基于神经网络的传爆药废水COD去除率预测研究  

Elman Model in Prediction of COD Removal Rate of Booster Explosive Wastewater

作　　者：刘玉存[1] 于国强[1] 王少华[2] 常双君[1] 

机构地区：[1]中北大学化工与环境学院,山西太原030051 [2]中国北方车辆研究所,北京100072

出　　处：《含能材料》2009年第3期361-364,共4页Chinese Journal of Energetic Materials

基　　金：山西省自然科学基金项目(20070113034);教育部科学技术研究重点项目(204020)

摘　　要：为预测超临界水氧化法处理二硝基重氮酚生产废水的COD(chemicalo xygen demand)去除率,采用HXDK-01-A间歇式超临界水氧化实验装置处理实际工业生产废水,主要考察反应温度、反应压力、停留时间和过氧量对COD去除率的影响。采用实验数据,以反应温度、反应压力、停留时间和过氧量为网络输入,COD去除率为网络输出,以Matlab为平台建立了Elman神经网络预测模型。神经网络模型预测的均方差为0.0418,单个最大误差为-0.3231,最小误差为0.0296;多元回归分析拟合数据的均方差为0.3149,单个最大误差为0.8830,最小误差为0.2200,神经网络预测结果明显优于多元回归分析结果。说明采用神经网络模型预测超临界水氧化法的废水COD去除率是可行的。In order to predict the chemical oxygen demand （COD） removal rate of the diazodinitrophenol （DDNP） wastewater treated by supercritical water oxidation （SCWO） , the HXDK-01-A intermittence type supercritieal water oxidation deviee was used to dispose the actual industrial production wastewater, and the effects of reaction temperature, reaction pressure, residence time, oxygen excess on COD removal rate were studied. A single hidden layer Elman prediction model was established by using the reaction temperature, reaction pressure, residence time, oxygen excess as input variables ,and using the COD removal rate as output. The MSE of the Elman model is 0.0418, the biggest error is -0.3231, and the least error is 0. 0296, the MSE of the multiple regression is 0. 3149, the biggest error is 0. 8830, and the least error is 0. 2200. The Elman neural network prediction results are better than that of multiple regression analysis. Results show that the Elman model can be adopted to predict the COD removal rate of the wastewater treated by SCWO.

关 键 词：环境科学 超临界水氧化 二硝基重氮酚(DDNP) ELMAN神经网络 废水处理 

分 类 号：X789[环境科学与工程—环境工程]