Aircraft parameter estimation using a novel hybrid Luus–Jaakola/Hooke–Jeeves neural-network based optimization technique

Abstract

The neural network–based aircraft parameter estimation techniques have gained prominence in the last decade. Neuro–Gauss–Newton (NGN) technique is a widely used neural network–based algorithm. Although the NGN technique is useful for aircraft parameter estimation, it has some limitations. The present study is motivated by the limitations of the NGN method in estimating aircraft parameters, and hence, a new hybrid Luus–Jaakola/Hooke–Jeeves (LJ/HJ) method is proposed. The hybrid LJ/HJ method is based on two direct search methods: the Luus–Jaakola (LJ) method and the Hooke–Jeeves (HJ) method. In the hybrid LJ/HJ method, a global search of the minima of the error cost function is performed using the Luus–Jaakola method in the first phase. After that, the final solution is obtained using the Hooke–Jeeves method. The parameters estimated by the hybrid LJ/HJ method are compared with those estimated using the NGN method. The results obtained from the hybrid method are accurate, and unlike gradient-based optimization methods, the convergence issues are not observed during the estimation of parameters.