A novel intelligent modeling framework integrating convolutional neural network with an adaptive time-series window and its application to industrial process operational optimization

With the increasing complexity of industrial processes, it becomes more and more difficult to set up process operational optimization models. Recently, the convolutional neural network (CNN) has been widely and successfully applied to extracting useful information due to its deep learning capability. Aiming at extracting useful information concerning operational optimization from complex industrial process data, this paper proposes a novel framework integrating CNN with an adaptive time-series window (ATSW-CNN). The proposed ATSW-CNN method is composed of four kinds of network layers, i.e. the proposed adaptive layer, convolutional layers, pooling layers and fully connected layers. The proposed adaptive layer provides CNN paradigms with a capability of adaptively selecting appropriate time-series windows for different steady-state operational optimization data. As a result, the proposed ATSW-CNN method can effectively extract steady-state optimal operating conditions from process time-series data. In order to validate the performance of the proposed ATSW-CNN, simulations on an industrial furnace are carried out. Simulation results verify the effectiveness of the proposed method, which demonstrates the proposed ATSW-CNN method is applicable for searching steady state operating strategies.