定量近红外光谱分析中应用自适应神经网络法建立校正模型

Application of Algorithm of Adaptive Neural Network to Construction of Calibration Model in Quantitative Near IR Spectrometric Analysis

论述了将自适应神经网络，结合内模控制，建立校正模型，应用于定量近红外光谱法测定石油化工产品中各组分的可行性。试验中，以dSPACE硬件平台为基础，以直馏柴油、加氢精制柴油和催化裂解柴油为校正模型的训练样本，对自适应神经网络校正模型作检验。试验结果表明：该方法响应速度快、误差小、鲁棒性强，在近红外长波区（800-2300nm）内，校正样品和验证样品的均方偏差均小于1×10^-6。

The feasibility of application of the algorithm of adaptive neural network, in combination with the internal model control, to the construction of calibration model in quantitative near IR-spectrometric analysis of petroleum and chemical products for their constituents was approached and discussed. In the testing, the hardware platform of dSPACE was taken as a basis, and samples of direct distilled diesel oil, hydrogenated refining diesel oil and catalytic-cracking diesel oil were taken as training samples, to test the validity of calibration model constructed by adaptive neural network. As shown by experimental results, the proposed method was proved to have quick response, small error and good robustness. In the IR spectral range of 800--2 300 nm, the values of mean square deviation of calibration samples and testing samples were found all less than 1×10^-6.