A hierarchical cooperative flight control system and method for transition section of tilt-rotor unmanned aerial vehicle

Abstract

The application discloses a layered cooperative flight control system and method for a transition section of a tilt-rotor unmanned aerial vehicle, and relates to the field of intelligent flight control of tilt-rotor unmanned aerial vehicles. The system comprises a decision layer, an optimization layer and a stabilization layer. The decision layer generates a dynamic safety corridor and a strategy parameter vector based on a dynamics model, environmental perception information and a health state and a predicted state sequence of the tilt-rotor unmanned aerial vehicle. The optimization layer embeds the dynamic safety corridor as a time-varying constraint into an optimization problem by using a Tube robust nonlinear model predictive control method, combines the strategy parameter vector and a warning signal, and realizes decoupling of nominal performance optimization and robust safety guarantee through constraint tightening and robust invariant set design to obtain optimal control instructions and the predicted state sequence. The stabilization layer suppresses disturbances and determines tracking errors in real time according to the instructions and the sequence, and generates the warning signal. The application improves the autonomy, environmental adaptability and task reliability of the tilt-rotor unmanned aerial vehicle in the transition flight phase.

Description

Technical Field

[0001] This application relates to the field of intelligent flight control for tiltrotor unmanned aerial vehicles (UAVs), and in particular to a hierarchical cooperative flight control system and method for the transition phase of tiltrotor UAVs. Background Technology

[0002] Tiltrotor unmanned aerial vehicles (UAVs) have demonstrated great potential in both military and civilian fields due to their unique combined vertical takeoff and landing (VTOL) and high-speed cruise capabilities. However, their core technological challenges lie in the dynamic transition between two flight modes. Flight control during this transition process is a recognized technical challenge due to strong nonlinear dynamics, multiple strict constraints, and significant uncertainties, and has become a key bottleneck restricting their widespread application. During the transition phase, the aircraft's dynamic characteristics exhibit dramatic and continuous time-varying features: there is strong nonlinear aerodynamic coupling between the complex downwash generated by the rotor and the fixed airfoil, and this disturbance dynamically evolves with changes in airspeed and tilt angle; simultaneously, the tilt motion of the rotor assembly causes real-time changes in the overall mass and inertia distribution, triggering significant inertial coupling effects; the control efficiency and range of action of each control surface (including rotor thrust vector, aerodynamic control surfaces, and tilt mechanism) continuously shift throughout the transition process, forming a highly complex control allocation problem.

[0003] Faced with this complex controlled object, existing technologies exhibit significant limitations. Traditional control methods, such as gain-scheduled PID control or feedback linearization, inherently rely on linearized models or precise dynamic cancellation, making them ill-suited to handle such strong nonlinearity and time-varying characteristics. Furthermore, they cannot explicitly satisfy multiple types of hard constraints and are relatively fragile in their robustness to model errors and external disturbances. The current mainstream serial architecture of offline trajectory planning + online tracking separates planning and control, resulting in generated reference trajectories that may be infeasible due to a lack of consideration for real-time control capabilities and disturbances. When encountering unforeseen situations, the system can only respond passively, lacking proactive adaptation and replanning capabilities, making it difficult to achieve optimal overall performance and safety.

[0004] With the continuous advancement of robust control theory and optimization algorithms, Tube-based robust model predictive control has demonstrated significant theoretical advantages in handling uncertain systems. This type of method provides a rigorously provable boundary for the state error of the disturbed system by constructing a robust invariant set (i.e., a "tube"-shaped region) around the nominal trajectory. However, existing theories still have significant limitations when applied to the specific problem of the transition phase of tiltrotor UAVs: most research focuses on linear systems or relatively simple nonlinear systems; at the technical level, existing solutions often emphasize the robustness design of the control algorithm itself, failing to systematically coordinate it with higher-level online trajectory planning and mission decision-making; particularly in key challenges such as the online generation of dynamic safety boundaries, the real-time trade-off between robust performance and optimal performance, and implementation under limited computational resources, a complete and usable technical system has not yet been formed.

[0005] Furthermore, as the demand for tiltrotor UAVs to perform diverse tasks in complex environments increases, the requirements for the safety, smoothness, and autonomy of the transition process are constantly rising. Traditional flight control strategies that rely on pre-designed, fixed parameters, or fragmented plan-track architectures, are proving inadequate in the face of model uncertainties, sudden wind disturbances, and real-time mission changes. There is an urgent need to develop an intelligent flight control method capable of online perception, decision-making, and adaptation. However, existing technologies are clearly insufficient in supporting this comprehensive demand: they either lack stringent safety guarantees or are computationally complex enough to not operate in real time, making it difficult to provide both safe and high-performance autonomous control capabilities in dynamic environments.

[0006] Based on the above description, the main systemic technical challenges currently faced by the transition phase control system of tiltrotor UAVs are as follows: traditional methods struggle to handle the strong nonlinearity and multimodal dynamics unique to cross-domain flight, resulting in limited control accuracy and stability; the traditional planning-control separation architecture lacks real-time environmental adaptability, and trajectory generation is often infeasible in dynamic environments; advanced nonlinear optimization methods suffer from real-time bottlenecks due to excessive computational complexity; existing robust solutions either lack strict safety guarantees or excessively sacrifice performance for conservatism; and insufficient coordination between functional modules fails to form an intelligent and complete control system. These problems severely restrict the safety, smoothness, and mission adaptability of transition flight. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this application provides a hierarchical cooperative flight control system and method for the transition phase of tiltrotor UAVs, offering an innovative systematic solution for highly autonomous and reliable flight of tiltrotor UAVs.

[0008] To achieve the above objectives, this application provides the following solution: In the first aspect, this application provides a hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle, comprising: a decision layer, an optimization layer, and a stabilization layer; The decision layer is used to generate a dynamic safety corridor based on the dynamic model, environmental perception information, and health status of the tiltrotor UAV, combined with the predicted state sequence obtained by the optimization layer, through reachability set analysis and intersection operation, and to formulate a strategy parameter vector to guide performance trade-offs based on the mission status. The optimization layer employs the Tube robust nonlinear model predictive control method, embedding the dynamic safety corridor as a time-varying constraint into the optimization problem. Combined with the policy parameter vector and early warning signal, and through constraint tightening and robust invariant set design, it decouples nominal performance optimization from robust safety assurance, obtaining the optimal control command for the current moment and the predicted state sequence for the next time step. The predicted state sequence is fed back to the decision layer in real time through the outer loop. The stabilization layer is used to suppress disturbances and determine the tracking error in real time according to the optimal control command and the predicted state sequence, and to generate the warning signal when the tracking error is located at the internal safety boundary and the external constraint boundary; the warning signal is fed back to the optimization layer in real time through the inner loop; the internal safety boundary and the external constraint boundary are determined by the dynamic safety corridor.

[0009] Secondly, this application provides a hierarchical cooperative flight control method for the transition phase of a tiltrotor unmanned aerial vehicle (UAV), applied to the aforementioned hierarchical cooperative flight control system for the transition phase of a tiltrotor UAV; the method includes: Based on the dynamic model, environmental perception information, health status and predicted state sequence of tiltrotor UAV, a dynamic safety corridor is generated through reachability set analysis and intersection operation, and a strategy parameter vector guiding performance trade-offs is formulated according to the mission status; the predicted state sequence is fed back in real time through the outer loop. The Tube robust nonlinear model predictive control method is adopted, in which the dynamic safety corridor is embedded as a time-varying constraint optimization problem. By combining the policy parameter vector and the early warning signal, the nominal performance optimization and robust safety guarantee are decoupled through constraint tightening and robust invariant set design, so as to obtain the optimal control command at the current time and the predicted state sequence at the next time. The early warning signal is fed back in real time through the inner loop. Based on the optimal control command and the predicted state sequence, disturbances are suppressed in real time and the tracking error is determined; when the tracking error is located at the internal safety boundary and the external constraint boundary, the warning signal is generated; the internal safety boundary and the external constraint boundary are determined by the dynamic safety corridor.

