中文摘要：

污水处理过程复杂多变的运行工况以及系统脆弱的抗负荷冲击能力,常常导致污水处理厂运行目标难以实现,有效识别污水操作工况的变化对污水处理过程安全运行和操作优化十分重要。为增强未知样本分类可靠性,在概率极限学习机二分类基础上,将其扩展到多分类概率极限学习机方法 （extreme learning machine）。该方法首先采用极限学习机建立污水处理过程实时变量和污水处理过程工况编码之间的预报模型,然后根据类别的输出预报值分别建立每个类训练样本潜在函数的均值,确定所有类的条件概率密度函数,非线性最小二乘辨识条件概率密度函数参数,最后根据贝叶斯原理计算所有类的后验概率,由后验概率最大值判别样本所属类别。以辽宁某城市污水处理厂实时数据为背景进行验证,实验结果表明多分类概率极限学习机分类的可靠性和准确性优于极限学习机分类方法。

英文摘要：

Due to the complex varying operational conditions and weak stability to resist the impact loads of influent flow rate and quality,it is difficult to satisfy operational objective of wastewater treatment plants.Therefore,the effective identification of the operational conditions for the complex wastewater treatment processes becomes one of the most important issues due to the potential advantages to be gained from reduced costs,improved productivity and increased production quality.A multi-classification probabilistic extreme learning machine（ELM）was proposed to enhance the reliability of classification by using the combination of extreme learning machine and the Bayesian decision theory.Extreme learning machine was used to build the output coding model between the real time process variables and the operational condition coding in the wastewater treatment process.ELM prediction values fluctuated around the class encoding following a normal Gaussian distribution.A potential function was calculated for each training sample based on the density methods.The potential functions of the training samples for each class were averaged to obtain the probability density function of each class.Parameters of probability density function for each class were estimated by the nonlinear least squares method.It could decrease the misclassification to classify the unknown samples due to the uncertainty of ELM predictions.The proposed method was verified with the data from a small scale industrial wastewater treatment plant,located in Liaoning province,China.Experimental results showed that the proposed method had a relatively better performance than ELM on classification accuracy.