基于贝叶斯优化—随机森林回归的燃煤锅炉NO_(x)预测模型

NO_(x) Prediction Model for Coal-fired Boiler Based on Bayesian Optimization-Random Forest Regression

根据某超超临界1 050 MW燃煤机组实际运行数据,采用随机森林(RF)算法建立燃煤锅炉炉膛出口烟气中NOx质量浓度预测模型,并利用贝叶斯优化(BO)进行超参数寻优,将BO-RF模型与网格搜索优化的RF模型(GSO-RF)进行对比。为了更好地评价预测模型,以平均绝对百分比误差δMAPE和决定系数R^(2)作为评价指标,将所建立的BO-RF模型与目前常见的基于贝叶斯优化的BP神经网络(BO-BPNN)模型、最小二乘支持向量机(BO-LSSVM)模型进行比较。结果表明:BO-RF模型比GSO-RF模型的预测精度更高,且BO-RF模型的δMAPE为1.478%,R2为0.916 2,均优于BO-BPNN模型和BO-LSSVM模型的预测结果,证明BO-RF模型具有更高的预测精度和更优的泛化性能。

Based on the actual operating data of a coal-fired boiler in a 105o MW ultra-supercritical unit,a random forest(RF)algorithm was used to establish a prediction model for NO,concentration in the flue gas at the furnace outlet of the coal-fired boiler,and Bayesian optimization(BO)was used to optimize the hyperparameters.Then the BO-RF model was compared with the grid search optimized RF model(GSORF).In order to better evaluate the prediction model,the established BO-RF model was compared with the current common error Bayesian optimized back propagation neural network(BO-BPNN)model and Bayesian optimized least square support vector machine(BO-LSSVM)model,using the average absolute percentage error OmAPe and the coefficient of determination R?as evaluation indicators.Results show that the prediction accuracy of the BO-RF model is higher than that of the GSO-RF model,and the MAPE of the BO-RF model is 1.478%,the R?is 0.9162,which are better than the prediction results of the BO-BPNN model and the BO-LSSVM model,indicating that the BO-RF model has higher prediction accuracy and better generalization performance.