基于ADE-ELM的涡轴发动机建模方法

提出了基于自适应微分进化-极端学习机(ADE-ELM)求解平衡方程的高精度涡轴发动机实时部件级模型建立方法.基于牛顿-拉夫逊(N-R)迭代模型,以迭代计算前模型平衡方程残差为输入,迭代收敛后平衡方程猜值修正量为输出,训练极端学习机,并采用自适应微分进化(ADE)算法优化极端学习机(ELM)参数,提高猜值修正量映射精度.ADE算法中采用sigmoid型自适应缩放因子,提高了微分进化算法的寻优能力.在涡轴发动机不同飞行状态下的测试结果表明,以N-R迭代算法模型为基准,基于ADE-ELM的发动机模型,最大建模误差约为一次通过算法的1/3,运算耗时约为一次通过算法的1/3,验证了算法的有效性. 

Turbo-shaft engine modeling method based on ADE-ELM

A modeling method based on adaptive differential evolution-extreme learning machine (ADE-ELM) was proposed to solve the balance equations for turbo-shaft engine models. The model achieved high accuracy and real time properties. The extreme learning machine (ELM) was trained by the data got from the Newton-Raphson (N-R) iteration model. The inputs were the residuals of balance equations before iteration, and the outputs were the adjustments of guess values after iteration. The parameters of the ELM were optimized by adaptive differential evolution (ADE) algorithm to enhance the mapping accuracy of the guess values adjustments. The sigmoid type adaptive scaling factor was adopted by ADE algorithm to improve its optimization ability. Taking the N-R iteration model as a criterion, the simulation results of ADE-ELM model under different flying states show that, the maximum modeling error is about 1/3 that of the single iteration per time step, and the time consumption is about 1/3 that of the single iteration per time step. The modeling method is valid.