Macro-micro Feature Aware Transformer for Dissolved Oxygen Prediction

The importance of dissolved oxygen parameters in industrial production processes is significant, as they impact the growth state and growth period of aquatic organisms. This paper proposes a dissolved oxygen parameter prediction method based on Transformer technology, which consists of three modules: 1) macro embedding: designed to capture the correlations between parameters and their time-related trends; 2) micro embedding: intended to learn the subtle differences in each time step characteristics; 3) lightweight macro-micro feature fusion module:aimed at integrating macro and micro features for predicting future changes in each parameter. Experimental verification demonstrated that this method has higher accuracy compared to traditional prediction methods, as well as strong generalization ability. It can predict multi-time-step or single-time-step dissolved oxygen concentrations and other parameters future trends, aiding industrial personnel in making decisions, and satisfying actual production requirements.