【摘要】 提出了一种解析分光光度同时测定数据的正交信号校正-插值-RBF(OSC-Interp-RBF)网络方法。该法将光谱阵用浓度阵正交,滤除光谱与浓度阵无关的信号,再用一维线性插值处理使训练集样本对待辩识空间形成较好的覆盖,使RBF网络能够更有效地提取信息,从而提高预测的准确性。将该算法用于模拟原料油中铁、镍、钒的测定数据解析及实际样的组分浓度预测的结果表明,对于合成样预测结果与组分实际浓度的相对误差,及对于实际样预测结果与按常规测定方法获得的结果的相对误差,绝对值均小于10%,结果令人满意。更多还原

【Abstract】 A new orthogonal signal correction-linear interpolation-RBF(radial basis function) neural networks(OSC-Interp-RBF) approach is proposed to deal with the data obtained by spectrophotometry.With this method,noises in visible spectra are eliminated by orthogonal signal correction and linear interpolation is used to make specimen of training samples cover recognizable room better.As a result,RBF can collect characteristic information more efficiently and the accuracy of the prediction is improved.The contents of Fe,Ni and V in synthetic samples and raw oil samples are determined simultaneously by the proposed method.Relative error(RE) of synthetic samples and raw oil samples are less than 10%.更多还原