| Due to the strong nonlinearity of dynamics,the uncertainty of model and environment,and various complex flight constraints,the trajectory optimization problem for the powered phase is very complex.Therefore,the reference trajectory in practice is often obtained by offline planning so far.However,because of the influence of modeling error and flight perturbation,there is always a certain difference between the performance indexes of the offline reference trajectory and the real flight trajectory.And in the case of large perturbation and extreme uncertainty such as propulsion system failure,the offline trajectory may lose its reference value.Under the pursuits of strong adaptability,high reliability and rapid response capability for the space launch mission,the bottleneck of poor real-time performance of offline trajectory optimization methods gradually emerges.In contrast,updating reference trajectory online can further exploit the performance of vehicles and improve the mission execution ability,and the online trajectory optimization thereby has become one of the difficult problems to be solved urgently in the field of advanced guidance and control technology.With the launch vehicle as the research object,based on traditional offline trajectory optimization methods,this thesis proposes an online trajectory generation method using the neural network for the powered phase,which can show a certain degree of adaptability to the deviations of thrust and aerodynamic parameters.The main research contents include:(1)The trajectory optimization model of the launch vehicle’s powered phase is established.The research object of this thesis is a three-stage solid launch vehicle.The motion equation for the powered phase is described in the launch inertial coordinate system.Then,the original trajectory optimal control problem is formulated with the maximum terminal speed as the performance index.And for the third stage flight in vacuum,the two-point boundary value problem is derived.In addition,considering the deviations of thrust and aerodynamic parameters,the corresponding deviation model is given.These work provide the model basis for the following research of the trajectory optimization methods.(2)The offline nonlinear optimization methods to get reference trajectory is studied.The hpadaptive pseudospectral method,immune clonal selection algorithm and sequential quadratic programming algorithm are used as the discretization tool,initial value generation tool and precise optimization tool,respectively.These three methods form an offline nonlinear optimization framework to obtain the trajectory samples for neural network.To verify the effectiveness of the above methods,under the nominal flight condition,the whole optimal trajectory of the powered phase is obtained by solving the original trajectory optimal control problem,and the third stage’s optimal trajectory is obtained by solving the two-point boundary value problem.(3)An online trajectory generation method based on the neural network is proposed.Considering the parameter deviation model,a large number of optimal reference trajectories under different non-nominal conditions are solved offline by the nonlinear optimization framework mentioned before.Taking these trajectories as samples,two multilayer feedforward neural networks called "state-control" network and "state-costate" network are trained offline.In the process of online application,the "state-control" network is applied in the first and second stage flight,and the "state-costate" is applied in the third stage flight.The inputs of the two networks are real-time flight states.The outputs of the two networks are control variables and costate variables at the current time,respectively.And taking the costate variables as the good initial values,the two-point boundary value problem can be solved online quickly to obtain the precise optimal control variables for the time-to-go,so then the cumulative error of trajectory can be corrected.Finally,the effectiveness of the above scheme is verified by simulations.The results show that while ensuring the real-time and accuracy,the scheme has a certain adaptability to the deviations of thrust and aerodynamics,which can meet the requirement of online application. |