[0010] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a hierarchical cooperative flight control system and method for the transition phase of tiltrotor unmanned aerial vehicles (UAVs). The constructed three-layer cooperative architecture (decision-planning-control) generates a dynamic safety corridor and strategy parameters at the upper layer (decision layer), performs real-time trajectory optimization within the corridor using Tube robust nonlinear model predictive control, and executes robust tracking and disturbance compensation at the lower layer (stabilization layer). A dual-loop information interaction mechanism is designed to achieve closed-loop cooperation. The dynamic safety corridor provides a provable safety state boundary, enabling deep integration of planning and control and supporting online replanning based on real-time state and disturbance observations. Tube robust design decouples complex optimization into efficient nominal optimization and robust compensation, combined with fast solution techniques to ensure airborne real-time performance. Through the dual-loop interaction mechanism, the system can intelligently adjust its control strategy based on environmental wind field, remaining energy, and mission requirements, autonomously balancing robustness, timeliness, and energy consumption. Compared to traditional serial or single controller solutions, this application decouples complex robust optimization problems and introduces dynamic safety boundary management and forward-looking early warning mechanisms, theoretically ensuring flight safety under bounded uncertainty. Ultimately, it forms a feasible intelligent control technology system that systematically solves the problems of safety, robustness, and adaptability in transition control, thereby improving the autonomy, environmental adaptability, and mission reliability of tiltrotor UAVs during the transition flight phase. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic flowchart of a hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle (UAV) provided in an embodiment of this application; Figure 2 A schematic diagram of the decision layer provided in an embodiment of this application; Figure 3 This is a schematic diagram of an optimization layer provided in an embodiment of this application; Figure 4 A schematic diagram of a stabilization layer provided in an embodiment of this application; Figure 5 This is a flowchart illustrating a hierarchical cooperative flight control method for the transition phase of a tiltrotor unmanned aerial vehicle (UAV) according to an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] Transition phase flight control technology, as the core of achieving safe mode transitions for tiltrotor UAVs, plays a crucial role in the entire flight system. Its main functions include: comprehensively processing multiple control inputs such as rotor thrust vector, aerodynamic control surfaces, and tilt mechanisms to achieve efficient and precise control allocation; achieving a safe and feasible flight trajectory while strictly adhering to multiple constraints such as the flight envelope and the physical limits of the actuators; and ultimately, effectively suppressing the influence of uncertainties such as aerodynamic interference and wind disturbances through a feedback mechanism, ensuring a smooth and stable transition of flight status along the desired trajectory. This technology not only directly determines the safety, speed, and ride quality of the transition process but is also a key enabling technology for unleashing the advantages of tiltrotor UAV configurations and expanding their mission capabilities. Therefore, to address the aforementioned shortcomings of existing technologies, developing an intelligent flight control method capable of online collaborative decision-making, planning, and execution, with proven safety, not only has significant theoretical innovation value but also has urgent practical significance for unleashing the full-envelope flight potential of tiltrotor UAVs and ensuring their reliable operation in complex missions. This control method needs to break through the limitations of traditional control architecture and algorithms, deeply integrate advanced theories such as robust predictive control, online optimization and hierarchical decision-making, and establish a complete intelligent control system from high-level strategy to low-level robust execution, so as to provide fundamental technical guarantee for the safe and efficient transition flight of tiltrotor UAVs.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] In one exemplary embodiment, this application provides a hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle (UAV), such as... Figure 1 As shown, it includes: a decision-making layer, an optimization layer, and a stabilization layer; The decision layer is used to generate dynamic safety corridors based on dynamic models, environmental perception information, and the health status of tiltrotor UAVs, combined with the predicted state sequence obtained from the optimization layer, through reachability set analysis and intersection operation, and to formulate strategy parameter vectors that guide performance trade-offs based on the mission status. The optimization layer employs the Tube robust nonlinear model predictive control method, embedding the dynamic safety corridor as a time-varying constraint into the optimization problem. By combining the policy parameter vector and early warning signal, and through constraint tightening and robust invariant set design, it decouples nominal performance optimization from robust safety assurance, obtaining the optimal control command at the current moment and the predicted state sequence for the next time moment. The predicted state sequence is fed back to the decision layer in real time through the outer loop. The stabilization layer is used to suppress disturbances and determine the tracking error in real time based on the optimal control command and the predicted state sequence. It is also used to generate early warning signals when the tracking error is located at the internal safety boundary and the external constraint boundary. The early warning signals are fed back to the optimization layer in real time through the inner loop. The internal safety boundary and the external constraint boundary are determined by the dynamic safety corridor.

[0017] Based on the above description, this application integrates robust predictive control theory and a hierarchical decision-making architecture to construct a three-layer collaborative intelligent control system of decision-making, planning, and control. This system generates dynamic safety corridors and strategy parameters through upper-level decision-making; performs real-time trajectory optimization within the corridor using Tube robust nonlinear model predictive control in the middle layer; and executes robust tracking and disturbance compensation in the lower layer. An internal and external dual-loop information interaction mechanism is designed to achieve closed-loop collaboration. Compared to traditional serial or single-controller schemes, this application, by decoupling the complex robust optimization problem and introducing dynamic safety boundary management and a forward-looking early warning mechanism, theoretically guarantees flight safety under bounded uncertainty. It systematically solves the challenges of safety, robustness, and adaptability in transition control, thus providing a novel solution for the highly autonomous and reliable flight of tiltrotor UAVs.

[0018] In one exemplary embodiment of this application, the decision-making layer, acting as the system core or highest command center, has core functions including safe corridor generation, strategy decision-making, and emergency response. It receives task instructions from external sources, environmental perception information (such as real-time wind fields), the health status of the UAV itself (such as remaining battery power), and predicted status feedback from the optimization layer. Through a three-stage progressive processing of reachability set analysis, safe corridor synthesis, and strategy decision-making, it obtains dynamic safe corridors and strategy parameters. Based on this, as... Figure 2 As shown, the decision-making layer provided in this application includes: an input module, a reachability set analysis module, a security corridor synthesis module, and a strategy decision-making module. The security corridor synthesis module is connected to the reachability set analysis module; the strategy decision-making module is connected to both the reachability set analysis module and the security corridor synthesis module.

[0019] The input module obtains user task instructions to determine the task status, and obtains environmental perception information (such as real-time wind field) and the health status of the drone itself (such as remaining battery power).

[0020] The reachability set analysis module is based on a computationally efficient dynamic model and uses the zonotope set operation method to prospectively determine the set of all reachable states from the current state within a future period of time, thus obtaining the reachability set. The safe corridor synthesis module performs intersection operations on the reachable set with geographical constraints (airspace boundaries, obstacles) and flight envelope constraints (minimum stall speed, maximum attitude angle, etc.), and generates safe corridor slices at each time step through conservative approximation. The safe corridor slices at each time step are connected along the time axis to construct a continuously evolving dynamic safe corridor. The strategy decision-making module, based on multi-objective optimization principles, comprehensively considers mission status, environmental perception information, and the health status of the tiltrotor UAV to determine the strategy parameter vector. This vector guides the lower layers in weighing different performance objectives (time, energy, robustness). The module also acquires the predicted state sequence from the outer loop feedback in real time to evaluate the effectiveness of the strategy parameter vector and adjust the dynamic safety corridor. Simultaneously, the strategy decision-making module receives predicted state feedback from the optimization layer, evaluates the effectiveness of the strategy, and makes forward-looking corridor adjustments, forming a closed-loop outer loop control.

[0021] Based on this, the decision-making layer outputs the dynamic security corridor and policy parameter vector to the optimization layer, while receiving the predicted state sequence from the optimization layer, forming an outer loop feedback to evaluate the effectiveness of the current policy and make forward-looking adjustments.

[0022] In an exemplary embodiment of this application, the optimization layer acts as a real-time planner. Based on the dynamic safety corridor and policy parameter vector obtained above, and the early warning signal fed back from the lower layer (i.e., the stable layer), it constructs an optimization problem for model predictive control and solves it using a solver to obtain the optimal control command at the current moment and the predicted state sequence for a future period. Based on this, as... Figure 3 As shown, the optimization layer includes a pipe model predictive controller module and a solver module. The pipe model predictive controller module is connected to the decision layer; the solver module is connected to the pipe model predictive controller module. The Tube Model Predictive Control (Tube MP) module employs the Tube Robust Nonlinear Model Predictive Control method, embedding the dynamic safety corridor as a time-varying constraint into the optimization problem to obtain a nonlinear optimization problem. At each iteration point, the solver module transforms the nonlinear optimization problem into a series of quadratic programming subproblems and uses the initial solution for hot start-up to obtain the predicted state and control commands. During the iteration process, the iteration step size is determined in real time through line search until the set conditions are met, and then the optimal control command at the current time and the predicted state sequence at the next time point are obtained.

