重型燃气轮机进气系统是蒸汽-燃气联合循环电厂中的关键通流部件，其进口滤网堵塞可能引发压气机旋转失速、燃烧不稳定及内物损伤等严重故障。现有方法仅通过压差判断滤网堵塞情况，但未能充分考虑气体工质的质量流量、比容等运行工况参数变化的影响。为此，本文提出了一种基于深度学习的燃气轮机进气滤网堵塞程度量化诊断方法。首先，采集进气系统正常运行时的气路历史参数，建立深度神经网络预测模型，实现复杂变工况下理论健康压差的实时计算；其次，基于通流部件内在气路故障模式与外在气路可测参数故障征兆的关系，构建无量纲化进气系统健康特征参数模型，量化滤网堵塞程度。通过稳态与瞬态仿真实验，该方法对进气系统不同堵塞程度计算值的均方误差不超过0.09%，验证了所提出算法的准确性和鲁棒性。

The heavy-duty gas turbine intake system is a key flow component in a steam-gas combined cycle power plant, and its inlet filter blockage may cause serious failures such as compressor rotation stall, combustion instability and internal material damage. The existing methods only judge the blockage of the filter screen by the pressure difference, but fail to fully consider the influence of the changes of operating parameters such as mass flow rate and specific volume of the gas working fluid. To this end, this paper proposes a deep learning-based quantitative diagnosis method for the blockage degree of gas turbine air inlet filter. Firstly, the historical parameters of the gas path during the normal operation of the air intake system were collected, and the deep neural network prediction model was established to realize the real-time calculation of the theoretical healthy pressure difference under complex variable working conditions. Secondly, based on the relationship between the failure mode of the internal air path of the flow component and the fault symptoms of the measurable parameters of the external air path, a dimensionless model of the health characteristic parameters of the air intake system was constructed to quantify the degree of filter blockage. Through steady-state and transient simulation experiments, the mean square error of the proposed method for the calculated values of different blockages of the air intake system is not more than 0.09%, which verifies the accuracy and robustness of the proposed algorithm.

国家自然科学基金面上基金项目(62076160) ；上海市重型燃气轮机领域联合创新计划(UIC计划)资助项目（GYQJ-2023-1-06）；上海市“曙光计划”项目（23SG55）；上海市青年科技启明星项目资助(23QA140380)