论文部分内容阅读
为实现高效短程生物脱氮及氨氮和亚硝酸盐氮的快速检测,采用主成分分析结合BP神经网络的方法建立短程生物脱氮工艺中氨氮和亚硝酸盐氮的近红外光谱定量分析模型(BP神经网络模型)。工艺运行结果表明:原水经过好氧阶段氨氮从45.3mg·L-1下降到2.7mg·L-1,亚硝酸盐氮从0.01mg·L-1上升到19.6mg·L-1,硝酸盐氮受到抑制;在缺氧段亚硝酸盐氮从19.6mg·L-1下降至1.2mg·L-1,系统实现了良好的短程生物脱氮效果。水样原始光谱主成分分析表明:前13个主成分代表了原始光谱数据的信息,其累计贡献率达到95.04%,排除了冗余信息且大大降低了模型的维数,光谱数据矩阵从192×2 203减少到192×13,大大降低了运算量并提高了模型的精度。BP神经网络模型校正结果显示:BP神经网络模型对氨氮、亚硝酸盐氮校正时的决定系数(R2)分别达到0.950 4和0.976 2,校正均方根误差(RMSECV)分别为0.016 6和0.010 9。BP神经网络模型预测结果显示:BP神经网络模型对氨氮、亚硝酸盐氮预测输出与期望输出之间的决定系数(R2)分别为0.974 0和0.981 4,预测均方根误差(RMSEP)分别为0.033 7和0.028 7,模型预测效果良好。研究表明,BP神经网络模型可以通过快速测定水样的近红外光谱数据预测短程生物脱氮工艺中氨氮和亚硝酸盐氮浓度,并根据氨氮和亚硝酸盐氮浓度变化及时、灵活地控制工艺的运行,为生物脱氮提供快速有效的检测技术和科学依据。
In order to achieve high efficiency short-range biological denitrification and rapid detection of ammonia nitrogen and nitrite nitrogen, a principal component analysis combined with BP neural network was used to establish a quantitative analysis model of near-infrared spectrum of ammonia nitrogen and nitrite nitrogen in short-range biological nitrogen removal process Neural network model). The results of process operation showed that the ammonia nitrogen concentration decreased from 45.3mg · L-1 to 2.7mg · L-1 and the nitrite nitrogen increased from 0.01mg · L-1 to 19.6mg · L-1 during the aerobic phase. Nitrate nitrogen The nitrite nitrogen decreased from 19.6mg · L-1 to 1.2mg · L-1 in the anoxic stage, and the system achieved a good effect of denitrification. The principal component analysis of the original spectrum of water samples shows that the first 13 principal components represent the original spectral data, and the cumulative contribution rate reaches 95.04%, eliminating the redundant information and greatly reducing the dimension of the model. The spectral data matrix changes from 192 × 2 203 reduced to 192 × 13, greatly reducing the amount of computation and improve the accuracy of the model. The results of BP neural network model calibration showed that the coefficients of determination (R2) of the BP neural network model for the correction of ammonia nitrogen and nitrite nitrogen reached 0.950 4 and 0.976 2 respectively, and the RMSECV of the BP neural network model were 0.016 6 and 0.010 9 . The results of BP neural network model prediction showed that the coefficients of determination (R2) of BP neural network model for prediction output and expected output of ammonia nitrogen and nitrite nitrogen were 0.974 0 and 0.981 4, respectively. The root mean square error of prediction (RMSEP) 0.033 7 and 0.028 7, the model predicted good effect. The results show that the BP neural network model can quickly determine the concentration of ammonia nitrogen and nitrite nitrogen in short-range biological nitrogen removal process by rapid determination of water near-infrared spectroscopy data and control the process flexibly according to the change of ammonia nitrogen and nitrite nitrogen concentration Operation, provide fast and effective detection technology and scientific basis for biological nitrogen removal.