Application of Artificial Neural Networks and Decision Trees for Optimizing Industrial-Scale Composting of Biodegradable Waste to Support Sustainable Waste Management

Résumé fourni par la source

Sustainable management of biodegradable waste is a key component of the circular economy and resource recovery strategies. Composting is a complex biological process whose efficiency depends on numerous operational and physicochemical factors. Under industrial conditions, continuous laboratory monitoring of waste properties is often limited by time and cost constraints, necessitating reliable predictive tools to support process management. This study investigates the use of artificial neural networks (ANNs), decision trees (C&RT), and principal component analysis (PCA) for optimizing the composting of biodegradable waste under industrial-scale conditions. The research was conducted at a full-scale mechanical–biological treatment facility in Poland processing both the organic fraction mechanically derived from mixed municipal waste and separately collected biowaste. A dataset containing 23 records was developed from operational parameters (airflow, water addition, turning frequency, and process duration) and physicochemical properties of composted waste, including moisture content (MC), loss on ignition (LOI), total organic carbon (TOC), respiration activity (AT4), and higher heating value (HHV). The best-performing neural model achieved a predictive accuracy of 0.999 (coefficient of determination R2 in the test set). For each of the neural networks, goodness of fit indices were also determined: MAE and RMSE. PCA confirmed strong relationships among key waste properties, while decision tree analysis identified airflow as the dominant operational factor affecting MC, LOI, and TOC, whereas turning frequency had the strongest influence on AT4. The results demonstrate that machine learning tools can effectively support industrial composting optimization by predicting operational parameters required to achieve desired waste stabilization characteristics, providing practical decision-support solutions for composting plant operators. It is recommended to implement single-output MLP models for dynamic, real-time process control and C&RT rules as emergency procedures. This study aligns with circular economy principles and the Sustainable Development Goals by demonstrating the potential of artificial intelligence to enhance sustainable biodegradable waste management, resource recovery, and industrial composting performance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.