基于动力学和PSO-SVM的废水厌氧处理产气量的混合软测量模型

Hybrid Model of Measuring Biogas Yield in Anaerobic Digestion Process Based on Incorporated Bio-Kinetic Model with Support Vector Machine Model

关键词:  厌氧消化  产气量  动力学模型  粒子群算法  支持向量机

Key Words:anaerobic digestion  biogas flow rate  kinetic model  Support Vector Machine model  particle swarm optimization

摘要:在实验室搭建了一套基于IC厌氧反应器的废水厌氧处理系统，自制有机废水（以葡萄糖、尿素、磷酸二氢钾按COD∶N∶P=200∶5∶1的比例配制，同时加入微量元素）进行实验，该系统运行2个月，采集159组运行数据作为元数据集，以进水有机负荷、反应器温度、反应器pH值、氧化还原电位、体系积累的乙酸和进水碱度为输入量，以产气量为输出量，建立PSO（粒子群算法）-SVM（支持向量机）传统模型。为提升模型预测精度，在传统模型基础上，将反应器温度、反应器pH值、体系积累的乙酸进行动力学模型量化后建立混合模型。仿真结果表明，PSO-SVM模型对预测废水厌氧处理体系产气量表现较好，测试集的预测数据与实际数据的相关系数为86.71%，引入动力学模型后的混合模型在产气量预测中的精度提升较大，线性相关性R由86.71%提升至95.73%，可为监控、优化和理解厌氧消化过程提供指导。

Abstract:Lack of AD process control and analysis is believed to be one of the main limitations for effective organic matter degradation. Biogas flow rate and component as commonly monitoring indicators indicate the overall process performance. The objective of this work was to implement a strategy to simultaneously monitor and predict the biogas flow rate using a hybrid model, which combined kinetic model and a traditional Support Vector Machine model (SVM) optimized by particle swarm optimization algorithm (PSO). For the training and verification of the models, a data set with 159 samples was used, which were obtained using a lab-scale AD reactor system. The results demonstrated that the hybrid model had a satisfying predicting performance. The R value of the traditional model was 86.71%. And compared with traditional model, the performance of the hybrid model was improved significantly the R value of the hybrid model was 95.73%. Furthermore, the hybrid model gave a successful window, which was a good reference for the modeling study of AD process.