脱氮过程中短程亚硝化工艺主导因子及最佳运行参数分析

ANALYSIS OF DOMINANT FACTORS AND OPTIMAL OPERATION PARAMETERS FOR NITROGEN REMOVAL BY SHORTCUT NITRIFICATION

以匹配后续厌氧氨氧化主体脱氮工艺进水为目的,采用亚硝酸型硝化工艺进行前期脱氮预处理。在前期试验及动力学分析基础上,利用带动量的自适应学习速率梯度下降算法,建立BPNN模型,预测系统温度、进水pH、碱度、进水氨氮质量浓度、曝气量5个生态因子对亚硝化过程影响。采用分割连接权值(PCW)和偏导数(PaD)2种方法,定量化分析网络各层神经元的连接权值,明确了既定进水条件下,匹配厌氧氨氧化的短程亚硝化过程主导因子依次为曝气量、温度及碱度。采用遗传算法对已建立BPNN模型寻优,结果表明系统最优运行参数为:温度28.5℃、进水pH为8.34、进水碱度值6 777 mg.L-1,进水氨氮质量浓度1 215.8 mg.L-1、曝气量0.24 m3.h-1,与实际试验具有较好一致性。同时表明加大曝气量可以一定程度上降低温度要求。

nitrous nitrification process is adopted as the primary denitrification treatment to mach＇ the subsequent denitrification of ANAMMOX. Based on large quantities of tests performed by the author at early stage, dynamic analysis and partial engineering applications, self-adaptive learning rate gradient descent algorithm with momentum is used for building BPNN model, in order to forcast the influence of five ecological factors including system temperature, influent pH, alkalinity, infiuent ammonia nitrogen concentration and aeration rate on nitrification process. Two approaches including partitioned connection weight （PCW） and partial derivative （PAD） are used for quantitative analysis of connection weights of neurons at each layer of network, in order to determine the dominant factors for shortcut nitrification process mached by anammox as aeration rate ,temperature and alkalinity according to priority. In the end, genetic algorithm optimizes built BPNN model. The results show that the optimal operation parameters of the system include temperature of 28.5 ~C, influent pH value of 8.34, influent alkalinity of 6 777 mg. L-1, influent concentration of 1 215.8 mg-L-1 and aeration rate of 0.24 m3.h-1.They are well consistent with actual test. Meanwhile, the results further show that improve aeration rate help to reduce temperature requirement.