Data-driven Modeling Research For Aircraft

Modeling is the basis of simulation experiments for aircrafts.Because of the complex intrinsic mechanism of aircrafts,it shows strong non-linearity,strong coupling and time-varying characteristics in flight process,and it is difficult for the mathematical models established by traditional mechanism modeling methods to meet application requirements.The development of intelligent technology provides a theoretical basis for non-linear system modeling.Researchers can obtain a large amount of test data from the hardware-in-the-loop simulation experiments and flight tests.These data provide a data basis for data-driven modeling of aircraft.Taking the altitude control system of small unmanned helicopter,electric actuator and guided bomb as the main research objects,this paper mainly studies the modeling problems in online and offline usage scenarios,and carries out the theoretical research and method innovation of aircraft data-driven modeling.The main work is as follows:(1)In this paper,the single-input single-output nonlinear system modeling problem in online usage scenario is studied,and a dual-neural network modeling method is designed for complex nonlinear system modeling of small unmanned helicopter.Based on the traditional neural network modeling method,the error compensation model is introduced and the height control identification model of small unmanned helicopter based on BP network is established.Finally,the effectiveness and feasibility of this method are verified by simulation experiments compared with traditional modeling methods.The simulation results show that the approximation accuracy and generalization ability of the model are greatly improved by using the method of double neural network modeling.(2)Aiming at the modeling problems of SISO system in offline usage scenario,the cuckoo search algorithm is improved and used to optimize the training process of neural network,and a neural network identification model of the actuator is established.The algorithm makes Levy flight mechanism adaptive by dynamically adjusting the step size.Through the simulation experiment,compared with the traditional algorithm,the validity and feasibility of this method are verified.The simulation results show that the improved cuckoo search algorithm can effectively balance the global search and local search,and effectively improve the convergence speed and optimization accuracy.(3)Aiming at the modeling of MIMO nonlinear system in offline usage scenario,a data-driven hybrid modeling method is proposed for guided bomb.Based on the actual flight test data,considering the uncertainties and nonlinear factors in the guided bomb model,the hybrid model of guided bomb is established by combining the neural network with mechanism modeling.This method can avoid the problems existing in the mechanism modeling of guided bomb and traditional data-driven modeling methods.It also provides a solution for the modeling of complex nonlinear dynamic systems in offline usage scenarios.Finally,the feasibility of the method and the credibility of the model are verified by simulation experiments and statistical analysis.