Efficient Quality Variable Prediction of Industrial Process Via Fuzzy Neural Network with Lightweight Structure

Quality Variables of industrial processes generally require to be obtained as fast as possible. In this paper, a correlation-wise self-organizing fuzzy neural network (CwSFNN) for efficient quality variables prediction of industrial process is proposed. Firstly, the correlation-wise self-organizing mechanism is developed by calculating the correlations between quality variables and fuzzy rules to optimize the network structure. The fuzzy rules of CwSFNN are generated or pruned systematically during the learning process, which can both improve the modeling performance and decrease the computational complexity. Moreover, the loss performance and convergence of CwSFNN are theoretically analyzed to ensure its successful application in practice. The benchmark Tennessee Eastman process (TEP) and real-world aluminum electrolysis process are presented to verify the effectiveness of CwSFNN. The experimental results show that the proposed CwSFNN performs better performance in both quality variable prediction and computation cost compared with some advanced methods. The source code of proposed CwSFNN is available at https://github.com/wjiecsu/CwSFNN .