Regarding the large-range variations of model parameters caused by configuration switching of high-speed morphing vehicle， a configuration switching flight vehicle attitude control method based on time-delay embedded dynamic mode decomposition （DMD） is proposed， in which a lifted state that includes the configuration state is firstly established via time-delay embedding， and a data-driven modeling procedure is used to realize the representation in a unified manner for attitude dynamics over multiple configurations， and a DMD-oriented prediction model suitable for controller synthesis is obtained. A backstepping-based angle-control loop is then designed so that the controller is required only to regulate the angular-rate dynamics to track a virtual nominal angular-rate command. Finally， a DMD-MPC controller with a cascaded architecture is developed. Regarding the features of relatively high prediction precision in finite time domain but prediction that diverges in infinite time domain， model predictive control is employed and closed-loop stability is verified. The simulation results demonstrate that the proposed DMD model achieves high prediction precision in representative morphing maneuvers， and satisfactory attitude tracking performance can be achieved by the DMD-MPC controller during configuration switching.
To address the issue of challenges in unmanned aerial vehicle trajectory planning that include strong dependence on initial conditions， difficulty in balancing global and local search capabilities， and susceptibility to premature convergence under complex constraints， a trajectory planning algorithm based on an improved hierarchical sparrow search algorithm is proposed. During the phase of population initialization， an improved chaotic mechanism combining the Logistic with Tent maps is introduced to enhance the diversity of initial solutions and reduce sensitivity to initial settings， and an improved Lévy flight strategy based on gradient control parameters is incorporated during the search process to achieve dynamic step-size adaptive adjustment， thereby global exploration and local exploitation are effectively balanced across different search phases. A two-layer collaborative optimization framework combining upper-layer global search with lower-layer local refinement is further established， in which the upper layer is used for global region search through the improved sparrow search algorithm， while the lower layer is responsible for high-precision local optimization near candidate solutions to enhance convergence quality and trajectory smoothness. This dual-layer structure is shown to significantly improve robustness and optimization performance in multi-constraint environments through adaptive information exchange. Finally， the effectiveness and superiority of the proposed method are validated through simulation and experiments.
According to heterogeneous aerial vehicle clusters task allocation under typical complex contraints， a task allocation method based on the Fisher market clearing （FMC） mechanism is proposed. By taking constraints such as time and task relationships under consideration， a mathematical model of heterogeneous aerial vehicle clusters task allocation that incorporates temporal window constraints with task execution relationships is established. Based on the FMC mechanism， the task allocation problem is mapped to the process of finding market equilibrium prices， and a proportional response algorithm is adopted to solve the task allocation matrix， and a discretization strategy is integrated to generate the final task allocation scheme. Simulation results demonstrate that compared with the classical genetic algorithm and particle swarm optimization， the proposed FMC-based method shows superior comprehensive performance in terms of computational efficiency and solution quality， and the effectiveness of the proposed method for heterogeneous resource task allocation in different scenarios is validated.
Regarding the limitations of traditional swarm intelligence optimization algorithms in three-dimensional path planning for unmanned aerial vehicle （UAV） under complex constraints， a novel swarm intelligence optimization algorithm capable of escaping local optima and improving convergence performance is proposed. Firstly， by comprehensively considering multiple constraints including path length， collision avoidance， risk zones， flight altitude and trajectory smoothness， the three-dimensional path planning issues for UAV is formulated as a multi-objective optimization problem， and then a tri-domain dynamic evolutionary algorithm （TDDEA） is designed. TDDEA divides the optimization population into three categories： global navigation domain， local search domain and roaming exploration domain， which achieves complementary advantages among domains and full utilization of limited information through individual division and dynamic migration mechanisms， and enhances the ability of escaping local optima， and improves planning accuracy and robustness. The simulation experiments are implemented in three-dimensional mountainous constraint environments with two different complexity levels， and TDDEA is compared with the genetic algorithm （GA） and holistic swarm optimization （HSO）. The results demonstrate that the proposed algorithm outperforms GA and HSO in both the quality and stability of generated paths. The requirements of safe， efficient and continuous UAV path planning can be met in complex environments， and the proposed algorithm has strong engineering application value.
