印染废水的神经网络法优化絮凝处理

Optimizing Preparation of New Flocculants for Dyeing Wastewater Treatment by Artificial Neural Network

为获得新型絮凝剂的最佳制备条件,采用人工神经网络(ANN)结合加速遗传算法(AGA)对絮凝剂性能影响因素进行全局寻优。在影响絮凝剂处理性能的主要因素Fe、Al与Si的摩尔比n(Fe+Al):n(Si),B、Mg和Si的摩尔比n(B+Mg):n(Si)以及熟化时间的有效作用范围内,用Box-Behnken Design(BBD)实验设计方法,产生15组影响因素组合作为输入样本,经对印染废水的絮凝处理实验得到相应的COD和色度去除率输出样本,由神经网络方法对15组输入、输出样本数据建模,得到反映絮凝剂制备条件和絮凝剂去除效果的响应关系,并采用加速遗传算法全局优化,得到最大COD去除率和色度去除率条件下的最优絮凝剂制备条件组合,即n(Fe+Al):n(Si)=5.08,n(B+Mg):n(Si)=0.55,熟化时间为2.1 d,对应的印染废水COD去除率为88.10%,色度去除率为95.37%。模型验证实验显示,实验值与模型预测值的最大相对误差不超过5%。相对于响应曲面法,用神经网络结合加速遗传算法优化得到的新型絮凝剂去除印染废水中COD和色度的能力更高,进一步验证了神经网络法在新型絮凝剂制备条件优化中的有效性。

To develop a new kind of flocculants, the artificial neural network （ANN） model and accelerating genetic algorism （AGA） method were used to optimize the preparation of flocculants at the laboratory. COD and colority were the parameters of the dyeing wastewater being treated with one of the flocculants, while the main parameters affecting the performances of the flocculant were the mole ratios i.e,, n （Fe＋Al）：n（Si）and n（B＋Mg）：n（Si） ,as well as the maturation time. By using Box-Behnken Design（BBD） method, 15 sets of influencing parameters were obtained as the input-output training samples, ANN modeling and optimization resulted in a set of optimized parameters, by which the best results for treatment of the dyeing wastewater were obtained, i.e., COD removal was 88.10%, colority removal 95.37%. In addition, the validation experiment showed that relative errors between the test data and the predicted data of the model were within 5%.