| 运用神经网络分析电流测定堆肥系统中的对苯二酚 |
| 摘要点击 3314 全文点击 2505 投稿时间：2007-09-27 修订日期：2007-12-10 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 神经网络 生物传感器 对苯二酚 堆肥 回归模型 |
| 英文关键词 artificial neural network biosensor hydroquinone compost regression model |
| DOI 10.13227/j.hjkx.20080950 |
|
| 中文摘要 |
| 针对生物传感器检测堆肥系统中对苯二酚时存在的检测范围较小和检测精度不高等问题，在分析漆酶生物传感器检测电流信号的基础上，以传感器响应电流为输入变量，以对苯二酚浓度为输出变量，建立了神经网络（ANN）模型，对网络进行了优化，确定了最佳的网络参数（最佳隐含层神经元数为7，最佳输入层神经元数为5，最优的传递函数组合为Tansig-Logsig，最优的算法为Levenberg-Marquardt算法），并将模型预测性能同非线性回归（NR）模型进行了比较.NR的RMSE（均方根误差）为14.441?9 μmol/L ，预测值与实测浓度之间的相关系数为0.994?4 ；与之相比，ANN的RMSE达到4.444?2 μmol/L ，预测值与实测浓度的相关系数为0.999?6 ，预测性能明显优于NR.对苯二酚的检测范围达到0.015～450 μmol/L ，在现有水平上进一步扩大了检测范围同时提高了检测精度. |
| 英文摘要 |
| To resolve the problems of small detection concentration scope and poor precision in amperometric determination of hydroquinone in compost system by laccase biosensor, an artificial neural network (ANN) model was established to analyze the current signal and predict hydroquinone concentration. The response currents and hydroquinone concentration were taken as input variables and output variable respectively. The optimal configuration of networks was obtained: hidden neuron number, 7; input neuron number, 5; transfer function, Tansig-Logsig; algorithm, Levenberg-Marquardt. The prediction performances of ANN were compared with nonlinear regression model (NR). The RMSE (root mean square of error) and correlation coefficient between calculated values and measured values of NR are 14.441?9 μmol/L and 0.994?4 respectively, while those of ANN reach 4.444?2 μmol/L and 0.999?6 , respectively, showing better prediction performances than NR. Meanwhile, the hydroquinone detection scope of 0.015-450 μmol/L is achieved. Both the detection scope and precision of hydroquinone in compost system were further improved on the present level. |