[0023] In practical applications, the optimization layer, as the core planning and optimization engine, is responsible for real-time trajectory optimization and predictive control within the safety boundary defined by the decision-making layer. It receives dynamic safety corridors and policy parameter vectors from the upper layer (i.e., the decision-making layer), as well as early warning signals from the lower layer. Its core is a Tube robust nonlinear model predictive controller (i.e., the tube model predictive controller module). This controller constructs an innovative optimization problem: minimizing a comprehensive performance index composed of tracking error, control cost, and robustness cost, while satisfying the constraints of the nonlinear dynamic equations and input constraints. The weights of each component are dynamically adjusted by the policy parameters. Crucially, the optimization problem explicitly embeds the safety corridor as a time-varying state constraint and ensures robustness through constraint tightening techniques, including a pre-computed robust positive invariant set calculated offline.

[0024] Furthermore, to achieve real-time airborne solution, the optimization layer employs a high-efficiency solver, such as the Real-Time Iteration (RTI) method. This method transforms the complex nonlinear optimization problem into a series of rapidly solvable quadratic programming (QP) subproblems by linearizing the area near the current nominal trajectory. It utilizes the solution from the previous time step for a warm start and, combined with convergence judgment techniques, ensures rapid solution completion. The output of the optimization layer includes the optimal control command for the current time step and the predicted state sequence for a future period. Both are issued to the stabilization layer for execution, and the latter is fed back to the decision layer to form an outer-loop closed-loop control.

[0025] In an exemplary embodiment of this application, the stabilization layer, acting as a precise executor, integrates a disturbance observer and a robust feedback controller to suppress disturbances in real time and ensure that the tracking error remains strictly within the safety boundary. Simultaneously, it provides closed-loop feedback through a monitoring mechanism. Based on this, the stabilization layer, as the final execution unit, is responsible for tracking the predicted state sequence generated by the optimization layer—the nominal trajectory—with high precision and robustness, according to the optimal control command and predicted state sequence obtained from the optimization layer. Figure 4 As shown, the stabilization layer includes: an interference observer module, a robust controller module, a control allocation module, and a monitoring module. The interference observer module is connected to the optimization layer; the robust controller module is connected to the interference observer module; the control allocation module is connected to the robust controller module; and the monitoring module is connected to both the interference observer module and the optimization layer.

[0026] The disturbance observer module estimates the total disturbance in real time online based on the tracking error between the actual state and the predicted state sequence (i.e., the nominal state sequence) in order to capture the system's uncertainties and external wind disturbances.

[0027] The robust controller module (using sliding mode control law) designs a feedforward-feedback composite control based on the tracking error and the estimated total disturbance to determine the robust compensation control quantity; the robust compensation control quantity is superimposed on the optimal control command to obtain the total control command; Under physical limiting constraints, the control distribution module distributes the overall control command to each actuator (such as motor, control surface, tilting mechanism, etc.) according to the control distribution command at the current rotor tilt angle, while satisfying the physical limiting constraints of each actuator.

[0028] In addition to providing status information, the monitoring module also constructs a status recognition mechanism based on tracking error. This module normalizes the real-time tracking error of the stabilization layer and compares it with preset multi-level dynamic thresholds to accurately determine the system's operating status. When the error remains within the safety boundary of the safety corridor, it is determined to be a normal operating condition, maintaining the normal collaborative working mode. When the error exceeds the internal safety boundary but has not yet reached the external constraint boundary, it is identified as a moderate disturbance operating condition, generating an early warning signal containing disturbance characteristics and parameter adjustment suggestions. When the error further exceeds the external constraint boundary, it is immediately determined to be a major abnormal operating condition, triggering the highest-level emergency warning and activating the corresponding safety response plan.

[0029] The three core operational layers (decision-making, optimization, and stabilization) collaborate closely through standardized data interfaces and calling protocols. The decision-making layer, acting as the strategic command center, is responsible for generating safe corridors and making strategic decisions, providing the system with forward-looking safety boundaries and performance guidance. The optimization layer, as the tactical planning engine, performs real-time trajectory optimization within the safety boundaries based on Tube robust NMPC, generating optimal control commands and predicted states, and feeding them back to the decision-making layer to form an outer closed loop. The stabilization layer, as the precise execution unit, completes the final control commands through robust tracking control and control allocation, and generates inner-loop early warning signals to feed back to the optimization layer through a monitoring module. This layered architecture ensures both the modularity and scalability of the system, while the dual-loop feedback mechanism ensures efficient and secure data processing.

[0030] In an exemplary embodiment of this application, based on the above description, the decision-making layer, as the strategic planning center, focuses on constructing a dynamic and safe flight corridor and formulating adaptive optimization strategies. The implementation process begins with the reachability set calculation module. This module, based on a validated tiltrotor dynamics model and mission instructions, employs an efficient centrally symmetric polyhedral set operation method. It uses a recursive formula to predict all possible arrival states at a given time in the future, providing a rigorous mathematical foundation for safety analysis. Subsequently, the safe corridor synthesis module loads preset three-dimensional airspace geographical constraints and aircraft physical envelope constraints in real time, performs online intersection operations with the calculated reachability set, and obtains the theoretically feasible region. To ensure the solvability of the downstream optimization problem, this module performs a conservative approximation on the theoretically feasible region: First, a large number of sampling time points are generated within the reachable set, and effective sampling time points that simultaneously satisfy all constraints are selected; second, the geometric center is calculated based on these effective sampling time points as the ellipsoid center, and the covariance matrix is ​​calculated to reflect the dispersion of the region in each direction; then, the expansion coefficient is determined by solving for the maximum Mahalanobis distance of all sampling time points, and the covariance matrix is ​​multiplied by this coefficient to obtain the ellipsoid shape matrix; finally, the time-varying ellipsoidal morphology safety corridor, characterized by the center vector and shape matrix, is output. Simultaneously, the strategy decision module continuously receives multi-source information such as aircraft status, real-time wind field intensity, and mission urgency, calculates mission urgency factors, environmental disturbance factors, and health status factors, and dynamically calculates and outputs a strategy weight vector based on a preset mapping function to balance the three objectives of time optimization, energy economy, and flight robustness. This module also receives predicted state sequences from the optimization layer. It evaluates the strategy by calculating a strategy effectiveness index and the minimum distance from the trajectory to the corridor boundary. When the index falls below a threshold or the trajectory is too close to the boundary, a forward-looking corridor adjustment is triggered, including corridor expansion, center shift, or contraction along the dominant error direction. Upon detecting a severe decline in health status or receiving a high-level warning, an emergency decision-making mechanism is activated, generating an emergency corridor pointing to an emergency landing point and conservative strategy parameters. This strategy vector and the dynamic safety corridor directly guide the performance trade-offs of the mid-level optimizer, ensuring the accurate implementation of the high-level intent. Based on this, the specific functions and implementation details of the decision-making layer in practical applications are as follows: (1) In the input module, the decision-making layer gathers four types of key input information, which constitute the data basis for its decision-making: (1-1) Flight targets from the mission management system include command parameters such as transition start and stop modes, target airspeed, altitude constraints, and mission priorities (e.g., time-optimal / energy-optimal / robust priority), which provide the final target guidance for the entire control process.

[0031] (1-2) Real-time external environment data (i.e., environmental perception information) obtained through airborne sensors or data links mainly include three-dimensional wind field estimation (wind speed, wind direction, turbulence intensity), air temperature and pressure, airspace restriction information, etc., which provide environmental constraints for reachability set calculation and strategy adjustment.

[0032] (1-3) Key parameters reflecting the aircraft’s own status (i.e., the health status of tiltrotor UAVs), including remaining battery power, actuator amplitude, sensor working status, current mass, etc., are used to assess the system’s remaining capacity and risk margin.

[0033] (1-4) The predicted state sequence from the optimization layer constitutes the core information of the outer loop feedback. This sequence enables the decision-making layer to foresee the flight trend in the future and realize the forward-looking assessment and dynamic adjustment of the effectiveness of the current strategy.

