机器学习强化干式厌氧消化工艺处理有机固废的工程实践

Engineering Practice of Dry Anaerobic Digestion Process for Organic Solid Waste Treatment by Machine Learning Reinforcement

摘要:

以西南地区某工程为例，提出面向厨余垃圾的机器学习强化的干式厌氧消化工艺，介绍了总体工艺技术框架、厌氧消化及脱水工艺流程、关键设计指标及主要设备规格。建立以挥发性脂肪酸与总碱度为核心的双因子状态评估模型，采用在线式最小二乘支持向量机算法构建数据驱动的状态预测模型，对高风险的调试阶段进行指标实时预测，强化干式厌氧消化工艺的抗风险能力，数据驱动模型的均方差为0.003 4。实际运行数据表明，该工艺在有机负荷、有机质降解率、沼渣与原料比方面表现良好。

Abstract:

Taking a project in the southwest region as an example, a machine learning enhanced dry anaerobic digestion process for kitchen waste is proposed. The overall technological framework, anaerobic digestion and dewatering process flow, key design indicators and main equipment specifications are introduced. A two-factor state evaluation model centered on volatile fatty acids and total alkalinity is established. An online least squares support vector machine algorithm is adopted to construct a data-driven state prediction model, which predicts the indicators in real time during the high-risk commissioning stage, thereby enhancing the risk resistance of the dry anaerobic digestion process. The mean square error of the data-driven model is 0.0034. The actual operation data show that this process performs well in terms of organic loading, organic matter degradation rate, and the ratio of biogas residue to raw material.