Data-Driven Soft Sensor Model Based on Deep Learning for Quality Prediction of Industrial Processes

摘要: Fermentation process is a time-varying, nonlinear and multivariable dynamic coupling system. Therefore, it difficult to directly measure the key biological variables using traditional physical sensors during of fermentation, which makes monitoring real-time control impossible. To resolve this problem, data-driven soft sensor modeling method based on deep neural network (DNN) proposed in paper. This suitable for large amount data enjoys high efficiency robustness. At same time, an adaptive moment estimation (Adam) algorithm used optimize hyper-parameters DNN model, technique efficient stochastic optimization that only requires first-order gradients with little memory requirement. The consistent correlation determine auxiliary model. penicillin l-lysine fermentation processes are taken as research object, substrate concentration, cell product concentration selected target variable. performance established model evaluated through indexes mean square error (MSE), root-mean-square (RMSE), absolute (MAE). simulation results show prediction DNN-Adam good compared gradient descent (SGD) momentum algorithm. It verified can make more accurate quality process, has higher accuracy than DNN-SGD method.