Aerospace Trajectory Design And Optimization Based On Neural Network Methods

Interstellar orbit transfer problem is a very important part of deep space exploration.In the interplanetary mission,the past often required a large incremental speed to achieve the orbit transfer,while the small thrust propulsion system compared to the traditional chemical thrust propulsion system due to the advantages of high specific impulse,small volume and ignitable multiple times which has shown the great potential for applications.The application of small thrust propulsion system has brought great rewards for deep space exploration as well as put forward new requirements and challenges for orbital transfer optimization technology.This dissertation aims at the trajectory design and optimization of small-thrust interstellar orbit transfer problem,and proposes a novel trajectory design and optimization method based on RBF neural network fitting and a neural network relying on augmented Lagrange multiplier method.Combining RBF neural network with small thrust trajectory to simulate the approximation to establish the orbit relationship between the state and time to avoid the traditional curve fitting strategy to solve the time constraints,the use of simulated annealing algorithm to quickly calculate the global optimal solution.After transforming the solved optimization results to get the initial value of the discrete model of pulse,the augmented Lagrange multiplier method was used to optimize the neural network and obtain the final result of the introduction of the acceleration path constraints,which completed the design of the shortest time orbit transfer and the most fuel-efficient orbit transfer problem with fixed transfer time.The proposed method was validated by taking the orbital transition of Earth to Mars as an example.The examples show that this method can effectively solve the problem of small-thrust interstellar orbit transfer.The neural network-based design optimization method is compared with the simplex method and the indirect method.It is concluded that there are many local minimum points in the small-thrust orbit transfer problem,the simplex is trapped in the local minimum point and the distance is far away from the optimal point.The indirect method is complicated to operate and requires not only complicated firstorder necessary conditions,but also requires a highly skillful method to solve the two-point,multi-point boundary value problem.Once the indirect method can be solved,the results are very close to the optimal solution.The results using the RBF neural network fitting and a neural network relying on augmented Lagrange multiplier method all obtain the approximate global optimum.