基于随机搜索算法优化XGBoost的过热汽温预测模型

Superheated Steam Temperature Prediction Models Based on XGBoost Optimized with Random Search Algorithm

建立过热器汽温对象的高精度性能预测模型,是实现过热汽温智能优化控制的基础。为此,针对某600 MW超临界机组仿真系统的历史运行数据,采用机器学习中的XGBoost(eXtreme Gradient Boosting)回归模型建立过热器汽温特性的预测模型,并分别采用网格搜索算法和随机搜索算法对模型参数进行优化。通过比较两种优化模型的预测效果,结果表明:随机搜索算法可以进行多个参数组合寻优,收敛速度快,全局寻优的效果更好,优化后的过热汽温模型具有更好的预测精度和泛化能力。

In order to realize intelligent optimized control of superheated steam temperature(SST),it is necessary to establish accurate prediction model for the controlled object.Therefore,the XGBoost(eXtreme Gradient Boosting)regression model in machine learning field is employed to establish the prediction model of SST characteristics with the historical simulation operating data of a 600 MW supercritical power unit.Grid search method(GSM)and randomized search method(RSM)are employed to optimize the predictive model’s parameters respectively.The results show that RSA can perform multiple parameter combination optimization with faster convergence speed and better global optimization effect,and the XGBoost model for SST optimized by RSM has better prediction accuracy and generalization ability.