基于改进自适应增强算法的混煤发热量预测方法

The Method for Predicting Calorific Value of Blended Coal Based on Improved Adaptive Boosting Algorithm

为解决传统多元线性回归(Multivariate linear regression,MLR)模型在煤炭发热量预测方面精度不足和适用性有限的问题,提出了一种基于改进自适应增强算法(Adaptive boosting,Adaboost)的煤发热量的预测模型。将随机森林(Random forest,RF)作为Adaboost的基学习器,以提高模型在工业煤质分析中的发热量预测精度和泛化能力。研究基于某电厂1万组入炉煤的工业分析数据,选取水分、挥发分、灰分和固定碳作为模型输入,建立煤炭低位发热量的预测模型。通过与传统的多元线性回归方程及其他非线性模型比较,模型展现出更高的预测精度和更好的泛化能力。大样本测试的实验结果表明,本模型的平均绝对百分比误差为0.5417%,均方根误差为0.1304 MJ/kg,拟合度(R^(2))达到0.9799,其在煤炭发热量预测方面优于其他模型。此外,200组真实的混煤工业分析数据的模拟验证,进一步确认了本模型较优的泛化性能。

To address the problems of insufficient accuracy and limited applicability of traditional multivariate linear regression(MLR)models in predicting the calorific value of coal,a predictive model of coal calorific value based on improved adaptive boosting algorithm(Adaboost)has been proposed.This model incorporates random forest(RF)as Adaboost’s base learner to increase the accuracy and generalization capability of calorific value predictions within industrial coal quality analysis.The study is grounded on industrial analysis data from 10000 sets of as-fired coal in a power plant,selecting moisture,volatile matter,ash,and fixed carbon as inputs to construct a predictive model for coal’s lower heating value.When compared to traditional MLR equations and other nonlinear models,the proposed model exhibits superior prediction accuracy and enhanced generalization.Notably,in experiments conducted with large sample sizes,the results demonstrate the model has a mean absolute percentage error of 0.5417%,a root mean square error of 0.1304 MJ/kg,and a coefficient of determination(R^(2))of 0.9799,indicating its superior performance in predicting the calorific value of coal over other models.Moreover,the model’s superior generalization capability was further validated through a simulation with 200 sets of real blended coal industrial analysis data.