A novel spatial-temporal fusion deep neural network for soft sensing of industrial processes

This paper proposes a novel spatial-temporal fusion deep neural network for soft sensing of industrial processes. Different from previous deep learning based methods, the proposed model considers a new structure by combining the gated recurrent unit(GRU) and deep convolutional neural network(DCNN). The temporal correlation is captured by the GRU framework and the spatial correlation/input-output relationship is revealed using the DCNN structure. The resultant features are then fused and fed into a dense layer to produce the final estimations. A backpropagating algorithm is used to optimize the parameters of the model. The developed soft sensor considers both the spatial and temporal information and hence are more accurate in soft sensing of important quality variables that are difficult to measure in industrial processes. The effectiveness and feasibility of the soft sensor is verified by an application to the sulfur recovery unit(SRU).