Abstract:

The wastewater treatment process is vulnerable to the impact of external shocks to cause sludge floating, aging, poisoning, expansion and other failure conditions, resulting in effluent deterioration and high energy consumption. It is urgent to quickly and accurately identify the operating conditions of wastewater treatment process. In the existing supervised learning methods all the data are labeled which are time consuming and expensive. A multitude of unlabeled data to collect easily and cheaply have rich and useful information about the operating condition. To overcome the disadvantage of supervised learning algorithms that they cannot make use of unlabeled data, a semi-supervised extreme learning machine algorithm based on manifold regularization is adopted to monitor the operation states of biochemical wastewater treatment process. The graph Laplacian matrix is constructed from both the labeled patterns and the unlabeled patterns. Extreme learning machine algorithm is adopted to handle the semi-supervised learning task under the framework of the manifold regularization. It constructs the hidden layer using random feature mapping and solves the weights between the hidden layer and the output layer, which exhibit the computational efficiency and generalization performance of the random neural network. The results of simulation experiments show that the fault identification method based on semi supervised learning machine has superiority to the basic extreme learning machine in improving the accuracy and reliability.

Key words: wastewater treatment, extreme learning machine, semi-supervised learning, manifold regularization

摘要：

污水处理过程容易受外界冲激扰动影响,引发污泥上浮、老化、中毒、膨胀等故障工况,导致出水水质质量差,能源消耗高等问题,如何快速准确识别污水操作工况故障至关重要。针对污水工况识别过程中现有监督学习方法未利用大量未标记数据蕴含的丰富操作工况信息,采用基于流形正则化极限学习机的半监督学习方法,监视生化污水处理过程操作运行工况。该方法在学习过程中,在标记和未标记数据输入空间构建图拉普拉斯算子,通过随机特征映射建立隐含层,在流形正则化框架下,求解隐含层和输出层之间的权重,保留随机神经网络的计算效率和泛化性能。仿真实验结果表明,基于半监督极限学习机的污水处理工况识别在准确率与可靠性方面相对优于基本极限学习机方法。

关键词: 污水处理, 极限学习机, 半监督算法, 流形正则化