Quantitative Analysis of XRF Iron Ore Grade Combining Morphology and Optimization Algorithms
WANG Lan-hao1, ZHU Zhen-yu2, ZHONG Xiao3, WANG Hong-yan3*, LI Zhao-peng4
1. State Key Laboratory of Coking Coal Resources Green Exploitation, China University of Mining and Technology, Xuzhou 221116, China
2. Artificial Intelligence Research Institute, China University of Mining and Technology, Xuzhou 221116, China
3. School of Information and Control Engineering, China University of Mining Technology, Xuzhou 221116, China
4. LONGi Magnet Co., LTD., Shenyang 110000, China
Abstract：This study addresses the limitations of existing X-ray fluorescence spectroscopy (XRF) technology for the online detection of iron ore grade, including strong background interference, difficulty in resolving overlapping spectral peaks, and insufficient modeling accuracy. It proposes a comprehensive solution that covers the entire process, from signal pre-processing and spectral peak decomposition to grade modeling. In the background subtraction stage, a method combining mathematical morphology and derivative-iterative polynomial fitting (Mor+DIPF) has been developed. This method significantly improves baseline smoothness and fitting accuracy, while also ensuring spectral fidelity in the region of characteristic peaks. This addresses the deficiency of traditional morphological algorithms, which do not sufficiently smooth this region. To address complex peak overlap, a particle swarm optimization (PSO) approach with adaptive parameter updates (APU-PSO) is combined with an expectation-maximization (EM) Gaussian mixture model (GMM) decomposition framework. This enhances global optimization capabilities and enables high-precision analysis, providing accurate peak parameters for subsequent quantitative analysis. To address non-linear errors caused by matrix effects, a fusion model combining transformers and bidirectional long short-term memory networks (BiLSTM) is constructed. Transformers capture long-range dependencies among process variables, while BiLSTMs enhance the learning of temporal dynamics. By fusing multi-source data, key factors influencing grade fluctuations are thoroughly identified, overcoming the challenge of accurately predicting iron ore grades using XRF. Experimental results demonstrate that the Mor+DIPF algorithm outperforms traditional methods in terms of root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2) across four different baselines (linear, sinusoidal, Gaussian, and exponential), with R2 achieving a maximum value of 99.96%. The APU-PSO-EM-GMM algorithm outperforms the comparison algorithm.
[1] FAN Song-mei, SHA Jing-hua, YAN Jing-jing, et al(范松梅, 沙景华, 闫晶晶, 等). Resources Science(资源科学), 2018, 40(3): 507.
[2] Zhao J S, Ni K, Su Y P, et al. Construction and Building Materials, 2021, 286: 122968.
[3] Zhou Y L, Jiang X, Wang X A, et al. Metallurgical and Materials Transactions B, 2022, 53(6): 4075.
[4] LU Dan(芦 丹). Mining Research and Development(矿业研究与开发), 2019, 39(12): 177.
[5] AN Xiao-qiang, ZHOU Chang-chun, LI Zhen, et al(安小强, 周长春, 李 振, 等). Light Metals(轻金属), 2008，(11): 52.
[6] ZHANG Yang, ZHENG Shi-li, WANG Xiao-hui, et al(张 洋, 郑诗礼, 王晓辉, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2010, 30(1): 251.
[7] WANG Min-lei(王敏蕾). Non-ferrous Mining and Metallurgy (有色矿冶), 2007, 23(4): 88.
[8] WANG Yong-bao(王永保). Chemical Enterprise Management(化工管理), 2020，(32): 69.
[9] Shard A G, Wright L, Minelli C. Biointerphases, 2018, 13(6): 061002.
[10] Tekin Z, Unutkan T, Erulas F, et al. Food Chemistry, 2020, 310: 125825.
[11] Nakano H, Komatani S, Matsuyama T, et al. Analytical Sciences, 2021, 37(10): 1447.
[12] Singh V, Sharma N. X-ray Spectrometry, 2022, 51(3): 304.
[13] Lagerstrom M, Ytreberg E. Talanta, 2021, 223(2): 121820.
[14] Ma Q, Pollard A M, Jiang J, et al. Archaeometry, 2021, 63(1): 194.
[15] Li Y, Wang Q, Sun N, et al. Spectroscopy Letters, 2013, 46(7): 507.
[16] Yan Y, Zhao X, Zhang D. Medicinal Plant, 2014, 5(6): 33.
[17] HU Jin-song, BAO Ji-long, ZHOU Fang-hao, et al(胡劲松, 鲍吉龙, 周方浩, 等). Chinese Journal of Analytical Chemistry(分析化学), 2006, 34(3): 418.
[18] Yang X Y, Zhang C C, Wang C Y, et al. Ultrasonics, 2018, 91: 231.
