A soft sensor model based on CNN-BiLSTM and IHHO algorithm for Tennessee Eastman process

Accurate measurement of key variables plays an important role in on-line monitoring and control optimization of industrial processes. Soft sensor provides an effective method for measuring key variables. Deep learning model has strong learning ability and non-linear mapping ability for high-dimensional data. Many scholars have applied it to soft sensor, but there are many parameters, so it's necessary to optimize the parameters. In addition, how to select appropriate features for soft sensor modeling is also a key issue. The selection of appropriate variables can help to improve the performance of soft sensor model. Therefore, this study proposed a soft sensor model based on convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) and improved Harris hawk optimization (IHHO) algorithm for Tennessee Eastman (TE) process. First, Random forest (RF) algorithm is used to select auxiliary variables. Secondly, the Circle mapping initialization and the search strategy of simulated annealing (SA) are introduced to improve performance of HHO algorithm. Finally, the CNN-BiLSTM model is constructed for soft sensor modeling of key variables for the TE process. The simulation results prove that the proposed soft sensor model has good performance and high accuracy, can meet practical application of industrial engineering.

Automated summary

This study proposes a soft sensor model based on convolutional neural network - bidirectional long short-term memory (CNN-BiLSTM) and improved Harris hawk optimization (IHHO) algorithm for the Tennessee Eastman (TE) process. The model uses random forest (RF) algorithm to select auxiliary variables and incorporates circle mapping initialization and simulated annealing (SA) search strategy to improve the performance of the HHO algorithm. The CNN-BiLSTM model is constructed for soft sensor modeling of key variables in the TE process.