近红外光谱-径向基神经网络在异烟肼片无损定量分析中的应用

应用异烟肼片粉末的近红外漫反射光谱数据分别结合偏最小二乘法(PLS)和径向基神经网络(RBFNN)建立定量分析模型,并用所建模型对预测集样品进行了预测,结果表明:应用RBFNN所建立的定量分析模型优于PLS模型,相关系数(r)值由0.99593提高到0.99734,交互验证均方根误差(RMSECV)值由0.00523下降到0.00423,预测均方根误差(RMSEP)值由0.00614下降到0.00501。

Non-destructive Quantitative Analysis of Isoniazid Tablets by Near Infrared Spectroscopy with Radial Basis Function Neural Network

A quantitative analysis model for the prediction of isoniazid content in the prediction set sample was set up.It was based on the combination of the data obtained from the isoniazid powder by near infrared diffuse reflectance spectrometry and the partial least square(PLS) and radial basis function neural network(RBFNN),respectively.The results showed that the model based on RBFNN was superior to that based on PLS model.The correlation coefficient was improved from 0.995 93 to 0.997 34,the root mean squares error of leave-one-cross-validation(RMSECV) was reduced from 0.005 23 to 0.004 23 and the root mean square error of prediction set(RMSEP) was reduced from 0.006 14 to 0.005 01.The relative deviation of the prediction results of isoniazid using the RBFNN model and the actual values was less than 1.012%.