基于分位数回归的煤炭发热量预测

赵先枝; 陈军林

doi:10.13272/j.issn.1671-251x.2022060023

基于分位数回归的煤炭发热量预测

doi: 10.13272/j.issn.1671-251x.2022060023

赵先枝^1,,
陈军林²

1.
内蒙古煤炭地质勘查(集团)一五三有限公司, 内蒙古呼和浩特　010010
2.
中国地质大学(北京) 地球科学与资源学院, 北京　100083)

基金项目: 四川省自然科学基金资助项目（2022NSFSC1734）。

详细信息

作者简介:
赵先枝（1965—），女，内蒙古呼和浩特人，副高级工程师，现从事地质工程勘查技术工作，E-mail：498738032@qq.com

中图分类号: TD94
计量
- 文章访问数: 175
- HTML全文浏览量: 37
- PDF下载量: 11
- 被引次数: 0
出版历程
- 收稿日期: 2022-06-08
- 修回日期: 2022-07-19
- 网络出版日期: 2022-07-06

Prediction method of coal calorific value based on quantile regression

ZHAO Xianzhi^1
,,
CHEN Junlin²

1.
Inner Mongolia Coal Geology Exploration (Group) 153 Co., Ltd., Hohhot 010010, China
2.
School of Earth Sciences and Resources, China University of Geosciences(Beijing), Beijing 100083, China

摘要

摘要: 目前应用较多的煤炭发热量预测模型以传统的线性回归模型为主，但存在难以表达较复杂的自变量和因变量关系、需要数据服从特定的分布假设、对异常值敏感等问题。针对上述问题，提出了基于分位数回归的煤炭发热量预测方法。选取全水分、灰分、挥发分等容易测量的煤炭工业分析指标，分别应用线性分位数回归和分位数回归森林2种分位数回归方法对煤炭发热量进行预测，并与传统的线性回归方法进行对比。结果表明：线性回归给出的煤炭发热量预测值仅是1个条件均值，而通过分位数回归能够给出煤炭发热量预测值的范围；分位数回归森林的预测效果优于线性回归和线性分位数回归方法；全水分对于煤炭发热量预测的重要程度远大于灰分和挥发分；全水分对低发热量煤炭的发热量预测影响大，对高发热量煤炭的发热量预测影响小；挥发分和灰分对低发热量煤炭的发热量预测影响小，对高发热量煤炭的发热量预测影响大。
- 煤炭发热量 /
- 发热量预测 /
- 线性分位数回归 /
- 分位数回归森林 /
- 线性回归
Abstract: At present, the traditional linear regression model is mainly used to predict the calorific value of coal. But it is difficult to express the complex relationship between independent variables and dependent variables. The model needs data to obey specific distribution assumptions. And the model is sensitive to abnormal values. In view of the above problems, a prediction method of coal calorific value based on quantile regression is proposed. The method selects the coal industry analysis indicators that are easy to measure, such as total moisture, ash and volatile matter. The method uses two quantile regression methods, linear quantile regression and quantile regression forest, to predict the calorific value of coal. The results are compared with that of the traditional linear regression method. The results show that the predicted value of calorific value of coal given by linear regression is only a conditional mean value. But the range of predicted value of calorific value of coal can be given by quantile regression. The prediction effect of quantile regression is better than linear regression and linear quantile regression. The importance of total moisture for the prediction of calorific value of coal is much greater than that of ash and volatile matter. Total moisture has great influence on the prediction of calorific value of low calorific value coal. But total moisture has little influence on the prediction of calorific value of high calorific value coal. Volatile matter and ash have little influence on the prediction of calorific value of low calorific value coal. But volatile matter and ash have a great influence on the prediction of calorific value of high calorific value coal.
- calorific value of coal /
- calorific value prediction /
- linear quantile regression /
- quantile regression forest /
- linear regression

HTML全文

图 1 不同回归分析方法下回归拟合线

Figure 1. Regression fitting lines under different regression analysis methods

下载: 全尺寸图片幻灯片

图 2 线性分位数回归系数随分位点变化曲线

Figure 2. Variation curves of linear quantile regression coefficients with quantiles

下载: 全尺寸图片幻灯片

表 1 煤质参数相关系数

Table 1. Correlation coefficients of coal quality parameters

煤质参数	相关系数
煤质参数	M_t	V_ad	A_sd	Q_net,ad
M_t	1.00	−0.20	−0.10	−0.92
V_ad	−0.20	1.00	−0.20	0.18
A_sd	−0.10	−0.20	1.00	−0.23
Q_net,ad	−0.92	0.18	−0.23	1.00

下载: 导出CSV

表 2 不同回归模型评价结果

Table 2. Evaluation results of different regression models

方法	τ	均方误差	平均绝对误差	均方根误差	决定系数
线性回归	—	0.856	1.322	1.150	0.969
线性分位数回归	0.1	1.391	4.380	2.093	0.898
	0.2	1.061	2.897	1.702	0.932
	0.3	0.908	2.083	1.443	0.951
	0.4	0.836	1.658	1.288	0.961
	0.5	0.822	1.457	1.207	0.966
	0.6	0.860	1.511	1.229	0.965
	0.7	0.936	1.734	1.317	0.960
	0.8	1.096	2.344	1.531	0.945
	0.9	1.336	3.236	1.799	0.925
分位数回归森林	0.1	1.451	3.128	1.769	0.927
	0.2	0.947	1.470	1.212	0.966
	0.3	0.717	0.937	0.968	0.978
	0.4	0.595	0.736	0.858	0.983
	0.5	0.562	0.705	0.840	0.984
	0.6	0.603	0.854	0.924	0.980
	0.7	0.707	1.150	1.072	0.973
	0.8	0.940	1.876	1.370	0.956
	0.9	1.456	3.872	1.968	0.910

