摘要
针对松江污水厂污水处理活性污泥系统,采用神经网络技术进行建模试验研究,在对实际运行数据剔除异常数据后,将样本数据随机分成训练样本、检验样本和测试样本。用试凑法确定合理的神经网络隐层节点数,用检验样本实时监控训练过程从而避免“过训练”现象,用多次改变网络初始连接权值求得全局极小点,从而建立了泛化能力较好的基于神经网络的活性污泥系统数学模型。利用建立的神经网络模型,对活性污泥系统运行情况的仿真与控制进行了分析研究。示例研究表明:神经网络技术能较好地应用于活性污泥系统的建模与控制,有很好的理论与实践意义。
The actual operation data of activated sludge system in Shanghai Songjiang Sewage Treatment Plant (SSSTP) were used to establish a neural network-based (NN-based) model of activated sludge system. The abnormal data were deleted according to the physical and scope principle and an effective data set was flied. The total data were divided into three parts, namely, training data set, verification (validation) data set and test data set. The proper number of neurons on hidden layer was determined by trail-and-error. The verification data set was used to monitor the training process real-timely and dynamically and to find out over-training phenomena. The global minimum of error-function was got by randomly changing the initial value of connection weights more than thirty times. The reasonable, reliable NN-based model for activated sludge system was thus set up in this paper. The simulation and operation control of activated sludge system was carried out by using the established NN-based model. The case study shows that neural network can be successfully applied to describe and control the non-linear and complicated activated sludge system. The NN-based model possesses good generalization.
出处
《环境污染治理技术与设备》
CAS
CSCD
北大核心
2006年第8期47-51,共5页
Techniques and Equipment for Environmental Pollution Control
基金
上海市教委高等学校科学技术发展基金资助项目(01H03)
关键词
活性污泥系统
神经网络
建模
仿真
控制
泛化能力
activated sludge system
neural network
modeling
simulation
control
generalization