Simultaneous Determination of Metal Ions in Zinc Sulfate Solution Using UV-Vis Spectrometry and SPSE-XGBoost Method

Fei Cheng, Chunhua Yang, Can Zhou, Lijuan Lan, Hongqiu Zhu, Yonggang Li

TL;DR

This paper introduces a novel "Expand-Extract-Combine" strategy using UV-Vis spectrometry and a SPSE-XGBoost model for the simultaneous, real-time determination of multiple heavy metal ions (Zn, Cu, Co, Ni) in industrial wastewater. This approach effectively handles spectral overlap and noise, significantly improving prediction accuracy and speed compared to traditional methods.

What Problem Does This Paper Address?

The discharge of heavy metal ions from industrial processes, particularly zinc smelting, poses a significant threat to environmental and human health. Current methods for monitoring these ions are often off-line, labor-intensive, time-consuming, and prone to errors, hindering effective wastewater control. Existing spectroscopic techniques, like UV-Vis, struggle with complex solutions due to severe spectral overlap and non-linearity, making accurate simultaneous determination of multiple ions challenging. This paper aims to bridge this gap by developing a rapid, accurate, and on-line method for simultaneous multi-metal ion detection.

Methods

The core of this work is the "Expand-Extract-Combine" strategy:

  • Expand (Preprocessing): The authors employ the Singular Perturbation Spectrum Estimator (SPSE) to generate multi-order derivative spectra (denoising, 1st-order, and 2nd-order). SPSE is a novel technique that uses an inverse Taylor series approach to enhance spectral resolution, suppress background noise, and separate overlapping spectral bands, thereby expanding the useful spectral information.
  • Extract (Feature Selection): The Extreme Gradient Boosting (XGBoost) algorithm, a powerful ensemble learning method, is used for feature selection. XGBoost calculates a "Variable Importance" (VI) score for each spectral feature, identifying the most relevant wavelengths for predicting ion concentrations. This allows for the extraction of sensitive feature bands.
  • Combine (Model Building): Feature subsets derived from different derivative spectra are then combined. This combined feature set is used to train the XGBoost model for predicting the concentrations of Zn, Cu, Co, and Ni. This combination strategy leverages the diversity of information captured by different derivative orders.

Datasets

The study utilized a dataset of 49 mixed solutions containing zinc, copper, cobalt, and nickel ions. These samples were prepared with varying concentrations: Zn(II) ranging from 10 to 70 mg/L, and Cu(II), Co(II), and Ni(II) ranging from 0.2 to 1.4 mg/L. The dataset was split into a training set (39 samples) and a test set (10 samples) using the Kennard-Stone algorithm.

Key Insights and Ideas

  • Derivative Spectra are Crucial: The paper strongly emphasizes that multi-order derivative spectra (especially 1st and 2nd order) are essential for resolving overlapping spectral bands and enhancing subtle spectral features that are critical for distinguishing between similar metal ions.
  • SPSE Outperforms Standard Derivatives: The SPSE method for generating derivative spectra is shown to be more effective than standard methods (like Savitzky-Golay) in suppressing noise and improving spectral quality, leading to better prediction accuracy.
  • XGBoost for Feature Selection and Prediction: XGBoost proves to be a highly effective tool not only for building predictive models but also for robustly selecting the most informative spectral features, significantly reducing dimensionality and improving computational efficiency.
  • The Power of Combination: Combining feature subsets from different derivative spectra (the "Combine" step) significantly boosts prediction accuracy, demonstrating that leveraging diverse spectral information leads to more comprehensive and robust models. This is a novel contribution to metal ion determination.
  • Real-time Monitoring Potential: The developed method's speed and accuracy make it a strong candidate for real-time, on-line monitoring in industrial settings, enabling immediate feedback for process control.

Key Results

  • The SPSE-XGBoost method achieved the lowest Root Mean Square Error of Prediction (RMSEp) among all tested methods for all four metal ions.
  • The combined feature subset approach yielded the best results, with MAPE values as low as 3.083% for cobalt and MaxAPE below 8.323% for all ions, meeting industrial accuracy requirements.
  • The R² values for the test set predictions were consistently high, exceeding 0.987 for all ions, indicating excellent correlation between predicted and actual concentrations.
  • The "Expand-Extract-Combine" strategy significantly improved prediction accuracy and computational speed by utilizing selected feature subsets instead of the full spectrum.
  • Comparisons with traditional methods like CARS-PLS and UVE-LS-SVM showed that the proposed SPSE-XGBoost approach offered superior performance.

Limitations and Future Work

  • Limitations: While the paper demonstrates strong performance, the study focused on a specific matrix (zinc sulfate solution) and a limited set of metal ions. The presence of other complex matrices or a wider range of interfering ions might affect performance. The paper also notes that the second-order derivative spectrum can sometimes amplify signal instability, particularly for weaker signals like nickel, potentially reducing prediction precision.
  • Future Work: The authors suggest that this developed approach has great potential for real-time and on-line detection of multi-metal ion concentrations in other industrial wastewater scenarios. Further research could explore its applicability to more complex matrices and a broader spectrum of heavy metals. Investigating more advanced feature selection and combination techniques could also be a direction for future work.

Summary updated

Abstract

Excessive discharge of heavy metal ions will aggravate environment pollution and threaten human health. Thus, it is of significance to real-time detect metal ions and control discharge in the metallurgical wastewater. We developed an accurate and rapid approach based on the singular perturbation spectrum estimator and extreme gradient boosting (SPSE-XGBoost) algorithms to simultaneously determine multi-metal ion concentrations by UV–vis spectrometry. In the approach, the spectral data is expanded by multi-order derivative preprocessing, and then, the sensitive feature bands in each spectrum are extracted by feature importance (VI score) ranking. Subsequently, the SPSE-XGBoost model are trained to combine multi-derivative features and to predict ion concentrations. The experimental results indicate that the developed “Expand-Extract-Combine” strategy can not only overcome problems of background noise and spectral overlapping but also mine the deeper spectrum information by integrating important features. Moreover, the SPSE-XGBoost strategy utilizes the selected feature subset instead of the full-spectrum for calculation, which effectively improves the computing speed. The comparisons of different data processing methods are conducted. It outcomes that the proposed strategy outperforms other routine methods and can profoundly determine the concentrations of zinc, copper, cobalt, and nickel with the lowest RMSE P . Therefore, our developed approach can be implemented as a promising mean for real-time and on-line determination of multi-metal ion concentrations in zinc hydrometallurgy.
BibTeX
@inproceedings{cheng2020simultaneous,
  author = {Fei Cheng and Chunhua Yang and Can Zhou and Lijuan Lan and Hongqiu Zhu and Yonggang Li},
  title = {{Simultaneous Determination of Metal Ions in Zinc Sulfate Solution Using UV-Vis Spectrometry and SPSE-XGBoost Method}},
  booktitle = {Sensors},
  year = {2020},
  url = {https://openreview.net/forum?id=jkREn0pwA7},
}
Figure