活性污泥系统神经网络建模与仿真研究

采用神经网络技术对松江污水厂污水处理活性污泥系统进行建模试验研究，在对实际运行数据按机理准则和范围准则剔除异常数据后，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，以避免采用过大或过小的网络结构，在训练过程中用检验样本实时监控从而避免“过训练”现象的影响，较好地解决神经网络方法建模的两大难题，从而建立可靠、有效的活性污泥系统神经网络模型。并应用建立的网络模型对活性污泥系统的运行情况进行了仿真研究。建模研究表明，神经网络技术能较好地应用于活性污泥系统的建模，模型具有较好的泛化能力，有很好的实用价值。

Neural network for modeling and simulation of activated sludge system

The actual operation data of the activated sludge system of Shanghai Songjiang Sewage Treatment Plant were used to establish a neural network-based (NN-based) model of the activated sludge system. After deleting abnormal data, the remaining data were divided into three sets: training data, verification (validation) data and test data. The reasonable neuron number on hidden layer was determined by trail-and-error. The verification data set was used to monitor the training process and to avoid the over-trained phenomena. A reliable NN-based model for activated sludge system was developed and was successfully employed for performance simulation of the non-linear and complicated activated sludge system. The NN-based activated sludge model is an attractive simulation tool.