Intelligent control for food waste composting using observation-assessment-decision-action cycles: From technical...

Intelligent control systems (ICS) based on sensor technology and machine learning (ML) can improve the inefficiency and instability of traditional food waste (FW) composting processes, yet quantitative, deployment-oriented techno-economic assessments remain limited. This study reviews and analyzes the performance of common ML models within the FW composting, utilizing Observation-Assessment-Decision-Action cycle framework. The result showed that decision tree-based models demonstrate superior accuracy, while artificial neural network-based models are better suited to address data distribution differences caused by FW heterogeneity. Based on this, comparative technical-economic and Monte Carlo-based economic risk analyses were conducted at lab and pilot scales for an ANN-based ICS and a PID-based general control system (GCS). Results showed that, compared with GCS, ICS reduced the unit direct operating-cost intensity by 64.31 % (lab) and 47.81 % (pilot). Capital-intensity (TCI per ton) was also reduced by 39.16 % (lab) and 32.11 % (pilot) under ICS. Multi-factor fluctuation risk analysis revealed that both strategies remained loss-making at the laboratory scale, but ICS reduced the mean loss magnitude by 22.35 %, while at the pilot scale ICS increased the mean net revenue by 67.79 % relative to GCS. Overall, ICS improves economic robustness and profitability potential at deployment-relevant (pilot) scale while narrowing the loss margin at laboratory throughput. This study provides references for the research and implementation of FW composting ICS and lays the foundation for its technological iteration.