I 摘要 混凝处理法是最为常用的预处理、物化处理工艺之一，其在工业污水、城市污水、 自然水体中被广泛应用。自上个世纪末以来，混凝法在新型药剂到与其他工艺的联用 上都有了长足的发展。但是在实际工程运行中，加药量大、水质波动等问题对混凝法 的影响依旧没有得到很好的解决。而且混凝法的设计中存在着对诸如药剂选择、药剂 组合、去除率范围等参数的最优化信息未知的局限性，这些因素制约着混凝法设计的 准确度及应用的效果。因此，作者通过对两类不同废水的混凝处理优化调试，探究混 凝法在实际中的处理效果并分析其遇到的问题及解决途径。同时，在针对解决混凝法 设计的局限性上，作者提出了基于决策树模型的数据分析方法，通过机器学习的方式 对混凝法工艺设计中参数进行分类并预测，以此建立设计智能化的初步探究，以达到 对已有工程工艺路线以及新工程设计的优化。主要研究内容及结果如下： 对洗毛废水处理中混凝气浮进行成本优化，通过小试发现使用 Ca(OH) 2 作为 pH 调节剂，可降低 PAC 药剂量 37.1%。最终经实际调试后，药剂费用降低约 27%，出 水 COD Cr 在 122~180mg/L，SS<110mg/L，达到地方污水厂接管要求且出水稳定。 对线路板废水处理中混凝沉淀法处理含铜废水出水波动情况进行优化，通过对反 馈信息进行分析并排查，小试对 Na 2 S+PAC+PAM 的加药路线进行可行性实验并发现 其对 Cu 2+ 去除率在 97%以上。通过对其工艺改造增加沉淀时间并增设加药系统后， 最终出水 Cu 2+ <0.25mg/L，满足排放要求。 通过对两类废水的实际调试中重要的参数进行总结，以及对已刊的文献资料中参 数进行整理，根据决策树模型所需要的数据类型对数据进行预处理。结果发现当前文 献中存在着大量的参数不完整，对模型的分类能力具有一定的影响。 通过对原始数据设计的分类方法进行决策树模型分析，结果表明，决策树模型对 混凝处理工艺中的参数具有良好的分类能力且与实际优化调试结果相吻合。这表明决 策树模型在混凝工艺参数的分类预测中具有可行性，其能通过数据处理的方式来对混 凝工艺进行优化，可由此建立混凝法参数设计平台。 采用随机森林进行监督学习，发现在测试集存在着大量的误差以及较低的灵敏度 和特异性，结合模型在训练集中展示的优良的精度。其原因可能为数据量不够、混凝 工艺参数缺失、变量分类的主观模糊性等。 通过对实际工程的优化以及对文献工艺参数分析的结果可知，系统化的混凝工艺 参数大数据库具有实际价值。其建立可提升数据分析的能力，并以此建立智能化专家 系统以辅助工程的设计、调试。 关键词：混凝法，优化调试，决策树模型，可行性苏州科技大学硕士报告 Abstract II Abstract Coagulation method is one of the most commonly used pretreatment and physicochemical processes. It is widely used in industrial wastewater, municipal wastewater and natural water. Since the end of last century, coagulation method has developed greatly in the developing of new agents and combination with other processes. However, in the operation of treatment projects, the effects of high dosage, water quality fluctuation and other problems have not been solved well. Meanwhile, such limitations as reagents selection, coagulants combination and unknown parameters in optimization of removal rate in the coagulation method existing in the scope of design, these factors restrict the design of coagulation method, the accuracy and application effect. Therefore, through the optimization of the coagulation treatment of two different kinds of wastewater, the author investigates the effect of coagulation method in practice and analyzes its problems and solutions. Simultaneously, to solve the limitation of design in the coagulation process, the author puts forward the data analysis method based on Decision Tree (DT) model, through the way of Machine Learning for classifying and forecasting the designing parameters, in order to establish a preliminary study of intelligent design, meanwhile, to achieve the optimization of the existing engineering process and the design of the new project. Main research contents and results are as follows: By the cost optimization of coagulation floatation in the treatment of wool washing wastewater, through lab tests found that the using of Ca(OH) 2 as pH regulator can reduce the dosage of PAC by 37.1%. After the actual adjustment, the dosage cost was reduced by approximately 27%. The outlet COD Cr was 122-180mg/L, SS<110mg/L, reached the requirements of the local sewage treatment plant and the water quality was stable. For the optimization of treatment of circuit board wastewater and solving the Cu 2+ concentration fluctuating. After analyzing and checking the feedback information, a feasibility experiment was conducted on the combination of Na 2 S+PAC+PAM showed it could reduce the Cu 2+ by the removal rate over 97%. After increasing the precipitation time and put one more drug adding system, the coagulation sedimentation method has a good effect on all kinds of pollutants in the waste water of the circuit board, and the final outlet of Cu 2+ was less than 0.25mg/L. Based on important parameters in two kinds of practical debugging of wastewater, together with the published literature of parameters for sorting, depending on the type of苏州科技大学硕士报告 Abstract III data needed for DT model for data preprocessing. It was found that there are a large number of parameters in the current literature are incomplete, thus, it would have a certain impact to classification ability of the DT model. Based on the original data, through a designed data classification method for the DT classification analysis, the result shows that the DT for analyzing the parameters in the coagulation process has good classification ability, by such a method, the reference platform can be built for coagulation process design. It also shows the feasibility of DT model using in the classification and prediction of coagulation process parameters, It can optimize the coagulation process by means of data processing. Through supervised learning through Random Forest, it is found that there is a large amount of error and low sensitivity and specificity in the test