基于数据驱动的燃气轮机剩余寿命预测

Data Driven based Prognostic of Remaining Useful Life for Gas Turbine

提出一种基于数据驱动的燃气轮机剩余寿命预测方法。将随时间呈上升趋势的参数(温度、压力、转速等)经过EMD方法分解,获得反映燃气轮机性能劣化的趋势特征;利用趋势特征和人为设定的若干运行状态训练神经网络,将实时运行数据输入神经网络进行分类以获得当前所处的运行状态;考虑到分类结果的离散性,利用最小二乘直线拟合方法将所有分类结果进行拟合,以获得预测的剩余使用寿命。上述方法通过C-MAPSS软件所提供的燃气轮机运行数据得到验证,取得较好的预测效果。

A data driven based approach of remaining useful life prediction for gas turbine is presented. Signals ascending with time such as temperature, pressure and rotating speed are decomposed by empirical mode decomposition, and the tendency features reflec-ting the deterioration process of gas turbine can be obtained. An artificial neural network is trained using the tendency features and op-erating state by artificial arrangement, then the online data is input into the trained neural network and the classification result can be calculated. Considering the discreteness of classification result, a linear least square method is applied to fit the classification result and the predicted remaining useful life is obtained. The above methodology is demonstrated by the datasets generated from the C-MAPSS software.