Intelligent operational control method integrating human expertise and real-time data for alumina digestion process

Abstract

The caustic ratio in discharge is a crucial quality index for the alumina digestion process (ALDP), reflecting both production efficiency and product quality. However, the caustic ratio cannot be measured online. Traditional human-based operational control often leads to delayed and inaccurate decisions due to inherent time delays of the ALDP and the prolonged duration of offline assays. To address this issue, an intelligent operational control method is proposed, comprising a caustic ratio endpoint prediction model and a fuzzy operational controller. The bi-directional gated recurrent unit (BiGRU) is employed within a dual-network paradigm to accurately predict the endpoint of the caustic ratio, adapting to time-varying operating conditions. A fuzzy system, informed by human expertise, is integrated with a closed-loop stable learning method to eliminate the impact of modeling errors on the stability of ALDP operation. The proposed method is capable of operating in both guidance and automated operational control modes. Industrial studies at a real-world alumina refinery in China demonstrated a significant improvement in the qualified rate of the caustic ratio compared to traditional manual operational control.