To address the issue of decreased orbit determination accuracy during satellite maneuvers， an adaptive process noise unscented particle filter （APNUPF） algorithm is proposed. This algorithm is based on the new information， in which the process noise can be adaptively adjusted during both non-maneuvering and maneuvering phases， and the estimation accuracy degradation caused by fixed noise parameters is resolved and the filter is enabled to more accurately characterize the system's dynamic state. Under Gaussian noise， APNUPF demonstrates superior orbit determination accuracy during maneuvers， which is compared with traditional unscented particle filters， and can maintain high-precision orbit determination even with abrupt state changes. The results of extensive Monte Carlo simulations show that the algorithm's root mean square error （RMSE） is only by 1.35 meters during maneuvers. Furthermore， in order to fulfill practical engineering requirements， the robustness and reliability of the algorithm are comprehensively verified for satellite maneuver orbit determination under non-Gaussian noise conditions. Through simulations， the algorithm's RMSE is validated by 1.42 meters during maneuvers in this scenario， which significantly outperform the extended Kalman filter.
Regarding the requirements of large-scale， highly dynamic， and heterogeneous unmanned swarm collaborative strikes against multiple targets， a heterogeneous unmanned swarms collaborative task allocation method is proposed， which is oriented to communication-constrained and load-imbalanced situations. Firstly， by focusing on the needs of multi-target concurrency and multi-swarm coordination strike and involving resource constraints， task requirement constraints， and time-window constraints for time-sensitive targets， a many-to-many bidirectional bidding decision model is established， which aims to minimize total allocation costs. Secondly， aiming at resolving the heavy communication overhead and uneven task distribution of the traditional contract net protocol （CNP）， an improved CNP algorithm is developed for effectively reducing communication volume and evening the load distribution， which introduces a probabilistic withholding mechanism and a load-balancing strategy to refine the decision logic in the bidding phase. Finally， the simulation results demonstrate that the proposed algorithm can efficiently handle task allocation for unmanned swarm strike missions and achieves significant optimization in both communication efficiency and load balancing.
Due to the nonlinearity and strong coupling caused by parameter uncertainties and external disturbances in quadrotor UAV systems， a compound control strategy is proposed， which integrates deep reinforcement learning with backstepping sliding mode active disturbance rejection technology. A backstepping sliding mode active disturbance rejection controller is designed to improve the robustness of the system against parameter perturbations and external disturbances. Meanwhile， a finite-time reduced-order state observer is established to accelerate the anti-disturbance response speed of the system within a finite time， and the finite-time convergence of the observer is proven through theoretical derivation procedure. On this basis， a deep deterministic policy gradient agent is introduced to realize the autonomous learning and dynamic online optimization of controller parameters. In the design process， the state space and action space are reasonably defined to achieve the global optimization of the system control performance. The simulation results show that compared with the traditional active disturbance rejection control scheme with fixed parameters， the trajectory tracking precision and anti-disturbance performance of the system are effectively improved by applying the proposed control strategy， and a better comprehensive control effect is presented.
To address the issue of formation maintenance difficulty and susceptibility to external wind disturbances in cooperative suspension transport systems composed of multiple quadrotor UAVs operating in complex environments， a distributed cooperative control scheme is proposed， which integrates disturbance observation with edge-consensus control. Firstly， regarding load control， a multi-body coupled dynamic model of the system is established， and a corresponding decoupling control law is developed to achieve accurate tracking of the payload center position and transportation trajectory. In order to resolve unpredictable wind disturbances in practical operations， a disturbance observer is introduced to estimate and actively compensate for external environmental disturbances in real time， thereby enhancing the system's disturbance rejection capability and trajectory-tracking robustness under complex meteorological conditions. At the formation coordination level， edge-consensus theory is incorporated to map the formation objective from the conventional node-state space to the edge-state space， and a distributed edge-state consensus controller is designed， which enables to maintain the prescribed formation configuration by only using relative observation information from neighboring UAVs in the system that effectively improves the distributed coordination capability of the system under directed communication topology and local information interaction constraints. The simulation results demonstrate that high-precision trajectory tracking and stable formation maintenance can be achieved by applying the proposed method for the cooperative suspension transport system in the presence of unknown disturbances.
