基于改进果蝇算法优化的GRNN航空发动机排气温度预测模型

Aero-engine exhaust gas temperature prediction model based on IFOA-GRNN

利用广义回归神经网络(GRNN)良好的非线性映射能力,对航空发动机排气温度(EGT)进行预测。由于GRNN的预测性能受宽度系数的影响,因此采用改进的果蝇算法优化广义回归神经网络(IFOAGRNN),并用优化后的GRNN对航空发动机的EGT进行预测。以某发动机为案例,选取相关参数作为预测模型的输入变量,EGT作为预测模型的输出变量。在相同的样本分配下,将FOA-GRNN(fruit fly optimization algorithm to optimize GRNN)、GRNN、自回归预测模型和优化的支持向量回归机作为对比算法。分析结果表明:IFOA-GRNN的收敛精度高于FOA-GRNN;IFOA-GRNN对EGT预测的平均相对误差为2.47%、拟合优度为0.850 6,其预测效果均优于其他对比算法;同时,IFOA-GRNN对噪声的敏感性也低于其他对比算法。

General regression neural network (GRNN) has a good nonlinear mapping ability.So exhaust gas temperature (EGT) is predicted by GRNN.But, its accuracy of prediction is affected by the width coefficient of GRNN.To address the problem,the GRNN optimized by the improved fruit fly optimization algorithm (IFOA-GRNN) was proposed. And it was used to predict EGT.Taking the engine as an example, some parameters were taken as input variables and EGT taken as output variable of prediction models.The forecast results of IFOA-GRNN, FOA-GRNN(fruit fly optimization algorithm to optimize GRNN), GRNN,auto-regressive and optimized support vector regression were compared under the same training samples and testing samples.The experiment results showed that the convergence accuracy of IFOA-GRNN was higher than FOA-GRNN. Average relative error of IFOA-GRNN for EGT prediction was 2.47%,and the goodness of fit was 0.8506, the prediction effect of IFOA-GRNN was better than other comparison algorithms.And it was more accurate than other methods in the prediction of aero-engine exhaust gas temperature under noisy and no-noise conditions.