作者：刘福国, 刘景龙, 张绪辉, 杨兴森

作者单位：国网山东省电力公司电力科学研究院，山东 济南 250002

关键词：高斯过程;煤;元素分析;工业分析;10折交叉验证

摘要：

利用煤的工业分析成分预测元素分析成分存在较多困难，现有文献给出的模型通常只对C、H和O等主要元素成分含量进行预测。而高斯过程能够解决复杂的机器学习问题，且可以对预测不确定性进行有效估计。该文建立基于高斯过程的煤元素含量预测模型，选用干燥无灰基挥发分和高位发热量作为随机过程的索引变量，分别对C、H、N、S元素含量进行高斯过程建模，O元素含量采用差减法得到，从而实现对元素分析全部成分含量的预测。采用10折交叉验证法对模型进行检验。结果表明，C、H、N和O元素含量预测的平均误差分别为1.46%、6.97%、16.80%和14.28%，但对于高硫煤，S元素含量的预测误差偏大，使用模型时应加以注意。

Prediction of comprehensive elemental compositions of coal based on Gaussian process
LIU Fuguo, LIU Jinglong, ZHANG Xuhui, YANG Xingsen
State Grid Shandong Electric Power Research Institute, Jinan 250002, China
Abstract: It is difficult to predict elemental compositions of coal using proximate analysis so far, and the models given in the existing literature usually predict the contents of the main elements such as C, H and O. Gaussian process models are routinely used to solve hard machine learning problems, and can effectively estimate the prediction uncertainty. In this paper, models based on Gaussian process are established to predict comprehensive elemental compositions of coal using proximate analysis. The dry ash free volatile matter and dry ash free basis high calorific value being selected as index variables, the Gaussian processes modeling are proposed to carbon, hydrogen, and oxygen compositions, respectively, and oxygen content is calculated by subtraction method, which realizes the prediction of all components in element analysis of coal. The model was tested and verified by 10 fold cross validation, results show that the averaged prediction errors of C, H, N and O are 1.46%, 6.97%, 16.80% and 14.28%, respectively. For coals with too high or low sulfur, the prediction error of S content is high, which means that one should be very careful to deploy the presented models to predict the elemental composition of coal with high or low sulfur contents.
Keywords: Gaussian process;coal;ultimate analysis;proximate analysis;10 fold cross validation
2021, 47(8):38-43  收稿日期: 2020-09-01;收到修改稿日期: 2020-10-11
基金项目: 国网山东省电力公司科技项目(520626200015)
作者简介: 刘福国(1969-),男,江苏邳州市人,高级工程师,硕士,主要从事电厂锅炉运行监测、诊断和优化研究工作
参考文献
[1] SAPTORO A, YAO H M, MO T, et al. Prediction of coal hydrogen content for combustion control in power utility using neural network approach[J]. Chemometrics and Intelligent Laboratory Systems, 2008, 94(2): 149-159
[2] 韩璞, 乔弘, 王东风, 等. 火电厂热工参数软测量技术的发展和现状[J]. 仪器仪表学报, 2007, 28(6): 1139-1146
[3] 卢毅, 宋朝晖. 瞬发伽马中子活化分析技术发展现状[J]. 核电子学与探测技术, 2013, 33(12): 1527-1531
[4] MESROGHLI S, JORJANI E, CHELGANI S. Estimation of gross calorific value based on coal analysis using regression and artificial neural networks[J]. International Journal of Coal Geology, 2009, 79(1-2): 49-54
[5] CHELGANI S C, HOWER J C, JORJANI E, et al. Prediction of coal grindability based on petrography, proximate and ultimate analysis using multiple regression and artificial neural network models[J]. Fuel Processing Technology, 2008, 89(1): 13-20
[6] PARIKH J, CHANNIWALA S A, GHOSAL G K. A correlation for calculating HHV from proximate analysis of solid fuels[J]. Fuel, 2005, 84(5): 487-494
[7] PARIKH J, CHANNIWALA S A, GHOSAL G K. A correlation for calculating elemental composition from proximate analysis of biomass materials[J]. Fuel, 2007, 86(12): 1710-1719
[8] YI L, FENG J, QIN Y H, et al. Prediction of elemental composition of coal using proximate analysis[J]. Fuel, 2017, 193: 315-321
[9] RASMUSSEN C, WILLIAMS C. Gaussian processes for machine learning[M]. US: MIT Press, 2006.
[10] SEEGER M. Gaussian processes for machine learning[J]. International Journal of Neural Systems, 2008, 14(2): 69-106
[11] SAPTORO A, VUTHALURU H B, TADÉ M O. A comparative study of prediction of elemental composition of coal using empirical modelling[J]. IFAC Proceedings Volumes, 2006, 39(2): 747-752
[12] NHUCHHEN R D. Prediction of carbon, hydrogen, and oxygen compositions of raw and torrefied biomass using proximate analysis[J]. Fuel, 2016, 180: 348-356