基于SSA-NARX的航空发动机动态特性参数辨识方法

A Methodology for Aero-engine Dynamic Characteristic Parameter Identification based on SSA-NARX

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 values δ_(max) 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) of validation results corresponding to parameters N_1,N_(2) and EGT are 0.29,0.18 and 1.50 respectively.The accuracy,real-time performance and robustness of the model meet the requirements for simulation.