摘要
提出了一套基于神经网络分类器的城市污水处理厂水力负荷冲击预警系统,以期对进水水量骤增现象进行提前1天的预报,使污水处理厂可根据预报结果提前采取水力冲击防护措施,从而保证各单元的平稳运行.根据进水水量的涨幅将某污水处理厂12年日进水水量监测数据分为“常规”和“冲击”两类,重点对“冲击”数据进行提前1天的预测,并采用冲击漏报率、冲击误报率和报准率对模型的预测精度进行评价;同时,基于同样的建模方法和不同的训练、验证样本建立了N（1）、N（2）和N（3）3个平行模型,以对模型的鲁棒性和建模方法的可重复性进行考察.结果显示,3个模型对2010年、2011年和2012年3年测试样本的预测效果良好,冲击漏报率和报准率两项指标数值均较为稳定,分别在0-0.167和0.981-0.995之间浮动,冲击误报率虽然在数值上的浮动较大,最低为0.143,最高为0.500,平均为0.310,但仍在工程上的可承受范围内.该结果表明,本研究基于神经网络分类器所建立的3个神经网络模型预测精度高、鲁棒性好,显示出良好的性能,有望为污水处理厂水力冲击防护工作提供有力参考.
An early-warning system for hydraulic shock loads in wastewater treatment plants （WWTPs） was developed based on artificial neural network （ANN） classifiers. The system can predict daily inflows one-day-ahead so that rapid increasing hydraufic shock loads due to sudden growth of inflows can be avoided. The daily inflow data of a WWTP in the past 12 years were classified into ＇regular＇ and ＇shock＇ loads, which were used for training the developed ANN model. To evaluate performance of the model, three evaluation functions （ ＇ false negative rates of shock loads＇ , ＇ false positive rates of shock loads＇ and ＇ accuracy rates＇ ） were used. In addition, to investigate robustness of the model and make the modeling method repeatable, three ANNs （ N （ 1 ）, N （2） and N （ 3 ） ） were developed based on the same modeling methods but different training and validation datasets. The results showed good prediction performance of the three ANN models for the validation dataset （from 2010 to 2012）. The false negative rates of shock loads and the accuracy rates were 0 - 0.167 and 0.981 -0.995, respectively. Although the false positive rates of shock loads varied significantly from 0.143 to 0.500, the variation is still within an acceptable range for engineering applications.
出处
《环境科学学报》
CAS
CSCD
北大核心
2015年第4期1224-1232,共9页
Acta Scientiae Circumstantiae
基金
国家自然科学基金(No.51208424)
西北大学"十二五""211工程"创新人才培养项目(No.YZZ13004)~~
关键词
人工神经网络
水力冲击负荷
城市污水处理厂
日进水量
时间序列
artificial neural networks
hydraulic shock loads
municipal wastewater treatment plants
daily flow
time series