[0034] (2) In the safe corridor synthesis module, the core multi-constraint intersection operation is performed, combining dynamic reachability with external hard constraints to generate time-varying dynamic safe corridors. Based on this, to facilitate optimization, a conservative approximation is performed on the reachable state region, usually using a polyhedron to wrap it, forming a safe corridor slice at that moment. Therefore, the process of performing an intersection operation between the reachable set and geographical constraints and flight envelope constraints, and generating safe corridor slices at each moment through conservative approximation, includes: (2-1) Perform an intersection operation between the reachable set and the geographical constraints and flight envelope constraints to obtain the reachable state region at each time step; for example: Define geographic constraints for: It is usually represented as the union of multiple convex polyhedra, where, A convex polyhedron constrained by geographical constraints such as airspace boundaries, obstacle protection zones, and no-fly zones. The number of convex polyhedra. For the sampling time point, Let be the constraint matrix of the convex polyhedron. Let be the constraint boundary vector of the convex polyhedron. This defines the system state. Flight envelope constraints are defined. for: These are the core physical limitations that ensure flight safety, including the minimum stall speed. Maximum speed Maximum pitch angle Maximum roll angle High safety margin wait, At the current speed, The current pitch angle, Let the current roll angle be denoted as . Performing the intersection operation between the reachable set and the above constraints yields the reachable state region (i.e., the theoretically feasible region), which is: .in, For the reachable state region, It is an reachable set.

[0035] (2-2) Obtain the sampling time points of the reachable state region, and determine the covariance matrix and expansion coefficient based on the sampling time points and the geometric center of the sampling time points; In practical applications, the intersection operation in step (2-1) above yields the theoretically safe and reachable state region at each time step. The covariance matrix is ​​represented as follows: The covariance matrix reflects the degree of dispersion of the theoretically feasible region in various directions. In the formula, Let covariance matrix be the variance matrix. The sampling time points were randomly selected. The geometric center of the sampling time point, This is the transpose of the matrix. This represents the total number of sampling time points within the reachable state area.

[0036] Furthermore, to obtain a conservative ellipsoid encompassing all sampling time points, the covariance matrix needs to be dilated. Based on this, the resulting dilation coefficient is expressed as... : , This means taking the maximum value of k from 1 to M, which means traversing all valid sampling time points and taking the maximum value of the Mahalanobis distance among all sampling time points. This represents the inverse of the covariance matrix. This dilation coefficient ensures that all sampling time points lie within the ellipsoid.

[0037] (2-3) To satisfy the conservatism requirement that all sampling time points are within the ellipsoid, an expansion coefficient is used to expand the covariance matrix to obtain a shape matrix, thereby achieving a conservative approximation of the reachable state region; where the shape matrix is ​​denoted as... Q , .

[0038] Security corridor slices are generated based on a shape matrix. These security corridor slices are denoted as follows: , Sampling time point The safety corridor slice, Sampling time point The geometric center, Sampling time point The shape matrix, Shape matrix The inverse matrix, Representing state To the geometric center The squared Mahalanobis distance is less than or equal to 1, meaning the state lies inside the ellipsoid.

[0039] The safety corridor slices at each discrete sampling time point are then connected sequentially along the time axis to form a continuously evolving dynamic safety corridor that spans the entire transition period, denoted as [missing information]. : , For a moment The system state vector, As the center of the safety corridor, At the start time of each slice, For the decision-making layer, predict the time domain length for any given time. , and For adjacent discrete time nodes for t Time-shape matrix The inverse matrix, , Sampling time point The shape matrix, This is the smoothing coefficient. Among them, the corridor ellipsoid parameters... Obtained through linear interpolation, we have: , For the decision-making level's time step, Sampling time point The geometric center.

[0040] This dynamic safety corridor defines a dynamic safe course for the entire transition flight of the tiltrotor UAV. Any trajectory generated by the optimization layer must be constrained within this course. The dynamic safety corridor upgrades control from tracking a line to autonomously optimizing within the safe passage, significantly improving system robustness and mission adaptability.

[0041] (3) In the strategy decision-making module, based on the principle of multi-objective optimization, and taking into account the urgency of the task, the wind field environment, and the aircraft status, a strategy parameter weight vector is calculated to guide the lower layer in weighing different performance objectives. Simultaneously, this module receives predicted state feedback from the optimization layer, performs strategy effectiveness evaluation and forward-looking corridor adjustments, forming an outer-loop closed-loop control. Based on this, in this module, based on the principle of multi-objective optimization, the process of determining the strategy parameter vector, taking into account the task status, environmental perception information, and the health status of the tiltrotor UAV, includes: (3-1) Determine the urgency factor based on the time requirements of the task status, as follows: , As an urgency factor, The task requires a completion time. This is the theoretical minimum transition time. This is the theoretical maximum transition time.

[0042] (3-2) Based on the real-time wind field information in the environmental perception information, the environmental correction factor is determined as follows: , As an environmental correction factor, For wind speed, For wind direction, This is the influence coefficient.

[0043] (3-3) Based on the health status of the tiltrotor UAV, the state factors are determined as follows: , For state factors, For battery power, This refers to the nominal battery capacity. This refers to the state of health.

[0044] (3-4) The policy parameter vector is obtained using a mapping function based on the urgency factor, environmental correction factor, and state factor. The policy parameter vector is represented as follows: , , () represents the mapping function, which can be implemented using a linear weighted summation method. For example, the state weight matrix can be represented as... , , The baseline state weight matrix, , and All of these are weighting adjustment coefficients.

[0045] Furthermore, the strategy decision-making module receives the predicted state sequence from the optimization layer. The nominal trajectory is used to evaluate the effectiveness of the current strategy. Based on this, the process of adjusting the dynamic safety corridor can be described as follows: determining the distance between the predicted state in the predicted state sequence and the boundary of the current dynamic safety corridor; determining whether the distance is less than a preset safety threshold; when the distance is less than the preset safety threshold, expanding the current dynamic safety corridor along its dominant direction; when the distance is greater than the preset safety threshold, contracting the current dynamic safety corridor along its dominant direction. Therefore, in practical applications, this process can be implemented as follows: Define the strategy effectiveness index as , In order to be in k Time prediction k+i The state at any given moment, This is the sign after contraction, that is, at time [time]. Predicted time The nominal system status; For a moment Dynamic safety corridor pipe slices, The prediction time-domain steps reflect the proportion of trajectories generated by the policy that satisfy safety constraints within the prediction time domain. For each prediction time step, the distance from the predicted state to the corridor boundary is... , , This is the tightened shape matrix. The minimum monitoring distance is... , For each prediction time, the distance from the predicted state to the corridor boundary is calculated. If the minimum monitoring distance is less than a preset safety threshold... This indicates that the trajectory is too close to the corridor boundary, posing a risk.

[0046] When the strategy effectiveness index is less than the preset index, when the environmental state changes significantly, or when the minimum monitoring distance is less than the preset safety threshold, i.e. when the interference condition is moderate, the corridor adjustment is triggered.

[0047] When the trajectory gets too close to the boundary of the safety corridor, the safety corridor is expanded along the dominant direction to obtain the expanded shape matrix: , For expansion gain, , The dominant error direction is determined by the direction of the predicted trajectory. When the predicted trajectory continues to deviate to one side of the safety corridor, the center of the safety corridor can be shifted, resulting in: , The center of the shifted safety corridor The average value of the predicted state. When the trajectory is far from the safety corridor boundary and the performance indicators are excellent, the corridor can be appropriately narrowed to improve control accuracy, such as: , It is the identity matrix. The shrinkage coefficient must be used to ensure that all theoretically feasible states are still included after shrinkage.

[0048] When the system detects a serious anomaly, the strategy decision module activates the emergency decision-making mechanism, which includes: , This is the emergency strategy parameter vector, used to guide the optimization layer and the stabilization layer to perform conservative control under emergency conditions; For emergency landing, For crash probability estimation. Finally, an emergency safety corridor pointing to the emergency landing point and emergency strategy parameters are generated, with more conservative speed and attitude limits and a larger safety margin.

[0049] Finally, the decision-making layer sends the generated policy parameter vector and dynamic security corridor to the optimization layer.

