Dynamic Soft Sensor Development Based on Hierarchical Spatiotemporal Extreme Learning Machine for Anaerobic Digestion Process

The effective approach for dealing with kitchen waste is anaerobic digestion (AD) technology. The concentration of volatile fatty acid (VFA) determines whether anaerobic digestion process can proceed smoothly. This paper introduces a novel approach, the Hierarchical Spatiotemporal Extreme Learning Machine (HSTELM), for real-time quality variables monitoring. The proposed model utilizes the adjacency matrix and positional encoding to obtain the spatial and dynamic characteristics of the process data, respectively. Additionally, deep features of the process data are extracted by stacking Spatiotemporal Extreme Learning Machine Autoencoder (STELM-AE). Finally, an AD process in kitchen waste plants is introduced as a case study. The actual industrial data of the plant verify the suggested model.