Opt Express. 2020 May 11;28(10):14198-14208. doi: 10.1364/OE.387858.
Laser-induced breakdown spectroscopy, coupled with advanced chemometric methods, was used to quantitate multiple elements in a seaweed-based fertilizer. The influence of important parameters was determined using partial least squares regression (PLSR), support vector regression (SVR) and random forest (RF) optimizations. Optimal results for Mg, K and P were obtained using PLSR, whereas RF yielded the best results for Mn, Cu, Sr and Ca. The best predictions for Ba levels were obtained with SVR. The lowest root mean square errors in the prediction sets for Mn, Cu, Sr, Ba, Mg, K, P and Ca were 48.27 µg/g, 36.90 µg/g, 0.37 mg/g, 40.32 µg/g, 1.99 mg/g, 2.03 mg/g, 4.81 mg/g and 14.08 mg/g, respectively, with average relative standard deviations of 13.65%, 2.68%, 19.80%, 5.17%, 3.32%, 2.98%, 1.82% and 5.81%. The results showed that the optimal multivariate model depended on the specific element being analyzed. The proposed method provides a rapid means of determining multielement concentrations in seaweed-based fertilizers.
激光诱导击穿光谱结合先进的化学计量学方法被用于定量分析海藻肥中的多种元素。通过偏最小二乘回归（PLSR）、支持向量回归（SVR）和随机森林（RF）优化确定了重要参数的影响。对于 Mg、K 和 P，PLSR 得到了最佳结果，而对于 Mn、Cu、Sr 和 Ca，RF 则得到了最佳结果。对于 Ba 水平，SVR 得到了最佳预测结果。Mn、Cu、Sr、Ba、Mg、K、P 和 Ca 的预测集的最低均方根误差分别为 48.27μg/g、36.90μg/g、0.37mg/g、40.32μg/g、1.99mg/g、2.03mg/g、4.81mg/g 和 14.08mg/g，平均相对标准偏差分别为 13.65%、2.68%、19.80%、5.17%、3.32%、2.98%、1.82%和 5.81%。结果表明，最优多元模型取决于分析的特定元素。该方法为快速测定海藻肥中的多种元素浓度提供了一种手段。
