证据理论组合模型用于三维荧光光谱水质分析

为了充分利用三维荧光光谱信息,提高分析模型对水质综合指标的预测精度及稳健性,提出了一种基于D-S（Dempster-Shafe）证据理论的模型组合方法。传统的模型组合方法主要依赖于对某个单一的预测性能评价指标进行优化来确定组合权值,没有对多个性能指标进行综合性考虑。将三维荧光光谱看作多个不同波长的激发光照射下的发射光谱,基于每个发射光谱建立子模型,将多个性能指标作为证据,确定子模型的概率分配函数,然后基于DS证据合成理论对子模型的可信度进行评价并确定组合权值。实验以40个不同来源的地表水样本为研究对象,建立组合分析模型,对总有机碳值（TOC）进行定量分析。实验证明,组合模型相对于子模型而言各个评价指标均有较大提高,充分说明了该方法的有效性。

Water quality analysis by three-dimensional fluorescence spectra based on multi-models combined by evidential theory

A multi-models was proposed to optimize combining method based on D-S （Dempster-Shafe） evidential theory the use of three-dimensional fluorescence spectral information so as to improve the precision and robustness of water quality analytic model. The combined weights in traditional combined method for forecasting are obtained with optimizing a single prediction performance index, with no considering of different aspects of the prediction model. The threedimensional fluorescence spectrum of a sample can be taken as a number of emission spectra issuing from exciting light with different wavelength. So multiple analytic sub-models were built based on emission spectra and several performance indexes were used to provide evidences then the credibility of every sub-model was evaluated based on D-S combinational weight. Forty surface water samples of variable origin （TOC） value were used as research objects and combined model was values. The experimental results showed that the combined model different performance indexes compared with the sub-models. of sub-models. And theory to determine the and total organic carbon built to predict the TOC improved appreciably at