Fast quality control of natural gas for commercial supply and transport utilities. (1st December 2021)

• Record Type: Journal Article
• Authors: Ferreiro, Borja; Andrade, José; López-Mahía, Purificación; Muniategui, Soledad; Vázquez, Cristina; Pérez, Andrés; Rey, María; Vales, Carlos
• Is Part Of: Fuel. Volume 305(2021)
• Publication Date: 2021-12-01
• Subjects: Natural gas composition -- Wobbe index -- Infrared gas measurement -- PLS prediction -- Spectral variable Selection; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2021.121500
• Languages: English
• ISSNs: 0016-2361

Abstract

Highlights: A quality procedure to determine the composition of natural gas was developed. Production samples instead of synthetic mixtures yielded accurate predictions. Selectivity ratio and iPLS were selected among five variable selection procedures. The prediction errors and limits of detection halved those previously published. Abstract: Quality control of natural gas frequently relies on off-line slow standardized chromatographic techniques. Previous implementations of new measurement approaches focused of synthetic mixtures without extensive industrial validation. Here, a fast alternative based on infrared spectra is presented to predict the gas constituents and a physical parameter, the Wobbe index. Commercial samples instead of synthetic mixtures were used to develop predictive models. Method performance parameters were calculated and ca. 100 % of the sample-specific confidence intervals for the predictions overlapped with those of the reference values and the approach was unbiased and precise. The limits of detection and quantification (classical and considering errors of type I and II) outperformed other approaches. Validation included commercial samples and primary mixtures. Furthermore, prediction models considering reduced sets of variables were sought for using Markov-chain Monte Carlo guided searches (uninformative variable elimination and random frog) and common (iPLS, UVE and SR) approaches. The prediction errors and limits of detection of these 'reduced' models outperformed those from other approaches. The methodology takes only minutes to analyse a sample, requires few sample and no reagents (only some argon), making this approach cost-effective and environmentally-friendly.