摘要
针对污水处理过程出水水质参数难以在线监测的问题,提出了一种多模型在线软测量方法。该方法根据污水处理过程中实时工况数据具有聚类和迁移属性的特点,利用在线减法聚类算法将实时工况数据样本进行划分,并根据实时工况数据在样本空间中的分布,采用模糊策略将相应子空间的实时工况数据分配给不同的子模型进行学习,最后通过动态集成各子模型的输出而得到最终软测量结果。以某污水处理厂实际运行数据对污水处理过程出水水质氨氮进行预测,实验结果表明,该方法确实能够以实时工况数据为驱动自组织构建多模型软测量模型,且用该方法构建出的多模型软测量模型无论在建模精度、建模速度以及跟踪实时工况的能力等方面都有所提高。
Considering the difficulty that the parameters of the water quality in wastewater treatment process is immeasurable, an online multi-model softsensing method is proposed. Based on the facts that the condition data of the wastewater treatment process exhibit the clustering and regime shifting properties, an online substractive clustering algorithm is used to partition the sample sapce. Then, a fuzzy strategy is applied to assign the corresonding sub-sample space learning data to different sub-neural network for learning. Finally, the output of each sub-neural network is dynamiclly integrated as the final softsensing result. The proposed method is applied to the predic- tion of annmonia nitrogen content of the effluent from a wastewater treatment plant. The experiment results shows that a multi-model soft- sensing can really bulid based on the real-time condition datas, and the modeling accuracy, learning speed and the track real-time con- ditions ability of this muhi-model are improved.
出处
《控制工程》
CSCD
北大核心
2014年第1期88-93,共6页
Control Engineering of China
基金
国家自然科学基金(60971048)
辽宁省教育厅科学研究一般项目(L2013129)
关键词
污水处理过程
多模型
软测量
在线减法聚类
wastewater treatment process
multi-model
softsensing
online subcluster