Regarding the flight stability of quadrotor unmanned aerial vehicles affected by unknown external disturbances and model uncertainties， a single-neuron fractional-order extended state observer （SNFOESO） is proposed and applied to the attitude control system of a quadrotor UAV. The historical information is incorporated into state and disturbance estimation by exploiting the memory and nonlocal properties of fractional-order derivatives in this method that effectively overcomes the estimation limitations of the traditional extended state observer （ESO） under time-varying disturbances and model mismatches， thereby improving observation accuracy and robustness. In addition， a single-neuron model is introduced to adaptively adjust the fractional orders and observer gains， which avoids the subjectivity and limitations of manual parameter tuning and enables the observer to adapt to dynamic disturbances and operating conditions in real time. The simulation and experimental results demonstrate that the proposed method significantly enhances disturbance rejection and robustness， thereby verifying the effectiveness and engineering feasibility of the algorithm. The proposed approach can serve as a new solution for high-precision attitude control of quadrotor UAVs in complex environments.
According to the selection dilemma of key parameters in balancing dynamic tracking performance with control input chattering suppression in the design of backstepping sliding mode control for a hypersonic morphing vehicle （HMV）， an intelligent parameter tuning strategy is proposed. Firstly， based on the introduction of a linear extended state observer and a command filter， a backstepping sliding mode attitude controller is designed for the HMV. Subsequently， an improved grey wolf optimizer based on a pheromone mechanism is proposed to enhance the global search capability of the wolf pack in the solution space， which effectively avoids local optima. Meanwhile， a composite fitness function incorporating system state tracking errors with rudder angle is designed to transform parameter tuning into a multi-objective constrained optimization problem. The simulation results demonstrate that the optimal controller parameter combination can be automatically identified by using the proposed strategy， while the high-precision attitude tracking of flying vehicle is ensured， and control surface chattering of the rudder can be significantly reduced， which present strong robustness and engineering application value.
Regarding the control allocation problem of over-actuated spacecraft systems， a games-in-games framework is proposed， which combines the evolutionary game with the Stackelberg game and can realize the multi-objective optimization for control allocation and optimal motion control simultaneously. Firstly， the optimal weights of the multiple objectives of control allocation are solved by using the evolutionary stable strategy. Then， the control allocation module is token as a part of follower in the Stackelberg game， which transmits its allocation strategy to the motion control module of the leader. On this basis， the motion control law is designed by the leader through application of the Stackelberg strategy. The results of numerical simulations and semi-physical experiments show the feasibility of the proposed method.
To address the issue of difficulty on existing guidance laws in balancing rapid response and terminal accuracy in short-duration interception scenarios， an intelligent parameter-tuning method is proposed， which combines finite-time control theory with maximum entropy deep reinforcement learning. By taking the nonlinear interception model with first-order autopilot delay as the object， a gain-adaptive regulator is designed by taking the stochastic policy exploration advantages of the soft actor-critic algorithm， which perceives missile-target states in real time and dynamically schedules guidance parameters. The simulation results demonstrate that under the conditions of a 15° initial heading error， a 25 maximum overload constraint， and a 0.12 s time constant delay， the line-of-sight rate is driven to converge to zero within 3 s by using the new algorithm， and a precise interception with a miss distance of 0.026 m is achieved. The defects of poor robustness of guidance laws under the conditions of short-duration， large time-delay， and large-error are effectively overcome， and the adaptive capability and guidance precision of the system are significantly improved.
