CAJ | 학술논문

以某化工企业MTBE装置的过程参数C4流量F1、CH3OH流量F2、混合原料预热温度T301为变量，用BP神经网络建立了关于醚化塔下部温度T303D的预测模型．经Pauta准则剔除样本数据异常点后建立的BP模型，训练、验证和预测计算结果的确定系数R^2分别为0．8873、0．8873和0．8582，而直接用原始数据建立的BP模型，训练、验证和预测计算结果的R^2则分别为0．8361、0．8148和0．7376．研究表明，运用Pauta准则剔除异常样本数据，可以较大幅度的提高模型的预测准确性．
이모화공기업MTBE장치적과정삼수C4류량F1、CH3OH류량F2、혼합원료예열온도T301위변량，용BP신경망락건립료관우미화탑하부온도T303D적예측모형．경Pauta준칙척제양본수거이상점후건립적BP모형，훈련、험증화예측계산결과적학정계수R^2분별위0．8873、0．8873화0．8582，이직접용원시수거건립적BP모형，훈련、험증화예측계산결과적R^2칙분별위0．8361、0．8148화0．7376．연구표명，운용Pauta준칙척제이상양본수거，가이교대폭도적제고모형적예측준학성．
With process parameters of a chemical enterprise MTBE device C4 flow F1, CH3 OH flow F2, mixing the ingredients preheating temperature T301 as variables, we have established an etherification tower temperature BPNN（Back Propagation Neural Network） prediction model about the T303D. The BP model was established after eliminating the abnormal values of the sample data by Pauta Criteria. The correlation coefficients（ R^2） of the training, validation and prediction were 0. 8873,0. 8873 and 0. 8582, respectively. And with the BP model established directly by original sample data, the R^2 calculation results of the training, validation and prediction were 0. 8361, 0. 8148 and 0. 7376, respectively. The research has shown that with the PaSta Criteria to eliminate the abnormal values of the sample data, the established BP model can improve the prediction accuracy significantly.