Background: Diclofenac (DCF) is an important widely used non-steroidal antiinflammatory drug. Disposal of expired formulation, excretion from administered dose, the poor performance of sewage treatment process, contributes to its frequent detection in environment. Analysis of DCF in environmental sample requires time consuming pretreatment, extraction steps. Though, UV absorption analysis of DCF is simple but spectral interference of soil organic matter is a problem. The aim of this paper is to establish appropriate partial least square chemometric model for DCF quantitation through variable selection, and validation of analytical method through multivariate figure of merit analysis.

Methods: Spectral data of DCF spiked soil solution is recorded and variants of partial least squares (PLS) regression viz., backward-interval PLS (biPLS), synergy-interval PLS (siPLS) and genetic algorithm (GA) based PLS models (GA-PLS) are developed from autoscaled and 2nd order differential spectrum. Prediction fidelity of the selected models was evaluated from a blind-folded semi-synthetic spectral data. The method was validated through figures of merit estimates, such as selectivity, analytical sensitivity, limits of detection and quantitation.

Results: The siPLS model developed offered the minimum root mean square error of crossvalidation (RMSECV) of 0.1896 mg/l and 0.1910 mg/l for autoscaled data (9 variables) and derivative spectra (12 variables), respectively. Refinement of the derivative spectrum with GA offered a simplified model (RMSECV:0.1712, 10 variable).

Conclusion: The GA based variable selection for PLS regression analysis offers robust analytical tool for DCF in environmental samples. Further research is warranted to model variable interference in spectral data unknown to analyst in priori.

背景：双氯芬酸（DCF）是一种重要的广泛使用的非甾体类抗炎药。过期配方的处理，给药剂量的排泄，污水处理过程的不良性能，导致其在环境中的频繁检测。分析环境样品中的DCF需要耗时的预处理，提取步骤。虽然，DCF的紫外线吸收分析很简单，但是土壤有机质的光谱干扰是一个问题。本文的目的是通过变量选择建立适当的偏最小二乘方化学计量学模型用于DCF定量，并通过多元品质因数分析验证分析方法。

方法：记录DCF加标土壤溶液的光谱数据，并使用偏最小二乘（PLS）回归，后向间隔PLS（biPLS），协同间隔PLS（siPLS）和基于遗传算法（GA）的PLS模型（GA）进行变体-PLS）是根据自动缩放和二阶差分频谱开发的。所选模型的预测保真度是通过半折叠半合成光谱数据进行评估的。该方法通过优值估算值进行了验证，例如选择性，分析灵敏度，检测限和定量限。

结果：对于自动定标数据（9个变量）和导数谱（12个变量），开发的siPLS模型提供的交叉验证最小均方根误差（RMSECV）分别为0.1896 mg / l和0.1910 mg / l。用GA细化导数谱提供了简化的模型（RMSECV：0.1712，10变量）。

结论：用于PLS回归分析的基于GA的变量选择为环境样品中的DCF提供了强大的分析工具。有必要进行进一步的研究，以对先验分析人员未知的光谱数据中的可变干扰进行建模。