This paper proposes an intelligent control method based on Soft Actor-Critic (SAC) to address uncertainties faced by flight vehicles during flight. The method effectively reduces aerodynamic loads and enhances the reliability of structural strength under significant wind disturbances. A specific launch vehicle is taken as the research subject, and its dynamic model is established. A deep reinforcement learning (DRL) framework suitable for the attitude control problem is constructed, along with a corresponding training environment. A segmented reward function is designed: the initial stage emphasizes tracking accuracy, the middle stage, with a detrimental effect due to the high-altitude wind region, focuses on load relief, and the final stage gradually resumes following tracking accuracy on the basis of maintaining the effect of load relief. The reward function dynamically switches between stages using a time factor. The improved SAC algorithm is employed to train the agent over multiple epochs, ultimately resulting in an intelligent load relief attitude controller applicable to the launch vehicle. Simulation experiments demonstrate that this method effectively solves the attitude control problem under random wind disturbances, particularly reducing the aerodynamic loads of launch vehicles in the high-altitude wind region.

This paper proposed a multi-view surface array to enlarge the field-of-view (FOV) from 45° × 45° to 72° × 75° and improve the estimation precision of guidance information. First, based on circular and rectangular FOV sensors, the superposition characteristics of FOV in the +-shaped layout and the X-shaped layout were explored and analyzed. Secondly, normalization processing was applied to obtain the equivalent measurement value of the central sensor and the corresponding error distribution. Based on such measurement value and error distribution, the filtering model could be constructed, which could effectively solve the problem of observation number variation caused by multiple sensors. Thirdly, a multivariate iterated extended Kalman filter (MIEKF) was proposed to make full use of multiple measurements. By iterating multiple unequal observations and making full use of the known error distribution information, the noise of filtered data was found to be effectively reduced and the estimation precision of guidance information was improved. Finally, based on a 6-DOF trajectory simulation, the correctness and effectiveness of the proposed method were verified. Simulation results show that MIEKF can improve the estimation accuracy of the line-of-sight (LOS) angle by at least 30% and the estimation accuracy of the LOS angle rate by nearly 80% compared with EKF.

At present, deep convolutional neural network (CNN) has been successfully applied to synthetic aperture radar (SAR) target recognition, and achieved good recognition effect. Compared with traditional methods, the recognition performance has been significantly improved. However, in practical applications, the resources of data processing platform are very limited, the computation and the memory cost of deep convolutional neural network are high. These two factors hinder its smooth deployment on embedded devices. This paper proposed a lightweight neural network design strategy combined with knowledge distillation for target recognition. First, a convolutional network model is designed based on the improved inverted residual structure, and a lightweight neural network is obtained, which is used as a student network. Then, the teacher network (a well-trained deep network model) is used to perform knowledge distillation, which affects the student network. training to improve the recognition accuracy. Finally, the trained student network is used to complete the 10-category target recognition in the MSTAR dataset.

In the existing impact time control guidance (ITCG) laws for moving-targets, the effects of time-varying velocity caused by aerodynamics and gravity cannot be effectively considered. Therefore, an ITCG with field-of-view (FOV) constraints based on biased proportional navigation guidance (PNG) is developed in this paper. The remaining flight time (time-to-go) estimation method is derived considering aerodynamic force and gravity. The number of differential equations is reduced and the integration step is increased by changing the integral variable, which makes it possible to obtain time-to-go through integration. An impact time controller with FOV constraints is proposed by analyzing the influence of the biased term on time-to-go and FOV constraint. Then, numerical simulations are performed to verify the correctness and superiority of the method.

For cruise missiles, this paper discusses finite-time spatiotemporal cooperative guidance with input saturation constraints. In order to improve the robustness, the method of distributed cooperative control is used. Based on the finite-time consensus theory, the longitudinal cooperative guidance instructions are derived to ensure that all of the missiles are able to attack the target at the same time in the line-of-sight (LOS) direction. An asymmetric saturation function is used to describe the input of the guidance system to satisfy the input saturation constraint. The normal cooperative guidance is derived to ensure missiles can attack at expected impact angles in the normal direction of the LOS. Also, the stability of the collaborative control has been demonstrated. Finally, the performance and correctness of the presented method are verified by simulation experiments.