【作者】 李伟奖；

【导师】 马邕文；

【作者基本信息】 华南理工大学 ， 环境工程， 2010， 硕士

【摘要】 随着工业的不断发展,工业废水的污染日益严重,治理污水的任务变得愈加繁重。造纸行业是污染我国环境的主要行业之一,解决造纸工业的水污染,成为了迫在眉睫的任务。而目前造纸废水的处理工艺主要采用“物化-生化”工艺处理,生化处理部分大多采用的是活性污泥法工艺处理的。由于活性污泥法处理造纸废水具有多变量、非线性、大滞后、不确定性和复杂性的特点,单纯人工的操作难以实现废水处理稳定性和良好的出水效果。传统的自动控制技术自身存在缺陷,建立的数学模型不够精确,控制效果不够理想,导致污水处理过程控制自动化水平相对落后。鉴于此,本论文致力于研究用模糊神经网络算法实现对废纸造纸废水生化处理系统曝气量的智能控制,得到以下成果:根据废纸造纸废水的特征,确定了实验室条件下的造纸废水处理工艺,完成了MCGS组态软件的组态和PLC程序的设计,在此基础上建立了造纸废水处理的自动控制系统,并在此自动控制系统下进行试验,获取大量的样本数据。再根据对废纸造纸废水处理特征和模糊神经网络的分析,提出了智能预测和控制模型方案。利用获取的样本数据,采用Takagi-Sugeno推理的网络建立A/O生化池出水COD预测模型,结合模糊C均值聚类和混合算法完成网络的结构辨识和参数辨识,借助MATLAB软件对建立的预测模型进行仿真分析,仿真结果表明,预测模型具有很好的学习能力和泛化能力,训练数据的相对误差绝对值范围为0~0.0073%,测试数据的相对误差绝对值范围为0~10.4636%。通过建立的Takagi-Sugeno预测模型与单纯的BP网络的预测模型进行性能比较,前者的训练数据和测试数据的平均绝对误差率MAPE分别为1.486×10-3 %、2.703%,而后者的训练数据和测试数据的平均绝对误差率MAPE分别为2.785%、11.53%。通过两者的性能对比可知,Takagi-Sugeno预测模型的平均绝对误差率MAPE远远小于单纯的BP网络的预测模型,说明了前者更适合用于预测生化池的出水COD值。采用基于Mamdani规则的模糊神经网络建立曝气量控制模型,利用出水COD值的变化量和出水COD值的变化率作为控制模型的输入变量,曝气量的修正量作为控制模型输出,用于调节曝气量。结合混合算法完成网络的结构辨识和参数辨识,确定控制模型的具体参数。在MCGS高级开发包生成的VB程序中编写了模糊神经算法,按照MCGS的接口函数规范将算法嵌入到MCGS中,实现了对造纸废水处理系统的智能控制。在实验室条件下进行验证试验,将期待出水COD值设定在100mg/L,结果表明在不同进水负荷下,生化池出水COD值在91~109mg/L范围,基本维持在100mg/L左右,说明基于模糊神经网络算法的智能控制系统能够有效地实现造纸废水处理系统的智能控制。本课题的研究成果为实现废纸造纸废水生化处理系统的自动控制提供了一种有效的解决方案,同时为污水处理过程实现优化控制提供了新的途径,具有一定的推广价值。更多还原

【Abstract】 As the constant development of the industry, the pollution of industrial wastewater becomes more and more serious. Sewage treatment has become an arduous task. The papermaking industry is one of the major pollutive industries in China, so solving the pollution of papermaking wastewater has become an urgent task. At present the treatment method used in paper wastewater treatment are mainly physicochemical and biochemical technology, in which activated sludge process is mostly adopted. In the wastewater process there are many characteristics, such as multi-variables, nonlinearity, hysteresis, uncertainty and complexity, etc, all of which cause poor stability and bad effluent quality. However, the traditional control owing to its shorting, is difficult to build precise mathematical model,so the control effect is not satisfactory. Therefore, in this paper the fuzzy neural network intelligent control system for aeration in wastewater treatment system was presented. And several valuable conclusions were reached.According to the characteristics of papemaking wastewater, the automatic control system based on Windows CE.NET OS and MCGS software of wastewater was constructed, and the design method and the hybrid intelligent structure were proposed with the consideration of Fuzzy neural network control. The predictive model of effluent COD value is based on the Takagi-Sugeno inferential network. In order to improve the network performance, fuzzy C-means clustering was used to identify model’s architecture and optimize fuzzy rule. The simulation indicated that the predictive model had good ability both in learning and generalization, with relative errors of training and test data are 0~0.0073% and 0~10.4636%, respectively.Camparing the predictive modle based on Takagi-Sugeno with the predictive modle based on BP network in performance, when training, MAPE (Mean Absolute Percentage Error)between the predicted and observed values was1.486×10-3 % using Takagi-Sugeno model, and it was 2.703% when testing But MAPE of the training and testing usingBP modle were 2.785% and 11.53%, repectively. Mape using the Takagi-Sugeno predictive model was lower than that using ANN. Therefore, the Takagi-Sugeno modle is more suitable to predict the effluent COD value.The control model was built up based on Mamdani network. The change and change rate of COD in effluent were considered as the input variables of control model, and correction of aeration was considered as the output variable, which was adopted for regulating aeration. Fuzzy neural network algorithm with MCGS development package using VB program was developed,and embeded it into MCGS according to MCGS interface function criterion, so that intelligent control system for papermaking wastewater treatmen was achived. And then The validation experiment was finally given in laboratory. The expectation COD value of effluent was setted to 100 mg/L. The results show that under different influent laod, the COD value of effluent was between 91 to 109 mg/L, remaining at about 100 mg/L. It proved that the intelligent control system based on fuzzy neural network was effective.The reasearch can provide an effective way to achive autocontrol for wastewater biochemical treatment system from papermaking and can provide guidance for the further study of intelligent control in the field of wastewater treatment and the popularization of wastewater treatment project with intelligent control.更多还原