[0050] In an exemplary embodiment of this application, based on the above description, the optimization layer, acting as a real-time trajectory planning engine, is responsible for solving the optimal and robustly guaranteed flight trajectory within the safety corridor. The core implementation involves constructing and solving a robust nonlinear model predictive control problem with constraint tightening. The optimization layer receives the dynamic safety corridor, policy weights, and warning signals from the upper layer (i.e., the decision layer), and adjusts the weights or replans based on the warning signals. Subsequently, a finite-time optimal control problem is assembled online, whose cost function integrates trajectory tracking accuracy, control energy consumption, control rate of change suppression, and terminal stability. The weights are dynamically injected from the policy vector. In the optimization layer, robust constraint processing is implemented: the original safety corridor is geometrically shrunk using a robust positive invariant ellipsoid obtained offline. The tightened state constraint shape matrix is ​​obtained by squaring the square root of the sum of the squares of the original shape matrix and the robust positive invariant shape matrix, while keeping the center unchanged. Similarly, the original control constraints are shrunk, and the tightened control boundary is obtained by adding and subtracting the maximum projection of the robust positive invariant set onto the control direction from the original boundary. When the optimization layer solves the nominal trajectory, it operates within this confined, conservative region, ensuring that the actual closed-loop trajectory remains strictly within the original safety boundary under feedback control. The solver module then employs a real-time iterative method to handle the nonlinear problem: First, it performs a warm-start and shift initialization using the optimal solution from the previous moment; it linearizes the nonlinear dynamics near the current trajectory, calculating the state Jacobian matrix and control Jacobian matrix; second, it approximates the original problem as a standard quadratic programming problem, constructing a quadratic objective function and linear constraint matrix; it quickly solves the quadratic programming subproblem using the interior-point method to obtain the search direction; then, it determines the appropriate step size through line search and updates the iteration points; this process is repeated until convergence. Finally, it performs forward integration on the nonlinear model to update the predicted trajectory. This process is iterative, outputting the current optimal control command and future state prediction in each cycle, achieving a balance between real-time rolling optimization and robust performance. Based on this, in practical applications, the implementation of the optimization layer can be described as follows: (1) Receive dynamic safety corridor and strategy parameter vectors from the decision-making layer, and early warning signals from the stability layer. Under normal operating conditions, the optimization layer continues to operate normally; however, under moderate disturbances, the optimization layer adjusts the state weight matrix as follows: , This is the adjusted state weight matrix. The original state weight matrix, The weight adjustment coefficient is used; when under major abnormal conditions, the optimization layer triggers online replanning, and the emergency nominal trajectory is solved by using the emergency safety corridor and emergency strategy parameters.

[0051] (2) Considering practical nonlinear systems ,in, For a moment The actual system state, To control the input, For a bounded disturbance, W This is a bounded set of perturbations. The nominal system (ignoring perturbations) is... , , This is the nominal system state vector at the current moment. This is the nominal control input vector at the current moment. And... , For robust positive invariant sets, ellipsoidal form is adopted. Its satisfaction , For Minkowski and operators, Let be the shape matrix of the robust positive invariant set ellipsoid. For the system matrix... and control matrix First, design the feedback gain. Make the closed-loop system matrix Schul stability is then obtained by solving the following linear matrix inequalities. : The estimation of the perturbation covariance matrix satisfy , This is the perturbation vector. All of the above can be obtained through offline calculation.

[0052] (3) Tube Robust NMPC Optimization Problem: To ensure the safety and reliability of the actual system under model uncertainty and external disturbances, the Tube Model Predictive Control (Tube MPC) module adopts a robust design strategy with tightened constraints. Within this contracted feasible region, an optimization problem is set: the objective function comprehensively considers trajectory tracking accuracy, control energy efficiency and terminal stability, while applying tightened state and input constraints.

[0053] To ensure the robustness of the actual system, it must operate within a tightened set of constraints. The initial state constraints include an ellipsoidal safety corridor issued by the decision-making level, denoted as... Tightened nominal state constraints satisfy For a robust positive invariant set in ellipsoidal form, the compacted ellipsoid shape matrix is: The center remains unchanged, that is The tightened nominal state constraints are as follows: The original control constraints are: Tightened nominal control constraints satisfy , For actual system control,u min This is the minimum value vector (lower bound) for controlling the input. u max Let be the maximum vector (upper bound) of the control input. For a robust positive invariant set in ellipsoidal form, the tightened control boundary is: , ,in .

[0054] To ensure that the obtained solution conforms to reality and is safe, constraints need to be set. The dynamic constraints are: , For at any time Predicted time The nominal system status, For at any time Predicted time The nominal system status, For at any time Predicted time The nominal control input, Let be the dynamic function of the nonlinear system. The initial state constraints are: The tightened state constraint is... , For at any time Predicted time The nominal system status, For a moment The center of the ellipsoid (center of the safety corridor). For a moment The inverse matrix of the tightened ellipsoid shape matrix. The tightening control constraints are: , For at any time Predicted time The nominal control input, This is the vector of minimum control values ​​after tightening. This represents the maximum vector of the tightened control quantity. The control rate of change constraint is... , For at any time Predicted time The nominal control input, To control the vector of minimum rate of change, To control the vector of maximum rate of change. Terminal constraint: ,in, This is the inverse of the tightened terminal cost matrix. , For the terminal cost matrix, Let the shape matrix be the robust positive invariant set ellipsoid. For at any time Predicted time The nominal system status (terminal status). This refers to the terminal target state.

[0055] At each sampling time, the following finite-time optimal control problem is constructed using comprehensive performance indicators: The decision variables are: , , For control sequences, contained in Time prediction from time At the time The nominal control input; It is a state sequence, contained in Time prediction from time At the time The nominal system status, For at any time Predicted time The nominal system state (initial state). The cost function is: ,in, and These are the state weight matrix and the control weight matrix, respectively, both dynamically adjusted by the policy parameters. For reference state trajectory, For reference control input, For at any time Predicted time The nominal control input, For at any time Predicted time The nominal system status, For the terminal cost matrix, The control rate of change weight matrix is ​​a positive definite matrix used to suppress drastic changes in the control variable. For all steps in the prediction time domain arrive Summation, To predict the number of time-domain steps, The squared weighted norm of a vector is defined as follows: , Let k be the rate of change of control at time k+i predicted at time k. To control the transpose of the rate of change vector, To control the rate of change weight matrix.

[0056] Efficient solver: The solver searches within this confined space to find the nominal trajectory and its corresponding nominal control sequence that are optimal under the nominal model.

[0057] To achieve real-time airborne solution, the optimization layer employs a real-time iterative method. This method transforms the complex nonlinear optimization problem into a series of quickly solvable quadratic programming subproblems by linearizing the area around the current nominal trajectory. Let the current iteration point be... , Representing the iteration For the first The state sequence of the next iteration includes the state from time [time]. At the time All nominal states; For the first The control sequence for the next iteration includes the sequence from time [time]. At the time All nominal control inputs. Linearize the dynamics at each point: ,in , For the first In the nth iteration The state Jacobian matrix of the step, For the first In the nth iteration The control Jacobian matrix of the step, For the first In the next iteration, at time... Predicted time The nominal condition, For the first In the next iteration, at time... Predicted time The nominal control input, For the first Step state deviation, For the first The control deviation of the step, For nonlinear system dynamics functions, For the dynamic function with respect to the state The partial derivatives, For the dynamic function of control The partial derivatives, For the point Differentiate at point 1. Define the state and control deviation as follows: , For at any time Predicted time The nominal condition, For at any time Predicted time The nominal control input, and the linearized dynamic constraints are: ,in To linearize the residuals, For the first In the next iteration, at time... Predicted time state, For the first Step state deviation, For the first In the nth iteration The state Jacobian matrix of the step, For the first In the nth iteration The control Jacobian matrix of the step.

[0058] The original problem is transformed into a problem concerning the control increment sequence. The problem of secondary planning, The nominal control sequence to be solved is... For the first The nominal control sequence for the next iteration. The objective function (quadratic form) is: The constraints (linear inequalities) are: ,in, For For the problem of minimizing decision variables, matrix The gradient vector is obtained by taking the second derivative of the cost function. The first derivative, This is the transpose of the gradient vector. To control the transpose of the incremental sequence, the constraint matrix... and constrain the right-hand side terms It consists of linearized dynamic constraints, state constraints, and control constraints.

