基于数据驱动的燃气轮机性能监测方法  

Data-driven Gas Turbine Performance Monitoring Method

作　　者：李永华[1] 陈伟华[1] 谢英柏[1] LI Yonghua;CHEN Weihua;XIE Yingbai(Department of Power Engineering,North China Electric Power University,Baoding 071003,Hebei Province,China)

机构地区：[1]华北电力大学能源与动力机械工程学院,河北保定071003

出　　处：《动力工程学报》2024年第12期1845-1853,共9页Journal of Chinese Society of Power Engineering

基　　金：国家自然科学基金资助项目(52072081)。

摘　　要：为分析机组负荷指令和环境参数等对燃气轮机运行特性的影响,研究了基于某电厂历史运行数据的性能监测方法。采用多层感知机(MLP)神经网络、极致梯度提升(XGBoost)算法和分类梯度提升(CatBoost)算法构建压气机效率、燃烧效率和透平效率的预测模型。结果表明:机组负荷指令每增加1 MW,基准燃料质量流量增加0.03 kg/s,基准压比增加0.0289,基准温比增加0.013;大气温度每增加1 K,基准压气机排气温度提升1.34 K;相比于MLP神经网络、CatBoost算法,XGBoost算法收敛速度更快,测试集的均方根误差分别降低0.019和0.001,平均绝对误差分别降低0.028与0.004,R^(2)分别增加0.041和0.002。To analyze the influence of unit load command and environmental parameters on gas turbine operating characteristics,the performance monitoring method based on the historical operating data of a power plant was studied.Multi-layer perceptron(MLP)neural network,extreme gradient boosting(XGBoost)algorithm and classification gradient boosting(CatBoost)algorithm were used to construct predict model of compressor efficiency,combustion efficiency and turbine efficiency.Results show that for each 1 MW increase in unit load command,the reference fuel mass flow rate increases by 0.03 kg/s,the reference pressure ratio increases by 0.0289,and the reference temperature ratio increases by 0.013.While the atmospheric temperature increases by 1 K,the reference compressor exhaust temperature increases by 1.34 K.Compared to the MLP neural network and CatBoost algorithm,the XGBoost algorithm converges faster,and the root mean square error of the test set is reduced by 0.019 and 0.001,the mean absolute error is reduced by 0.028 and 0.004,and R^(2)is increased by 0.041 and 0.002.

关 键 词：性能监测 数据驱动 机器学习 燃气轮机 工程应用 

分 类 号：TK477[动力工程及工程热物理—动力机械及工程]