of the number of hidden layer nodes and the number of iterations on the test performance of the soft sensor model. The number of nodes in the hidden layer is analyzed, and some experience values are obtained. The convergence of the maximum likelihood estimation in the training process of the model is determined by the square sum of the residuals of the two adjacent layers of the visible layer, which improves the efficiency of the iteration. At the end of the paper, a deep belief network with multiple hidden layers is studied and the model performance of deep belief network is analyzed. Soft sensor modeling is realized for 4-CBA content. The simulation results show that the prediction accuracy of the 4-CBA soft sensor model based on the deep belief network is higher.