[0059] To improve solution efficiency, a warm start is performed using the optimal solution from the previous time step, resulting in: ,in, For the first The initial nominal control sequence at each sampling time (the initial guess of the 0th iteration), which includes the previous time ( From the optimal solution predicted at time (i.e., the first time) Time to the The amount of control at any given moment The symbol representing the optimal solution. Represents the initial value. To predict the number of time-domain steps, the corresponding state trajectory is obtained through forward simulation: , , For at any time Predicted time The nominal initial state value (take the current actual state). For at any time Predicted time The nominal initial value, For at any time Predicted time The nominal initial value, For the first The initial control sequence at the sampling time is the first... The nominal control quantity of the step.

[0060] Solving the quadratic programming problem yields the control search direction. Then, the step size is determined through line search. : , For the state search direction, In order to be in Within the range, select the variable values ​​that minimize the objective function. Let be the cost function (optimization objective). Then, update the iteration points, with: , For the first The state sequence of the next iteration (the updated state sequence). For the first The control sequence for the next iteration (the updated control sequence).

[0061] The iteration stops if any of the following conditions are met: 1) , The minimum factor is set. To control the search direction The vector norm; 2) Reaching the maximum number of iterations ; 3) Cost function rate of decline ,in The set decline rate factor, For the first The cost function value of the next iteration. For the first The cost function value of the next iteration. To take the absolute value.

[0062] The final optimization layer will generate the optimal control command for the current moment. With the predicted state sequence This refers to the nominal state sequence, which is also sent to the stable layer. For predicting the state sequence (nominal state sequence), it contains time... Predicted from time At the time All predicted states, To predict the number of time-domain steps. For at any time Predicted time The optimal nominal control input is the first component of the optimal control sequence obtained by optimization.

[0063] In an exemplary embodiment of this application, based on the above description, the stabilization layer, as the execution and safeguarding unit, bears the heavy responsibility of high-precision tracking commands and resisting real-time interference. Its implementation begins with the nonlinear interference observer module. The nonlinear interference observer module estimates the equivalent concentrated disturbance online through the designed extended state observer equation, treating model uncertainty and external wind disturbance as a unified total disturbance for real-time estimation. The observer gain is configured according to the bandwidth parameter to ensure that the estimation error converges quickly. Based on this estimated value and the actual state error, the robust feedback controller module begins to work. The robust feedback controller module adopts a feedforward-feedback composite structure: the feedforward part calculates the feedforward compensation amount based on the disturbance estimate and the pseudo-inverse of the control matrix, directly canceling the observable disturbance; the feedback part is designed using a sliding mode control method, constructing a sliding mode surface and designing a sliding mode control law containing a saturation function. The sliding mode gain is adaptively adjusted online according to the disturbance estimate to ensure that the upper bound of uncertainty can be overcome, thereby theoretically guaranteeing that the tracking error is stably suppressed. The synchronously operating monitoring module normalizes the tracking error, calculates the Mahalanobis distance based on the ellipsoid shape matrix as the normalized error, and compares it with preset multi-level dynamic thresholds to accurately determine the system's operating status. When the tracking error exceeds the threshold, a corresponding level of early warning signal is generated, and the dominant error direction vector is calculated and transmitted to the decision-making and optimization layers. Under severe conditions, adaptive gain adjustment or emergency response plans are triggered. Finally, the control allocation module transforms the overall control command calculated by the stabilization layer into the specific actions of each actuator. The control allocation module constructs the relationship between virtual control commands and actual actuator commands based on the control efficiency matrix that varies with the tilt angle, and solves it using the weighted least squares method. When the actuator reaches its physical limit, a cascaded allocation strategy is initiated, limiting the amplitude of saturated actuators and reducing the problem for redistribution. While meeting physical limits such as rudder efficiency and speed, it accurately matches the overall control requirements, completing the final mapping from control law to physical action. Based on this, the implementation process of the stabilization layer can be described as follows: (1) Receive the nominal state sequence (i.e., the predicted state sequence) and the optimal control command from the optimization layer, as well as the actual state feedback from the sensor, to ensure that the stabilization layer can track the nominal state sequence generated by the optimization layer with high accuracy and high robustness through these data.

[0064] (2) The extended state observer designed in the disturbance observer module extends the total system disturbance (including model uncertainty and external disturbance) into a new state for real-time estimation, such as: First, we set up the extended state observer structure, considering a nonlinear system, we have: , For the nominal model, The total disturbance is unknown. Define the extended state. The augmented system is obtained as follows: , The system state derivative, For the extended state derivative, Given an unknown bounded function, design a linearly expanding state observer as follows: ,in, This is feedback on the actual status. and All are observer gain matrices. The system state estimated by the extended state observer. The total perturbation estimated by the extended state observer, The system state derivative estimated by the extended state observer, The total perturbation is estimated for the extended state observer. Then the observer poles are configured at... ,have: ,in, This is the observer bandwidth, typically taken as 5-10 times the controller bandwidth. for 3D identity matrix. Sampling time. The discretized form below is: , , for Constantly estimate the state. for Constantly estimate the state. for Constantly estimate disturbances. for Time-based disturbance estimation. State estimation error. With disturbance estimation error They are respectively It must satisfy: For bounded The estimation error is bounded and varies with Increases and decreases.

[0065] (3) In the robust controller module, a feedforward-feedback composite controller is designed based on the disturbance estimated by the extended state observer, as follows: First, define the tracking error as... e , , To optimize the nominal state issued by the layer, the error dynamic equation is: , For nominal system dynamics, To track the derivative of the error. Linearizing it near the nominal trajectory, we have: ,in , For nominal control, Let be the state Jacobian matrix at the nominal trajectory. Let be the control Jacobian matrix at the nominal trajectory.

[0066] Secondly, the sliding surface is designed as s. ,in, For the sliding surface parameter matrix, it is necessary to ensure Reversible. The selection of sliding surface parameters should ensure that the sliding motion... It possesses ideal dynamic characteristics. On the sliding surface, the equivalent control satisfies... The equivalent control is obtained by solving the problem. , The sliding mode control law is designed as follows: , ,in, For sliding mode gain, For boundary thickness, Let be a saturation function used to suppress chattering, then: In the formula, It is a symbolic function.

[0067] To ensure slip mode accessibility, the following must be met. , For robustness margin, For matrix norm, To estimate the disturbance, the boundary layer thickness is required. Balancing chatter suppression with tracking accuracy. Typically, the following value is used: ,in As the reference thickness, These are adaptive coefficients.

[0068] To improve adaptability to different disturbance intensities, adaptive gain adjustment of the sliding mode gain is adopted, namely: , This is the adaptive gain adjustment coefficient. The derivative of the sliding mode gain. The differential form of the sliding mode gain minimum / maximum is: , For the stabilization layer sampling time, for The synovial surface at any moment, for Sliding mode gain at time step for The sliding mode gain at any given time. When the sliding surface deviation is consistently greater than the boundary, the gain automatically increases; when the deviation is small, the gain decreases to avoid over-control.

[0069] Subsequently, feedforward compensation based on disturbance estimation is calculated. ,have: , This is the pseudo-inverse of the control matrix. This feedforward compensation... It directly counteracts the estimated disturbance, significantly reducing the burden on feedback control.

[0070] (3) The control allocation module distributes the synthesized total control command to each actuator (motor, control surface, tilt mechanism) according to the control allocation strategy under the current rotor tilt angle, while satisfying the physical limit constraints of each actuator. Based on this, the implementation process is as follows: First, synthesize the overall control command. ,have This overall control command can quickly counteract observable disturbances, suppress residual errors and unmodeled dynamics, and adjust the control strength according to real-time operating conditions.

[0071] Define the master control command With actual actuator instructions Relationship: ,in, To follow the tilt angle The variable control efficiency matrix. Control efficiency matrices for different operating conditions are obtained in advance, a lookup table is created, and the matrix is ​​retrieved online based on the tilt angle.

[0072] Subsequently, the weighted least squares method is used to solve the control allocation problem, resulting in: , ,in, This is the executor priority weight matrix. For preferred locations (such as locations with minimum energy consumption), the analytical solution in the unconstrained case is: .

