Flight control parameter cluster grouping optimization and constraint projection safety injection system and method

Abstract

This invention belongs to the field of flight control simulation technology, and relates to a system and method for flight control parameter cluster grouping optimization and constraint projection safety injection. It addresses the problem that existing technologies struggle to achieve efficient, robust, and fully automated optimization of flight control parameters in a simulation environment. Based on time-scale separation theory, this invention divides control parameters into fast, medium, and slow clusters and designs a periodic elite migration optimization mechanism. It completes a two-layer feasibility verification through boundary projection adaptive penalty and seven types of dynamic association rules. It achieves piecewise adaptive smooth parameter injection by combining simulation dynamic activity and parameter changes. It constructs an improved robust index that integrates fluctuation and discrete characteristics, and completes multi-scenario quantitative evaluation with staged scenario screening and adaptive index weighting, automatically iterating and outputting the optimal control parameters throughout the closed-loop process. This invention achieves fully automated and efficient optimization of flight control parameters.

Description

Technical Field

[0001] This invention belongs to the field of flight control and simulation technology, specifically relating to a flight control parameter cluster grouping optimization and constraint projection safety injection system and method. Background Technology

[0002] The tuning and optimization of flight control parameters is a crucial step in the design of flight control systems, directly impacting the aircraft's handling characteristics, stability, and flight safety. Currently, the tuning and optimization of flight control parameters heavily relies on engineers' manual experience. The typical process involves engineers setting initial parameters based on experience, running tests in a flight simulation environment, visually observing response curves (such as step response or frequency domain plots) or judging performance based on subjective feelings, then manually adjusting the parameters and repeating this process. This constitutes a time-consuming and inefficient cycle of "manual trial and error."

[0003] In existing technologies, some scholars have proposed using intelligent optimization algorithms such as Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) for automatic tuning of flight control parameters, which has improved tuning efficiency to some extent. However, existing technologies still have significant shortcomings in the following aspects:

[0004] First, there is the problem of quality search and reliance on experience. Existing methods adopt a uniform optimization strategy for all control parameters without considering the differences in physical characteristics of different parameters. The parameter tuning results are severely limited by the initial parameter settings and algorithm parameter configuration, which easily leads to local optima and makes it difficult to obtain parameter combinations with better global performance. Second, there is the issue of simulation stability. During automated optimization, sudden changes or improper combinations of candidate parameters can easily cause numerical divergence in the high-fidelity simulation model, leading to simulation interruption or even collapse, which disrupts the continuity of the automated process. Existing methods lack an effective parameter safety injection mechanism and cannot achieve long-cycle automated optimization while ensuring simulation stability.

[0005] Third, there is the issue of physical feasibility. Pure black-box optimization algorithms lack the use of knowledge in the field of flight control, which may produce theoretically optimal but physically infeasible or engineering impossible parameter combinations, such as causing actuator saturation, lateral coupling instability, etc. Fourth, there is a problem with insufficient robustness verification. Existing methods usually only adjust parameters for a few typical design points, lacking systematic and automated verification across the entire flight envelope in multiple scenarios, thus failing to guarantee the adaptability and robustness of the parameters.

[0006] To address the aforementioned issues, this invention proposes a flight control parameter cluster grouping optimization and constraint projection safety injection system. Summary of the Invention

[0007] To address the aforementioned problems in existing technologies, namely the low efficiency, lack of simulation stability, absence of physical feasibility constraints, and insufficient robustness verification of current flight control parameter optimization methods, making it difficult to achieve efficient, robust, and fully automated optimization of flight control parameters in a simulation environment, this invention, in its first aspect, proposes a flight control parameter cluster grouping optimization and constraint projection safety injection system. This system includes: The parameter and scenario configuration module is configured to: define the flight control parameter set and its feasible domain boundary, and define a multi-scenario test set; The intelligent optimization engine is configured to: divide the flight control parameter set into multiple parameter clusters according to the physical modal response speed of the aircraft; and perform collaborative optimization on the multiple parameter clusters to generate candidate parameter vectors by exchanging information between different parameter clusters; The parameter cluster management and security injection module is configured to: perform boundary compliance checks on the candidate parameter vectors; select appropriate constraint processing paths for non-compliant parameters based on their attributes; obtain parameter vectors that pass the boundary checks and violation metrics; and dynamically determine the smooth transition time of each parameter based on the parameter vectors used in the current simulation, the parameter vectors that pass the boundary checks, and the real-time state data of the current simulation, and generate corresponding smooth transition sequences. An automated simulation executor is configured to: inject parameter values ​​from the smooth transition sequence into the flight simulation model step by step according to the simulation step size, drive the flight simulation model to run all test cases in the multi-scenario test set, and record simulation data; The multi-dimensional objective performance evaluation module is configured to: quantitatively evaluate the simulation data and construct an improved multi-scenario robustness index, and generate a comprehensive feedback cost after weighted normalization and feed it back to the intelligent optimization engine; The results output module is configured to: filter the optimal parameter combination based on the comprehensive feedback cost, and generate an optimization report containing convergence curves and performance analysis.

