A Hybrid-driven Soft Sensor with Complex Process Data Based on DAE and Mechanism-introduced GRU

With the increasing complexity of industrial processes, process big data inevitably has strong nonlinear, dynamic and noise problems, which restrict the accurate establishment of data-driven or mechanism-driven soft sensor models. Therefore, a soft sensor based on the denoising autoencoder and mechanism-introduced gated recurrent units is proposed. The denoising autoencoder is used to denoise the data. The mechanism-introduced gated recurrent units are used to introduce the information contained in the mechanism model and extract the dynamic characteristics of the process data by deep learning. This soft sensing method can deal with various problems of complex process data and introduce mechanism model for hybrid-driven modeling, which improves the prediction performance of the soft sensor. The effectiveness and superiority of the method are verified by the industrial case of predicting the thermal deformation of an air preheater rotor.