[0073] When actuator limiting exists, a cascaded allocation strategy is used to calculate the unconstrained solution (i.e., the actual actuator instruction). ); detect and process saturated actuators, and define saturated sets. , For the unconstrained solution corresponding to the first The component subvectors of each executor For the first The maximum allowed instruction value (upper bound) for each actuator. For the first The minimum allowed instruction value (lower bound) for each actuator. For the executor index, if the saturated set If not empty, the saturation actuator will be fixed at the limit value: After reducing the control allocation issue, the remaining control commands are: ,in, Let be the efficiency matrix of the saturated actuator, and let be the efficiency matrix of the remaining actuators. Re-evaluating the remaining actuator allocation results, we have: , Here is the weight matrix for the remaining executors. For the preferred positions of the remaining actuators, The instruction is used to fix the saturation actuator after the limit value; check if the new solution meets the constraints, and if not, repeat the above steps. Finally, it is assigned to each actuator (motor, rudder, tilting mechanism, etc.).

[0074] (4) The monitoring module monitors the tracking error in real time based on the ellipsoidal tubular region, enabling accurate judgment of the system's operating status and generating corresponding early warning signals. Based on this, the implementation process of this module is as follows: First, based on the ellipsoidal tubular region and robust positive invariant sets (i.e., ellipsoidal form) Define the normalized tracking error as: : In the formula, This is the actual state. For the nominal state, this error metric has a clear geometric meaning, representing the Mahalanobis distance of the actual state from the nominal trajectory, taking into account the different allowable deviations in different state directions. The specific calculation process of the normalized error is as follows: first, obtain the tracking error vector calculated above; then... Perform Cholesky decomposition as , Perform Cholesky decomposition of the lower triangular matrix; then calculate , To track the error vector, the normalized error also needs to be decomposed to identify the dominant error channel. Perform eigenvalue decomposition: ,in The main axis direction (eigenvector matrix). For the first Eigenvectors (principal axis direction). It is an eigenvalue diagonal matrix. The square of the allowable deviation in the corresponding direction is used. The error vector is then projected onto each principal axis: , Let be the error vector projected onto the i-th principal axis direction. Let be the i-th principal axis direction. The normalized error components in each principal axis direction are: : Its dominant error channel is IMAX. The corresponding direction vector is Yes, the dominant direction requiring early warning is obtained, and this parameter is fed back to the decision-making and optimization layers along with the early warning signal.

[0075] Secondly, multiple levels of dynamic thresholds are set to accurately determine the system's operating status: internal safety boundary (warning threshold). : ,when At that time, the system was considered to be operating normally. This is the warning threshold coefficient, typically set to 0.85. External constraint boundary (critical threshold) : ,when At that time, the state was considered to be close to the pipe boundary, in a state of moderate disturbance. This is the critical threshold coefficient, typically taken as 0.95. Theoretically, it guarantees the boundary value. .when or At this point, the state has exceeded the theoretically guaranteed range, and the system may face the risk of instability. The threshold can be adaptively adjusted according to the flight phase as follows: , The original warning threshold, This is the dynamic threshold adjustment coefficient. This is the absolute value of the rate of change of tilt angle, meaning that the warning threshold should be appropriately relaxed when the tilt angle changes drastically.

[0076] Then, based on the normalized tracking error, the system operating status is identified and corresponding early warning signals are generated: 1) Normal operating conditions ( ): Maintaining the normal collaborative working mode, the stabilization layer independently completes disturbance suppression, the optimization layer performs rolling optimization according to a fixed cycle, and the decision-making layer keeps the current corridor unchanged; 2) Moderate interference conditions ( The stabilization layer sends an early warning to the optimization layer, the optimization layer adjusts the weight matrix according to the direction of the dominant error, and the decision layer evaluates the predicted trajectory and adjusts the safety corridor if necessary. 3) Critical operating conditions ( or The decision-making layer generates emergency corridors, the optimization layer triggers online replanning to solve for emergency trajectories, and the stabilization layer executes emergency control.

[0077] The stabilization layer ultimately outputs the actuator commands obtained after control allocation, as well as the corresponding early warning signals generated after identifying the system's operating status.

[0078] Based on the above description, this application breaks through the limitations of traditional control architectures and algorithms, deeply integrating advanced theories such as robust predictive control, online optimization, and hierarchical decision-making to establish a three-layer collaborative intelligent control architecture from high-level strategy to low-level robust execution: decision-making, planning, and control. The decision layer, as the core of the system, generates a forward-looking dynamic safety corridor based on accurate dynamic models and environmental perception information through reachability set analysis and intersection operations, and formulates a strategy parameter vector guiding performance trade-offs according to the task state. The optimization layer, as a real-time planner, adopts the Tube robust nonlinear model predictive control method, embedding the safety corridor as a time-varying constraint into the optimization problem. Through constraint tightening and robust invariant set design, it achieves decoupling between nominal performance optimization and robust safety assurance. The stabilization layer, as a precise executor, integrates a disturbance observer and a robust feedback controller, suppressing disturbances in real time and ensuring that the tracking error is strictly within the safety boundary, while providing closed-loop feedback through a monitoring mechanism. Furthermore, this application achieves dynamic closed-loop collaboration through internal and external dual-loop feedback. The outer loop feeds back the predicted trajectory from the optimization layer to the decision layer for proactive assessment of safety margins and triggers active replanning; the inner loop feeds back the tracking health information from the stability layer to the optimization layer, driving it to adjust weights or constraints online to enhance environmental adaptability. This dual-loop mechanism endows the system with closed-loop intelligence from perception to adjustment, enabling it to autonomously adjust control strategies based on real-time operating conditions.

[0079] Furthermore, under normal operating conditions, the intelligent control architecture provided in this application operates collaboratively at differentiated frequencies to form steady-state and efficient control. When faced with moderate disturbances, the inner loop early warning triggers the optimization layer to quickly adjust weights and trajectories, while the outer loop collaboratively relaxes the safety boundary to achieve dynamic rebalancing. When encountering major anomalies, the decision-making layer initiates an emergency response, switching to a conservative control mode to achieve a smooth system degradation while ensuring safety. Smooth switching logic between modes ensures control continuity.

[0080] Furthermore, to achieve airborne real-time performance, the optimization layer employs real-time iteration and efficient numerical solution techniques, decomposing nonlinear optimization into a rapidly computable sequential quadratic programming problem. Warm-start and convergence checks ensure rapid response. The stabilization layer meets high-frequency execution requirements through lightweight algorithm design. The system adopts a modular architecture and standardized interface design, supporting independent development and system integration of algorithms at each layer, achieving a systematic balance between security, adaptability, and real-time performance.

[0081] In summary, compared with the prior art, this application has at least the following advantages: First, verifiable safety assurance: This application constructs a full-chain active protection system, which realizes the transformation from passive response to active protection through a triple mechanism of dynamic safety corridor, Tube theoretical error envelope and real-time disturbance compensation, providing strict theoretical assurance for the safe flight of the system under uncertainty.

[0082] Second, coordinated optimization of control performance: Breaking the limitations of the traditional separation of planning and control, this application significantly improves system energy efficiency and trajectory tracking accuracy, thereby improving flight quality, through the deep integration of online trajectory planning and robust tracking control, while ensuring safety.

[0083] Third, environmental adaptive intelligent adjustment: Based on the dual-loop feedback mechanism, this application can autonomously adjust the multi-objective weights of the control strategy according to the real-time environmental status and mission requirements, realize intelligent and smooth switching of flight modes, and maintain basic stability under abnormal conditions, thereby enhancing the mission reliability in complex environments.

[0084] Fourth, the advantages of modular architecture design: adopting a clear three-layer decoupled architecture, it enables independent development and collaborative operation of each functional module, which can significantly improve the scalability and portability of the system.

[0085] Based on the same inventive concept, this application also provides a hierarchical cooperative flight control method for the transition phase of a tiltrotor unmanned aerial vehicle (UAV) applied to the system provided above. The solution provided by this method is similar to the implementation scheme described in the above system. Therefore, the specific limitations of one or more embodiments of the hierarchical cooperative flight control method for the transition phase of a tiltrotor UAV provided below can be found in the limitations of the hierarchical cooperative flight control system for the transition phase of a tiltrotor UAV described above, and will not be repeated here.