[0008] In a second aspect, the present invention proposes a method for flight control parameter cluster grouping optimization and constraint projection safety injection. Based on the aforementioned flight control parameter cluster grouping optimization and constraint projection safety injection system, the method includes the following steps: S1: Initialize configuration, define the flight control parameter set and its feasible domain boundary, and define the multi-scenario test set; S2: Initialize the optimization process by dividing the flight control parameter set into multiple parameter clusters according to the physical modal response speed of the aircraft; and perform collaborative optimization on the multiple parameter clusters by exchanging information between different parameter clusters to generate candidate parameter vectors; S3: Parameter cluster secure injection and intelligent smooth mapping, perform boundary compliance checks on the candidate parameter vectors, select appropriate constraint processing paths for non-compliant parameters based on their attributes, and obtain parameter vectors that pass the boundary checks and violation measures; Based on the parameter vectors used in the current simulation, the parameter vectors that have passed the boundary check, and the real-time state data of the current simulation, the smooth transition time of each parameter is dynamically determined, and a corresponding smooth transition sequence is generated. The parameter values ​​in the smooth transition sequence are injected into the flight simulation model step by step according to the simulation step size; S4: Automated simulation and objective evaluation, driving the flight simulation model to run all test cases in the multi-scenario test set and recording simulation data; The simulation data is quantitatively evaluated and an improved multi-scenario robustness index is constructed. After weighted normalization, a comprehensive feedback cost is generated. S5: Optimize feedback and convergence judgment. Update the population and generate the next generation candidate parameter vector based on the comprehensive feedback cost. The process of updating the population includes: performing the collaborative optimization, updating the parameters by exchanging information between different parameter clusters based on the comprehensive feedback cost; iteratively executing steps S3 to S5 until the preset convergence condition is met. S6: Output the global optimal solution, select the optimal parameter combination based on the comprehensive feedback cost, and generate an optimization report containing the convergence curve and performance analysis.

[0009] The beneficial effects of this invention are: 1) Based on the separation of the time scale of the aircraft, a three-cluster hierarchical elite migration collaborative optimization architecture is constructed, which automatically iterates and updates the control parameters without the need for manual trial and error to adjust the parameters. This solves the problem that traditional unified optimization is prone to getting trapped in local optima and has low tuning efficiency, and realizes intelligent optimization of flight control parameters across the entire domain, which greatly shortens the development cycle of control laws. 2) The design of a piecewise adaptive smooth injection mechanism integrates dynamic activity and parameter change, and adopts a monotonically bounded S-curve to progressively update simulation parameters, eliminating parameter step shocks and solving the problems of simulation numerical divergence and frequent interruption of automated processes caused by sudden changes in candidate parameters, thus supporting unattended long-term continuous optimization iteration. 3) Establish a boundary adaptive penalty + seven types of dynamic rules dual-layer verification system, implement gradient buffer correction or maximum cost shielding for illegal parameters, solve the problem of implicit infeasible parameters such as only limiting the upper and lower limits of parameters, easy to generate coupled oscillations, critical instability, etc., and ensure that the output parameters fully meet the requirements of flight dynamics and engineering control. 4) Construct an improved robust quantitative index that integrates fluctuation and discrete characteristics, and match it with a phased scenario screening and index adaptive weighting strategy to complete multi-condition evaluation with a unified numerical cost. This solves the problems of single-condition parameter tuning, subjective evaluation, and difficulty in quantitatively verifying the robustness of the full envelope, and realizes standardized and objective evaluation of parameter adaptation capability. 5) The system is equipped with a graded penalty and feasible solution adaptive reinforcement mechanism to quickly filter out high-risk invalid individuals and reduce redundant simulation computing power consumption; the entire link forms an automated closed loop and automatically outputs standardized analysis reports, which simultaneously solves the derivative problems of slow optimization convergence and high cost of manual data sorting, and improves the overall standardization level of flight control design. Attached Figure Description

