基于改进回声状态神经网络的出水总磷软测量研究

发布时间：2018-01-28 04:57

  本文关键词： 总磷 软测量模型 回声状态网络 自适应变异粒子群算法　出处：《北京工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着社会的进步与经济的蓬勃发展,环境污染和生态恶化等问题愈发严峻,随之水污染问题越发凸显。加强水质监测,不仅关乎国民经济的发展,对于人们的身体健康也具有现实意义。水体富营养化的机理过程复杂,影响因素众多,难以得到有效控制以致现阶段发生率高,而其关键水质参数指标污水总磷(Total Phosphorus,TP)难以进行在线监测。水质监测是水体评价的前提,对于污染防治起到预警作用。近年来,基于人工神经网络的软测量模型应用广泛,能够准确建立复杂系统的模型。针对污水处理系统具有复杂动态特性多噪声、非线性的时变系统,建立基于回声状态网络的出水总磷软测量模型。由于递归神经网络能够以任意精度逼近非线性函数以及良好的动态信息处理能力,该模型能够有效模拟污水处理系统的非线性动态变化过程,实现对于出水水质参数TP的在线预测。本文主要的研究工作包括以下几点:1.提出出水TP软测量模型设计。论文中针对TP软测量模型的设计概括为以下步骤,首先数据的采集以及样本数据的预处理,然后通过主成分分析法对于TP相关相关辅助变量精选,最后建立神经网络软测量模型。本文详细出水TP软测量模型的建立过程,并证实其有效性。2.提出一种自适应变异粒子群算法。论文中针对回声状态网络在训练过程中使用伪逆算法对输出权重进行训练,难以保证回声网络的稳定性,影响网络的稳定性和预测精度。依据回声状态网络结构特点在标准粒子群算法的基础上,采用自适应变异策略,提出一种改进粒子群算法。通过对于标准测试函数进行测试,验证该算法具有搜索速度快,能够有效避免陷入局部最优中。3.基于改进回声状态网络TP软测量模型设计。结合出水TP的特点和软测量技术的研究,提出基于改进回声状态网络建立出水TP的软测量模型。通过Mackey-Glass混沌时间序列预测的预测,有效证明其应用在非线性系统的有效性,为接下来的TP软测量模型提供基础。结合之前的研究,基于改进回声状态网络软测量模型应用到出水TP预测,证明所设计的出水TP软测量模型设计的有效性。
[Abstract]:With the development of society and economy, the problems of environmental pollution and ecological deterioration become more and more serious, and the problem of water pollution becomes more prominent. Strengthening water quality monitoring is not only related to the development of national economy. The mechanism of eutrophication of water body is complex and the influence factors are many, so it is difficult to get effective control and the incidence of eutrophication is high at the present stage. However, it is difficult to carry out on-line monitoring of the key water quality parameters, total total phosphorus phosphate (TP), and water quality monitoring is the premise of water body evaluation. In recent years, the soft sensor model based on artificial neural network is widely used, which can accurately establish the model of complex system. The sewage treatment system has complex dynamic characteristics and many noises. The soft sensing model of total phosphorus in effluent based on echo state network is established for nonlinear time-varying system. The recurrent neural network can approach nonlinear function with arbitrary accuracy and has good dynamic information processing ability. The model can effectively simulate the nonlinear dynamic process of sewage treatment system. On-line prediction of effluent quality parameters TP is realized. The main research work in this paper includes the following points:. 1. The design of effluent TP soft sensor model is proposed. The design of TP soft sensor model is summarized as follows. First, the collection of data and sample data preprocessing, and then through the principal component analysis of TP related auxiliary variables selected. Finally, the soft sensing model of neural network is established, and the process of setting up the soft sensor model of effluent TP is discussed in detail in this paper. And verify its validity. 2. An adaptive mutation particle swarm optimization algorithm is proposed. In this paper, pseudo-inverse algorithm is used to train the output weight in the training process of echo state network. It is difficult to ensure the stability of echo network and affect the stability and prediction accuracy of the network. Based on the characteristics of echo state network structure and the standard particle swarm optimization algorithm, adaptive mutation strategy is adopted. An improved particle swarm optimization (PSO) algorithm is proposed, which is proved to be fast by testing the standard test function. It can effectively avoid falling into local optimum. 3. The design of TP soft sensor model based on improved echo state network, combined with the characteristics of effluent TP and the research of soft sensing technology. A soft sensing model of effluent TP based on improved echo state network is proposed. The validity of its application in nonlinear systems is proved by the prediction of Mackey-Glass chaotic time series. Based on the previous research, the improved echo state network soft sensor model is applied to TP prediction of effluent. It is proved that the designed effluent TP soft sensor model is effective.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP183;X832
，

本文编号：1469828