[0086] In one exemplary embodiment, such as Figure 5 As shown, a hierarchical cooperative flight control method for the transition phase of a tiltrotor unmanned aerial vehicle (UAV) is provided, including: Step 100: Based on the dynamic model, environmental perception information, health status and predicted state sequence of the tiltrotor UAV, a dynamic safety corridor is generated through reachability set analysis and intersection operation, and a strategy parameter vector guiding performance trade-offs is formulated according to the mission status; the predicted state sequence is fed back in real time through the outer loop. Step 101: The Tube robust nonlinear model predictive control method is adopted, and the dynamic safety corridor is embedded as a time-varying constraint optimization problem. Combined with the policy parameter vector and the early warning signal, the nominal performance optimization and robust safety guarantee are decoupled through constraint tightening and robust invariant set design, so as to obtain the optimal control command at the current time and the predicted state sequence at the next time. The early warning signal is fed back in real time through the inner loop. Step 102: Based on the optimal control command and the predicted state sequence, suppress disturbances in real time and determine the tracking error; when the tracking error is located at the internal safety boundary and the external constraint boundary, generate an early warning signal; the internal safety boundary and the external constraint boundary are determined by the dynamic safety corridor.

[0087] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0088] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0089] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0090] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0091] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (RRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0092] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0093] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0094] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle (UAV), characterized in that, include: Decision-making layer, optimization layer, and stability layer; The decision layer is used to generate a dynamic safety corridor based on the dynamic model, environmental perception information, and health status of the tiltrotor UAV, combined with the predicted state sequence obtained by the optimization layer, through reachability set analysis and intersection operation, and to formulate a strategy parameter vector to guide performance trade-offs based on the mission status. The optimization layer employs the Tube robust nonlinear model predictive control method, embedding the dynamic safety corridor as a time-varying constraint into the optimization problem. Combined with the policy parameter vector and early warning signal, and through constraint tightening and robust invariant set design, it decouples nominal performance optimization from robust safety assurance, obtaining the optimal control command for the current moment and the predicted state sequence for the next time step. The predicted state sequence is fed back to the decision layer in real time through the outer loop. The stabilization layer is used to suppress disturbances and determine the tracking error in real time according to the optimal control command and the predicted state sequence, and to generate the warning signal when the tracking error is located at the internal safety boundary and the external constraint boundary; the warning signal is fed back to the optimization layer in real time through the inner loop; the internal safety boundary and the external constraint boundary are determined by the dynamic safety corridor.

2. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 1, characterized in that, The decision-making body includes: The reachability set analysis module is used to determine the set of all reachable states from the current state within a future time period based on the dynamic model and using the centrally symmetric polyhedral set operation method, thus obtaining the reachability set. The safe corridor synthesis module, connected to the reachability set analysis module, is used to perform intersection operations between the reachability set and geographical constraints and flight envelope constraints, and to generate safe corridor slices at each time moment through conservative approximation. The safe corridor slices at each time moment are connected along the time axis to construct a continuously evolving dynamic safe corridor. The strategy decision module, connected to the reachability set analysis module and the safety corridor synthesis module, is used to determine the strategy parameter vector based on the multi-objective optimization principle, comprehensively considering the task status, environmental perception information, and the health status of the tiltrotor UAV. The strategy decision module is also used to acquire the predicted state sequence of the outer loop feedback in real time to evaluate the effectiveness of the strategy parameter vector and adjust the dynamic safety corridor.

3. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 2, characterized in that, In the safe corridor synthesis module, the process of performing an intersection operation between the reachability set and the geographical constraints and flight envelope constraints, and generating safe corridor slices at each time point through conservative approximation, includes: The reachable set is intersected with the geographical constraints and the flight envelope constraints to obtain the reachable state region at each time step. Obtain the sampling time points of the reachable state region, and determine the covariance matrix and inflation coefficient based on the sampling time points and their geometric centers; The covariance matrix is ​​expanded using the expansion coefficient to obtain a shape matrix, thereby achieving a conservative approximation of the reachable state region; The security corridor slice is generated based on the shape matrix.

4. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 3, characterized in that, The covariance matrix is ​​expressed as follows: ； The expansion coefficient is expressed as: ； In the formula, Let covariance matrix be the variance matrix. This represents the total number of sampling time points within the reachable state region. For the k-th sampling time point, The geometric center of the sampling time point, This is the transpose of the matrix. The coefficient of thermal expansion is 1 / 3. This indicates that the maximum Mahalanobis distance among all sampling time points is obtained by iterating through all sampling time points. This represents the inverse of the covariance matrix.

5. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 2, characterized in that, In the policy decision-making module, based on the principle of multi-objective optimization, considering the task status, environmental perception information, and the health status of the tilt-rotor UAV comprehensively, the process of determining the policy parameter vector includes: Determining an urgency factor according to the time requirement of the task status; Determining an environmental correction factor according to the real-time wind field information in the environmental perception information; Determining a status factor according to the health status of the tilt-rotor UAV; Using a mapping function to obtain the policy parameter vector based on the urgency factor, the environmental correction factor, and the status factor.

6. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 2, characterized in that, In the policy decision-making module, the process of performing the dynamic safety corridor adjustment includes: Determining the distance between the predicted state in the predicted state sequence and the boundary of the current dynamic safety corridor; Determining whether the distance is less than a preset safety threshold; When the distance is less than the preset safety threshold, expanding the current dynamic safety corridor along the dominant direction of the current dynamic safety corridor; When the distance is greater than the preset safety threshold, contracting the current dynamic safety corridor along the dominant direction of the current dynamic safety corridor.

7. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 1, characterized in that, The optimization layer includes: A tube model predictive controller module, connected to the decision-making layer, for using the Tube robust nonlinear model predictive control method to embed the dynamic safety corridor as a time-varying constraint into the optimization problem to obtain a nonlinear optimization problem; A solver module, connected to the tube model predictive controller module, for converting the nonlinear optimization problem into a series of quadratic programming subproblems at each iteration point and using the initial solution for warm start to obtain the predicted state and control command, and in the iterative process, determining the iteration step size in real time through line search until the set condition is reached, and obtaining the optimal control command at the current moment and the predicted state sequence for the next time.

8. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 1, characterized in that, The stabilization layer includes: A disturbance observer module, connected to the optimization layer, for estimating the total disturbance online in real time based on the tracking error between the actual state and the predicted state sequence; A robust controller module, connected to the disturbance observer module, for designing a feedforward-feedback composite control to determine the robust compensation control amount based on the tracking error and the estimated total disturbance; superimposing the robust compensation control amount onto the optimal control command to obtain the total control command; A control allocation module, connected to the robust controller module, for allocating the total control command to each actuator according to the control allocation command under the current rotor tilt angle under the physical limit constraint condition; A monitoring module, connected to the disturbance observer module and the optimization layer respectively, for feeding back the state information and generating the warning signal based on the state recognition mechanism of the tracking error.

9. The hierarchical cooperative flight control system for the transition phase of a tiltrotor unmanned aerial vehicle according to claim 1, characterized in that, ​ ​ 10. A hierarchical cooperative flight control method for the transition phase of a tiltrotor unmanned aerial vehicle, characterized in that, ​ Based on the dynamic model, environmental perception information, health status and predicted state sequence of tiltrotor UAV, a dynamic safety corridor is generated through reachability set analysis and intersection operation, and a strategy parameter vector guiding performance trade-offs is formulated according to the mission status; the predicted state sequence is fed back in real time through the outer loop. The Tube robust nonlinear model predictive control method is adopted, in which the dynamic safety corridor is embedded as a time-varying constraint optimization problem. By combining the policy parameter vector and the early warning signal, the nominal performance optimization and robust safety guarantee are decoupled through constraint tightening and robust invariant set design, so as to obtain the optimal control command at the current time and the predicted state sequence at the next time. The early warning signal is fed back in real time through the inner loop. Based on the optimal control command and the predicted state sequence, disturbances are suppressed in real time and the tracking error is determined; when the tracking error is located at the internal safety boundary and the external constraint boundary, the warning signal is generated; the internal safety boundary and the external constraint boundary are determined by the dynamic safety corridor.

Images