基于GA-BPNN模型的脱氮组合工艺中UASB工艺主导因子及最佳运行参数研究

Analysis of dominant factors and optimal operation parameters for UASB by denitrification combined technique based on GA-BPNN model

以匹配后续主体脱氮工艺为目的,采用UASB工艺进行高浓度养殖废水前期厌氧预处理。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、系统有机负荷、进水pH值、碱度、进水氨氮浓度、COD、SS 7个生态因子对UASB厌氧过程的影响。采用分割连接权值(PCW)和偏导数(PaD)两种方法定量化分析网络各层神经元的连接权值,从而明确了既定进水条件下,匹配后续脱氮工艺的UASB厌氧过程的主导因子依次为温度、碱度及系统有机负荷。最后采用遗传算法对已建立的BPNN模型寻优,确定了系统最优运行参数。结果表明,UASB系统最优运行参数为:系统反应温度55℃,进水pH值8.2,进水碱度值2 649mg/L,有机负荷1.8 kgCOD/(m^3·d),进水COD 7 000 mg/L,进水氨氮质量浓度844.3 mg/L,SS为2 983.9 mg/L。这表明高温、高COD进水、高pH值及高碱度、高SS进水、低有机负荷、低氨氮进水质量浓度有利于提高系统有机物去除率。

UASB was adopted as primary anaerobic pretreatment for high concentration breeding wastewater to serve the goal of subsequent denitrification. Based on the considerable tests performed by the au- thor at the early stage, the dynamic analysis and partial engineering applications as well as self-adaptive learning rate gradient descent al- gorithm with momentum were used for the BPNN model construction. The model was used to forecast the influence of seven ecological fac- tors including system temperature, system organic loading, influent pH, alkalinity and influent ammonia nitrogen concentration, COD as well as SS on the UASB anaerobic process. Two approaches including partitioned connection weight （PCW） and partial derivative （PaD） were used for quantitative analysis of connection weights of neurons at each layer of network. Hence, the dominant factors for UASB anaero- bic process matched by subsequent denitrification were identified as temperature, alkalinity and system organic loading according to the priority. In the end, genetic algorithm was employed to optimize the built BPNN model. The results showed that the optimal operation pa- rameters of UASB system were as follows： system temperature of 55 ℃, influent pH of 8.2, influent alkalinity of 2 649 mg/L, system organic loading of 1.8 kgCOD/（m3 · d）, influent COD of 7 000 rag/L, influent ammonia nitrogen concentration of 844.3 mg/L and SS of 2 983.9 mg/L. The research also showed that high tempera- ture, high influent COD, high pH value, high alkalinity, high influ- ent SS, low organic loading and low influent ammonia nitrogen con- centration were helpful to improve the organic removal efficiency of the system. These factors could be the basis for production control enhancement.