Abstract:

A new method for the nondestructive quantitative prediction of flunarizine hydrochloride powder drug based on shortwave nearinfrared (SWNIR) diffuse reflectance spectral data. With the analytical data of high performance liquid chromatography (HPLC) taken as the reference value, a quantitative analysis model for describing the relationship between the spectral characteristic and component was established via combining genetic algorithm (GA) with neural networks. The neural network topological framework was constructed automatically by means of GA. The optimal network parameters and spectral region were selected. Moreover, the influences of different spectral preprocessing methods to the prediction ability of neural network were discussed. The mass ratio of the test sample was predicted. The results of the experiment show that the firstderivative spectrum is the optimal pretreatment method of modeling, by which the rootmeansquareerrors (RMSE) for test set was 0.198 7%, and the correlation coefficients R for test set was 0.986 3.

Key words: shortwave nearinfrared spectroscopy, diffuse reflectance, neural networks, genetic algorithm, flunarizine hydrochloride