基于GMM-ICSA的光谱重叠峰分解研究

针对X射线荧光光谱法(XRF)中Pb元素和As元素的重叠峰导致的元素浓度建模困难、乌鸦搜索算法(CSA)鲁棒性较弱等问题,将高斯混合模型(GMM)与改进型乌鸦算法(ICSA)相结合来实现重叠峰的分解。ICSA相比于CSA的改进主要有:引入了“乌鸦反哺”的特性,将GMM与ICSA更好的衔接在一起;将固定的意识概率更改为梯度型,使种群迭代更有多样性;适当调整了全局的优化策略,使算法更稳定收敛更快。将GMM模型与GMM-ICSA模型作对比实验,可得优化后的模型分解精度提升了4.93%;同时以均方误差和迭代时间为衡量搜索算法的标准,将ICSA与同类型的4种算法做了对比实验得出:CSA的均方误差和迭代时间均优于其他算法,从而表明改进后的算法能处理重叠峰问题的可行性。

Research on Decomposition of Spectral Overlapping Peaks Based on GMM-ICSA

Aiming at the difficulties in modeling the element concentration caused by the overlapping peaks of Pb and As in X-ray fluorescence spectroscopy (XRF) and the weak robustness of the crow search algorithm (CSA), the Gaussian mixture model (GMM) is combined with the improved the crow search algorithm (ICSA) to achieve the decomposition of overlapping peaks. Compared with CSA, ICSA has three main improvements: the introduction of the “crow feeding” feature to better connect GMM and ICSA; the fixed probability of awareness is changed to a gradient type to make the population iteration more diverse; appropriate adjustments the global optimization strategy makes the algorithm more stable and converges faster. The comparison experiment between the GMM model and the GMM-ICSA model shows that the decomposition accuracy of the optimized model is increased by 4.93%. At the same time, the mean square error and iteration time are used to measure the quality of the search algorithm, and ICSA is compared with the same type comparative experiments on the four algorithms show that the mean square error and iteration time of ICSA are better than other algorithms, which shows the feasibility of the improved algorithm in dealing with the problem of overlapping peaks.