基于机器学习的飞灰含碳量预测模型比较研究

Comparison of prediction models of carbon content of fly ash based on machine learning

锅炉飞灰含碳量是衡量锅炉燃烧效率的重要指标之一,基于机器学习构建了一套能够准确预测飞灰含碳的模型。首先,借助随机森林算法解决了飞灰含碳实测值与其他输入特征频率不一致的问题;其次,采用基于随机森林的递归特征消除方法,从30个原始输入特征中提取出9个输入特征,在降低模型计算量的同时提高了预测准确度;最后,以某电厂330 MW机组锅炉实际运行数据,建立了线性回归、决策树、KNN、随机森林、Catboost、XGBoost 6个机器学习模型对飞灰含碳量进行预测。预测结果发现:决策树、KNN、随机森林和XGBoost模型预测效果较好,均方误差分别为0.010、0.009、0.006和0.006,线性回归模型表现最差;构建的预测模型在锅炉低、中、高负荷下均保持稳定。

The carbon content of fly ash in boilers is one of the important indicators of combustion efficiency.This study employs machine learning models to accurately predict the carbon content of fly ash.Firstly,random forest is employed to adjust the frequency of fly ash carbon content data to once per minute,aligning it with the input features to address the issue of imbalanced data collection frequency.Then,a recursive feature elimination method based on random forest is used to extract nine important features out of the original 30 features,reducing feature correlation and improving model accuracy.Subsequently,six machine learning models(linear regression,decision tree,K-nearest neighbors(KNN),random forest,Catboost and XGBoost)are compared for prediction.The results indicate that decision tree,KNN,random forest and XGBoost models perform well,MSE of which on the test are 0.010,0.009,0.006 and 0.006,respectively,while linear regression exhibits the poorest performance.The prediction models remain robust under low,medium,and high boiler loads.