CAJ | 학술논문

分别采用 BP 人工神经网络算法及多元线性回归法，以实验所得的36组数据为样本，建立了以吸附时间、活性炭投加量及甲基橙废水浓度为输入变量，以活性炭吸附处理后甲基橙溶液的吸光度为输出变量的吸附预测模型，并进行了两模型预测效果的对比。结果表明，BP 神经网络模型获得了比多元线性回归更好的拟合预测效果。使用 BP 神经网络模型可以实现同时考虑三个操作因素条件下活性炭吸附特性的预测，而且预测结果与实验数据吻合度较高，其预测样本最大和最小相对偏差分别为2.92%和0.029%，残差绝对值小于0.0505。
분별채용 BP 인공신경망락산법급다원선성회귀법，이실험소득적36조수거위양본，건립료이흡부시간、활성탄투가량급갑기등폐수농도위수입변량，이활성탄흡부처리후갑기등용액적흡광도위수출변량적흡부예측모형，병진행료량모형예측효과적대비。결과표명，BP 신경망락모형획득료비다원선성회귀경호적의합예측효과。사용 BP 신경망락모형가이실현동시고필삼개조작인소조건하활성탄흡부특성적예측，이차예측결과여실험수거문합도교고，기예측양본최대화최소상대편차분별위2.92%화0.029%，잔차절대치소우0.0505。
Selecting the adsorption time, the dosage of activated carbon and the concentration of methyl orange solution as input items, the absorbance of different concentrations of methyl orange solution adsorbed by the activated carbon as output item, two models to forecast the activated carbon adsorption of methyl orange wastewater were established by BP artificial neural network algorithm and multivariate linear regression, respectively. Taking 36 groups experimental data as training and checking samples, the two models compared. The results show that the BP neural network model-predicted results are in better agreement with the experimental data than those of statistical model. The prediction of adsorption characteristics of activated carbon can be realized under considering three operation factors by using BP neural network model. The maximum and minimum relative deviations of prediction sample are 2.92% and 0.029% respectively, and the absolute value of residual is less than 0.0505.