基于BP神经网络的水厂加药凝絮过程辨识研究  

Research on identification of dosing coagulation process in water plant based on BP network

作　　者：郭佩佩[1] 甘艳珍[1] 朱学峰[1] 

机构地区：[1]华南理工大学自动化科学与工程学院,广东广州510641

出　　处：《计算机与应用化学》2008年第7期889-892,共4页Computers and Applied Chemistry

基　　金：广东省科技计划项目(2005B10201005)

摘　　要：城市供水中的加药凝絮过程是一个大惯性、大时滞、非线性、时变以及随机干扰多的难控过程。从生产数据来对该过程建模,并采用先进的控制策略对其进行有效的控制,一直是控制技术人员的追求目标。本论文以某水厂的实际生产数据为基础,采用动态BP神经网络的建模方法,通过辨识步骤,确定了动态BP网络结构,得到了具有较好拟合与泛化能力的神经网络模型。在已获取的神经网络模型上,进行了以待滤水浊度为输出,投矾量为输入的阶跃响应实验,以此取得了投药凝絮过程的一阶惯性加时滞模型。该模型的获取为今后采用先进的控制策略对加药凝絮过程进行高级控制打下了良好的基础。Dosing coagulation in water plant is a large time constant, big time delay, non-linear and time-varying process with many disturbances. Thus it is very difficult to be controlled. It is the expected goal for many researchers and engineers to model the process with industrial data and to control it effectively with modern control techniques. In this paper, the process has been identified by a dynamic BP network based on the acquired data. Firstly, the structure of dynamic BP network was determined via the industrial data of the water plant. The simulations had shown that the obtained ANN model had good fitting and predicting performances. Then from the ANN model the step response experiments were carried out in which the input and output variables were the dosing amount and the turbidity of water before filtering respectively. The obtained model from the step response is of first order plus time delay. Without doubt that will provide the good foundation for controlling the dosing coagulation process with advanced control strategies in the future.

关 键 词：加药凝絮过程 系统辨识 动态BP网络 阶跃响应 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]