针对航空发动机动态特性的建模问题，提出一种基于麻雀搜索算法（SSA）优化NARX神经网络的动态特性参数辨识方法。利用SSA对NARX网络的权值与偏置进行迭代寻优，使网络具备更高的准确度与泛化能力；利用优化后的NARX网络进行动态参数辨识；使用航空发动机飞行测试数据集进行了仿真测试。结果表明：SSANARX方法明显优于NARX和PSONARX方法。SSANARX方法的输出参数N1，N2和排气温度（EGT）与真实值的最大相对误差绝对值δmax分别降低至3.81%，1.24%和3.47%；动态特性指标Ti与Tt与真实值的相对误差均小于5%；经10次交叉试验，参数N1，N2和EGT的测试结果均方根误差均值RMSEm分别为0.29，0.18和1.50。模型的准确性、实时性与稳健性均满足了仿真需求。

A dynamic characteristic parameter identification method based on the sparrow search algorithm (SSA) and nonlinear autoregressive exogenous (NARX) neural network was proposed for the aeroengine dynamic characteristics modeling. SSA was used to optimize the weights and biases of the NARX network iteratively to enhance its accuracy and generalization capability; the optimized NARX network was used for identifying dynamic characteristic parameters; simulation tests were conducted using flight test dataset. The results show that SSANARX method is better than the NARX and PSONARX methods obviously. By the SSANARX method, the maximum relative error absolute valuesδmaxbetween the output parameters N1, N2 and exhaust gas temperature (EGT) and actual values are reduced to 3.81%, 1.24% and 3.47% respectively; the relative errors between the dynamic characteristic indicators Ti and Tt and actual values are less than 5%;through ten cross tests, the mean values of root mean square errors (RMSEm)

TK231

国家自然科学基金面上项目(12072196)