set, and the fine precision of the model is shown in the training set. The reasons can be insufficient data, lack of parameters of coagulation process, subjective ambiguity of variable classification. Through the optimization of the actual project and the results of the analysis of the literature process parameters, it can be seen that the systematic coagulation process parameter big database has practical value. Its establishment can improve the ability for data analysis and establish the intelligent expert system to assist engineering design and debugging. Keywords: Coagulation method, Optimization of debugging, Decision Tree model, Feasibility苏州科技大学硕士报告 目录 IV 目 录 摘要 ................................................................I ABSTRACT.......................................................... II 第一章 绪论 ..........................................................1 1.1 混凝法工艺简介 .................................................... 1 1.1.1 混凝剂的发展及应用............................................ 1 1.1.2 絮凝剂的发展及应用 ............................................ 2 1.1.3 混凝工艺的研究 ................................................ 3 1.2 混凝处理的应用 .................................................... 5 1.2.1 新型混凝工艺 .................................................. 5 1.2.2 工艺的联用 .................................................... 6 1.2.3 自动化设备的进展 .............................................. 6 1.3 基于决策树的大数据分析在环境领域的应用............................ 7 1.3.1 大数据分析技术 ................................................ 7 1.3.2 决策树模型简介 ................................................ 8 1.3.3 随机森林分类器简介 ........................................... 10 1.3.4 决策树模型在环境领域的应用 ................................... 10 1.4 报告的背景、研究内容、目的及意义................................. 11 1.4.1 研究背景 ..................................................... 11 1.4.2 报告的主要内容 ............................................... 12 1.4.3 报告的研究目的和意义 ......................................... 12 第二章 研究方法.....................................................14 2.1 工程优化及问题思考 .............................................. 14 2.1.1 混凝处理工程优化调试 ......................................... 14 2.1.2 问题探讨 ..................................................... 14 2.2 文献数据收集与分析 ............................................... 14 2.2.1 数据选择及收集 ............................................... 14 2.2.2 参数预处理 ................................................... 15 2.2.3 决策树模型建立 ............................................... 15 第三章 两种废水混凝优化实例研究 ..................................17 3.1 洗毛废水混凝气浮药剂量降低优化研究............................... 17 3.1.1 工程概况 ..................................................... 17 3.1.2 处理工艺构筑物参数及尺寸 ..................................... 17苏州科技大学硕士报告 目录 V 3.1.3 处理效果及药剂成本问题 ....................................... 19 3.1.4 混凝气浮成本优化 ............................................. 19 3.2 线路板废水混凝沉淀优化研究 ....................................... 22 3.2.1 工程概况 ..................................................... 22 3.2.2 处理工艺及主要构筑物参数 ..................................... 23 3.2.3 处理效果问题及原因分析 ....................................... 25 3.2.4 问题探究及改进优化 ........................................... 26 3.2.5 优化后运行调试结果 ........................................... 28 3.3 两类废水混凝优化结果讨论 ......................................... 28 3.3.1 洗毛废水优化处理讨论 ......................................... 28 3.3.2 线路板废水优化处理讨论 ....................................... 29 3.4 本章小结 ......................................................... 30 第四章 决策树法分析结果及讨论 ....................................32 4.1 数据收集及模型预测路线 ........................................... 32 4.1.1 数据收集方法 ................................................. 32 4.1.2 设计分析路线 ................................................. 33 4.2 参数预处理 ....................................................... 33 4.3 数据分析及结果 ................................................... 35 4.3.1 决策树输出结果 ............................................... 36 4.3.2 决策树结果讨论 ............................................... 39 4.4 随机森林分析结果及讨论 ........................................... 40 4.4.1 随机森林结果 ................................................. 40 4.4.2 结果讨论 ..................................................... 43 4.5 大数据平台建立的探讨 ............................................. 44 4.5.1 当前数据获取、分析的难点 ..................................... 44 4.5.2 混凝处理工艺大数据库建立探讨 ................................. 45 4.6 本章小结 ......................................................... 46 第五章 结论与建议 ..................................................47 5.1 结论 ............................................................. 47 5.2 建议 ............................................................. 47