提出了机理模型和基于神经网络的数据驱动模型相结合的燃机混合模型结构，利用燃机实际运行数据，通过数据驱动方法修正由于特性线偏差导致的机理模型仿真误差。以天然气长输管道压气站的某型双轴燃气轮机为研究对象，设计了3种机理-数据混合模型结构。通过对比3种混合模型与机理模型的性能仿真结果，说明与燃机机理模型相比，混合模型精度相对更高。同时基于燃机整体并联补偿的混合模型结构对截面参数的仿真精度最高，各参数的平均相对误差基本在1.5%以内，其中压气机出口和动力透平出口温度的相对误差小于0.5%。由于数据驱动方法可以针对特定燃机利用运行数据训练出特定的补偿网络，对于不同的机组所设计的混合结构模型都能具有较高的准确性，为提高燃机模型精度提供了参考手段。

As the working characteristics of gas turbine are difficult to obtain,and the unknown offset of its component characteristics occurs with the operation of the gas turbine,the current gas turbine modeling methods usually do not have matching characteristic lines which affects the accuracy of the gas turbine model.Aiming at the above problems,hybrid modeling structures of gas turbine based on mechanism model and datadriven model are proposed.Datadriven method is used to eliminate or reduce the error of mechanism model.In this paper,with a marine twoshaft gas turbine model as an example,the effect of three different hybrid modeling structures is studied.The simulation results of the three hybrid models and the mechanism model show that the hybrid models of gas turbine are more accurate than the mechanism model.In this case,the accuracy of parallel hybrid model of gas turbine is the highest,with the average relative error of less than 1.5%,and the relative error of less than 0.5% under certain operating conditions.As the datadriven method can train a specific compensation network for a specific gas turbine according to the historical operation data,the hybrid structure model designed in this paper has high accuracy for different units,which provides a reference means for improving the accuracy of the gas turbine model.

TK14

国家自然科学基金（51876116，51706132）;基础科研重大项目（JCKY2017208A001）