Abstract：
Aerobic composting stands as a widely-adopted method for treating organic solid waste (OSW), simultaneously producing organic fertilizers and soil amendments. This biologically-driven biochemical reaction process, however, presents challenges due to its complex non-linear metabolism and the heterogeneous nature of the solid medium. These characteristics inherently limit the simulation accuracy and efficiency optimization in aerobic composting. Recently, significant efforts have been made to simulate and control composting process parameters, as well as predicting and optimizing composting product quality. Notably, the integration of machine learning (ML) in aerobic composting of organic waste has garnered considerable attention for its applicability and predictive capability in exploring the complex non-linear relationships of organic waste composting parameters. Despite numerous studies on ML applications in OSW composting, a systematic review of research findings in this field is lacking. This study offers a systematic overview of the application level, current status, and versatility of ML in OSW composting. It spans various aspects, such as compost maturity, environmental pollutants, nutrients, moisture, heat loss, and microbial metabolism. The survey reveals that ML-intervention predominantly focuses on compost maturity and environmental pollutants, followed by nutrients, moisture, heat loss, and microbial activity. The most commonly employed predictive models and optimization algorithms are artificial neural networks (47%) and genetic algorithms (10%). These demonstrate high prediction accuracy and maximize composting efficiency in the simulation and prediction of organic waste composting, alongside regulation of key parameters. Deep neural networks and ensemble learning models prove effective in achieving superior predictive performance by selecting feature variables in compost maturity and pollutant residue prediction of organic waste composting in a simpler and more objective manner.
摘要：
好氧堆肥是一种广泛采用的处理有机固体废物（OSW）的方法，同时生产有机肥料和土壤改良剂。这种生物驱动的生化反应过程，然而,由于其复杂的非线性代谢和固体培养基的异质性质，提出了挑战。这些特性固有地限制了好氧堆肥中的模拟准确性和效率优化。最近,已经做出了巨大的努力来模拟和控制堆肥过程参数，以及预测和优化堆肥产品质量。值得注意的是,机器学习（ML）在有机废物好氧堆肥中的集成因其在探索有机废物堆肥参数的复杂非线性关系方面的适用性和预测能力而引起了广泛关注。尽管对ML在OSW堆肥中的应用进行了大量研究，缺乏对该领域研究成果的系统回顾。本研究对应用层面进行了系统概述,当前状态,ML在OSW堆肥中的多功能性。它涵盖了各个方面，如堆肥成熟度，环境污染物，营养素,水分,热损失,和微生物代谢。调查显示，ML干预主要集中在堆肥成熟度和环境污染物上，其次是营养素，水分,热损失,和微生物活动。最常用的预测模型和优化算法是人工神经网络（47％）和遗传算法（10％）。这些在有机废物堆肥的模拟和预测中证明了较高的预测准确性并最大程度地提高了堆肥效率，除了关键参数的调节。通过以更简单，更客观的方式选择堆肥成熟度和有机废物堆肥污染物残留预测中的特征变量，深度神经网络和集成学习模型可以有效地实现卓越的预测性能。
