连分式扩充的粒子群神经网络压气机特性重构方法

对航空发动机压气机原始二维等转速线的数据点进行连分式扩充,通过两次网络训练,增加转速特性数据,在三维空间中进行BP(back propagation)网络模型重构.根据压气机特性数据空间分布的特点,引入压力比函数,调整计算区域,定义网络的输入输出数据,利用试探法确定隐含层维数.采用基于趋利避害原则的粒子群算法对网络的初始权值和阈值进行优化,建立了压气机压比和效率特性的整体代理模型.最后以某型发动机的低压压气机为例进行了压气机特性模型的重构.通过模型的校核与验证表明:采用这种方法建立的模型精度较高,优于普遍采用的传统二维插值方法和普通BP神经网络模型.最终建立的重构模型对于采用选配法、坐标法和部件法等以压气机通用特性曲线为基础的发动机模型的求解,可提高计算精度和迭代速度,具有一定的工程应用价值.

Neural network reconstruction method of compressor characteristics with continued fraction expanded data

Original two-dimensional constant rotating speed curve data points of aero-engine compressor were expanded using continued fraction interpolation.Rotating speed characteristic data through twice network training was increased and BP(back progagation) neural network model reconstruction in three-dimensional space was performed.According to the space distribution of compressor characteristics data,pressure ratio function was introduced,computational domain was adjusted,input and output data were defined and dimension of hidden layer was determined by cut-and-try method.Particle swarm optimizing algorithm based on seek advantage and avoid disadvantage principle was used to optimize the initial weight and threshold of neural network,and to establish the integral surrogate model of compressors was taken pressure ratio and efficiency characteristics.In the end,some aero-engine low pressure compressor for example to conduct the model reconstruction method.The check and verification results indicate the surrogate model established by the method proposed is more accurate than traditional two-dimensional interpolation and common BP neural network model.The reconstruction model can increase computational accuracy and boost iteration speed when used in solving aero-engine mathematic model based on general characteristic curve of compressor such as trial and error procedure,coordinate method and components method,therefore it is of engineering value.