An Improved Extreme Learning Machine Based on Auto-Encoder for Production Predictive Modeling of Industrial Processes

Abstract

Industrial process data has the characteristics of complexity, variability and noisy, which brings challenges to data-driven production predictive modeling for industrial processes basing on the traditional extreme learning machine (ELM). Therefore, this paper proposes an improved ELM based on auto-encoder (AE) (AE-ELM). The AE can extract the main features with lower-dimension by eliminating the linear correlation among the original complex data. Then, the main features are used as the inputs of the ELM. For the purpose of verifying the effectiveness of the proposed method, the AE-ELM model has been experimented on the production prediction of the pure terephthalic acid (PTA). The experimental results prove that the AE-ELM is less sensitive to the structure of the traditional ELM and principal components extraction based robust ELM (PCE-RELM). Moreover, the modeling accuracy can be improved by 2.4%, which has certain guiding significance for process modeling and production prediction.