下载: 导出CSV

表 3 不同分位点下线性分位数回归系数

Table 3. Linear quantile regression coefficients under different quantiles

τ	M_t回归系数	V_ad回归系数	A_sd回归系数
0.1	−0.767	−0.054	−0.391
0.2	−0.748	−0.073	−0.391
0.3	−0.726	−0.079	−0.394
0.4	−0.706	−0.083	−0.397
0.5	−0.684	−0.087	−0.401
0.6	−0.668	−0.090	−0.404
0.7	−0.650	−0.090	−0.406
0.8	−0.621	−0.088	−0.405
0.9	−0.599	−0.083	−0.402

下载: 导出CSV

参考文献(15)

[1]	李纯毅. 煤质分析[M]. 北京: 北京理工大学出版社, 2012. LI Chunyi. Coal quality analysis[M]. Beijing: Beijing Institute of Technology Press, 2012.
[2]	李英华. 煤质分析应用技术指南[M]. 北京: 中国标准出版社, 1991. LI Yinghua. Technical guide for application of coal quality analysis[M]. Beijing: Standards Press of China, 1991.
[3]	郝锡林. 杏花矿洗混煤发热量回归方程的建立[J]. 煤炭技术,2006,25(3):75-77. doi: 10.3969/j.issn.1008-8725.2006.03.040 HAO Xilin. Establishment of tropic equation for the quality of heat of washed-coal in Xinghua Coal Mine[J]. Coal Technology,2006,25(3):75-77. doi: 10.3969/j.issn.1008-8725.2006.03.040
[4]	郝飞. 影响煤炭发热量测量的常见因素分析[J]. 煤质技术,2019,34(5):61-64. doi: 10.3969/j.issn.1007-7677.2019.05.015 HAO Fei. Analysis of the common influencing factors which affect coal calorific value determination[J]. Coal Quality Technology,2019,34(5):61-64. doi: 10.3969/j.issn.1007-7677.2019.05.015
[5]	李大虎,李秋科,王文才,等. 基于MIV特征选择与PSO−BP神经网络的煤炭发热量预测[J]. 煤炭工程,2020,52(11):154-160. LI Dahu,LI Qiuke,WANG Wencai,et al. Prediction of coal calorific value based on MIV characteristic variable selection and PSO-BP neural network[J]. Coal Engineering,2020,52(11):154-160.
[6]	李大虎,韦鲁滨,朱学帅,等. 基于SVR与特征变量选择方法的煤炭发热量预测[J]. 煤炭学报,2019,44(增刊1):278-288. doi: 10.13225/j.cnki.jccs.2018.1268 LI Dahu,WEI Lubin,ZHU Xueshuai,et al. Prediction of coal calorific value based on SVR and characteristic variables selection method[J]. Journal of China Coal Society,2019,44(S1):278-288. doi: 10.13225/j.cnki.jccs.2018.1268
[7]	潘红光,宋浩骞,苏涛,等. 基于SVM的煤炭低位发热量软测量[J]. 西安科技大学学报,2021,41(6):1130-1137. doi: 10.13800/j.cnki.xakjdxxb.2021.0622 PAN Hongguang,SONG Haoqian,SU Tao,et al. Soft sensor of coal net calorific value based on SVM[J]. Journal of Xi'an University of Science and Technology,2021,41(6):1130-1137. doi: 10.13800/j.cnki.xakjdxxb.2021.0622
[8]	KOENKER R,BASSETT G. Regression quantiles[J]. Econometrica:Journal of the Econometric Society,1978,46(1):33-50. doi: 10.2307/1913643
[9]	PALMER C, OMAN C, PARK A, et al. The US geological survey coal quality（COALQUAL） database version 3.0[R]. Reston: US Geological Survey, 2015.
[10]	陈建宝,丁军军. 分位数回归技术综述[J]. 统计与信息论坛,2008(3):89-96. doi: 10.3969/j.issn.1007-3116.2008.03.018 CHEN Jianbao,DING Junjun. A review of technologies on quantile regression[J]. Statistics & Information Forum,2008(3):89-96. doi: 10.3969/j.issn.1007-3116.2008.03.018
[11]	关静. 分位数回归理论及其应用[D]. 天津: 天津大学, 2009. GUAN Jing. The theory of quantile regression and applications[D]. Tianjin: Tianjin University, 2009.
[12]	MEINSHAUSEN N,RIDGEWAY G. Quantile regression forests[J]. Journal of Machine Learning Research,2006,7(2):983-999.
[13]	PRADEEPKUMAR D,RAVI V. Forecasting financial time series volatility using particle swarm optimization trained quantile regression neural network[J]. Applied Soft Computing,2017,58:35-52. doi: 10.1016/j.asoc.2017.04.014
[14]	HE Yaoyao,QIN Yang,WANG Shuo,et al. Electricity consumption probability density forecasting method based on LASSO-quantile regression neural network[J]. Applied Energy,2019,233:565-575.
[15]	LIAW A,WIENER M. Classification and regression by random forest[J]. R News,2002,2(3):18-22.