View in EDS

MICE-PSO-RF Model for Predicting Coal Spontaneous Combustion Temperature Based on Multiple Imputation by Chained Equations.

Saved in:

Bibliographic Details
Title:	MICE-PSO-RF Model for Predicting Coal Spontaneous Combustion Temperature Based on Multiple Imputation by Chained Equations.
Authors:	Zheng, Xuezhao^1,2,3 (AUTHOR), Li, Peihua¹ (AUTHOR) 24220089058@stu.xust.edu.cn, Cai, Guobin^1,2,3 (AUTHOR) 22120089018@stu.xust.edu.cn, Guo, Jun^1,2,3 (AUTHOR), Liu, Yin^1,2 (AUTHOR)
Source:	Combustion Science & Technology. 2026, Vol. 198 Issue 8, p2225-2242. 18p.
Subject Terms:	Multiple imputation (Statistics), Random forest algorithms, Coal combustion, Machine learning, Particle swarm optimization, Temperature measurements, *Missing data (Statistics)
Abstract:	To address the issue of missing indicator gas data in coal spontaneous combustion temperature prediction models in practical applications, a novel prediction model is proposed, which integrates the MICE (Multiple Imputation by Chained Equations) method with PSO (Particle Swarm Optimization)-optimized RF (Random Forest) algorithm. A coal spontaneous combustion heating experiment was conducted, selecting O2, CO, CO2, CH4, and C2H6 as characteristic gases for coal combustion. Five sets of data with randomly introduced missing values were designed, and four imputation methods – KNN, RF, PSO-SVR, and MICE – were used to impute the missing values. The imputed data were then used to train three temperature prediction models (BP, RF, and SVR), with parameters optimized using PSO. The imputation and prediction results were compared. The results indicate that, across five different missing data rates, the MICE method outperforms the other imputation techniques. Without parameter tuning, all three models showed overfitting. However, after PSO optimization, the PSO-RF model achieved the highest prediction accuracy and demonstrated good stability. In practical applications, the MICE-PSO-RF model yielded an average temperature prediction error of 2.69°C across 15 test sets, demonstrating reliable performance in predicting coal temperature. [ABSTRACT FROM AUTHOR]
Database:	Energy & Power Source

Description
Abstract:	To address the issue of missing indicator gas data in coal spontaneous combustion temperature prediction models in practical applications, a novel prediction model is proposed, which integrates the MICE (Multiple Imputation by Chained Equations) method with PSO (Particle Swarm Optimization)-optimized RF (Random Forest) algorithm. A coal spontaneous combustion heating experiment was conducted, selecting O2, CO, CO2, CH4, and C2H6 as characteristic gases for coal combustion. Five sets of data with randomly introduced missing values were designed, and four imputation methods – KNN, RF, PSO-SVR, and MICE – were used to impute the missing values. The imputed data were then used to train three temperature prediction models (BP, RF, and SVR), with parameters optimized using PSO. The imputation and prediction results were compared. The results indicate that, across five different missing data rates, the MICE method outperforms the other imputation techniques. Without parameter tuning, all three models showed overfitting. However, after PSO optimization, the PSO-RF model achieved the highest prediction accuracy and demonstrated good stability. In practical applications, the MICE-PSO-RF model yielded an average temperature prediction error of 2.69°C across 15 test sets, demonstrating reliable performance in predicting coal temperature. [ABSTRACT FROM AUTHOR]
ISSN:	00102202
DOI:	10.1080/00102202.2025.2500516