[0010] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0011] Figure 1 This is an architecture diagram of the flight control parameter cluster grouping optimization and constraint projection safety injection system of the present invention; Figure 2 This is a flowchart of the flight control parameter cluster grouping optimization and constraint projection safety injection system of the present invention. Detailed Implementation

[0012] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0013] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0014] To more clearly explain the flight control parameter cluster grouping optimization and constraint projection safety injection system of the present invention, the following will be combined with... Figures 1 to 2 The steps in the embodiments of the present invention will be described in detail below.

[0015] Flight control parameter cluster grouping optimization and constraint projection safety injection system, see [link to documentation]. Figure 1 The system includes: The parameter and scenario configuration module is configured to: define the flight control parameter set and its feasible domain boundary, and define a multi-scenario test set; In this embodiment, the user performs initialization settings through the parameter and scenario configuration module, first defining the set of flight control parameters to be optimized. and specify each parameter The feasible region boundary, also known as the physical feasible lower bound. and the Upper Realm , forming the lower bound vector and upper bound vector These boundaries are determined by the actuator saturation limit, the control theory stability boundary, and engineering experience; the flight control parameter set includes angular rate damping gain. The feasible region is [0.5, 3.0]; servo gain The feasible region is [1.0, 4.0]; attitude scaling gain The feasible region is [1.0, 5.0]; attitude integral gain The feasible region is [0.1, 1.0]; the gain is highly controllable. The feasible region is [0.1, 2.0]; throttle gain The feasible region is [0.5, 3.0]. Correspondingly, construct lower bound vectors and upper bound vectors: ; Simultaneously define multi-scenario test sets Each scenario defines the initial conditions, input commands, and evaluation period for the simulation. The multi-scenario test set includes: level flight acceleration scenario (altitude 3000 m, speed 0.6 Ma), climb scenario (altitude 2000 m → 5000 m), turning scenario (slope 30°), descent approach scenario, and gust disturbance scenario. Configure parameters such as population size and number of iterations for the intelligent optimization engine, and preset the initial weights for each performance metric. j and robustness weights For example, configure the intelligent optimization engine as Particle Swarm Optimization (PSO), with a population size Npop of 50 and a maximum number of iterations. Robustness weights ; The intelligent optimization engine is configured to: divide the flight control parameter set into multiple parameter clusters according to the physical modal response speed of the aircraft; and perform collaborative optimization on the multiple parameter clusters to generate candidate parameter vectors by exchanging information between different parameter clusters; specifically, the flight control parameter set is divided into fast-variable clusters, medium-variable clusters, and slow-variable clusters according to the physical modal response speed of the aircraft, and the migration weights of the three clusters decrease progressively. A periodic elite individual migration mechanism is executed between the three clusters, and the medium-variable cluster simultaneously receives optimal solution information from the fast-variable cluster and the slow-variable cluster to obtain candidate parameter vectors. In this embodiment, to improve optimization efficiency and physical meaning, the concept of parameter clusters is introduced. The flight control parameter set is functionally grouped according to the physical modes it controls. Based on the time scale separation theory, the flight control parameter set is divided into fast-variable clusters (such as rudder loop gain), medium-variable clusters (such as attitude angle control), and slow-variable clusters (such as airspeed / thrust control) according to the response speed of the physical modes it controls. A complete parameter cluster division example for a typical aircraft is shown in Table 1 below: Table 1

[0016] For cross-cluster coupling parameters, they are assigned to the corresponding cluster according to the principle of "dominant physical mode priority", and the performance index is corrected through the coupling channel during the evaluation stage. For cross-cluster coupling parameters (i.e., control parameters that simultaneously affect multiple physical modes, such as attitude-height coupling parameters) ), and are assigned to the corresponding cluster according to the principle of "dominant physical mode priority" (e.g., if If the primary influence is on the altitude trajectory (slowly varying mode), then it is classified into the slowly varying cluster. If the attitude angle is significantly affected simultaneously (medium-variable mode), the cross-cluster impact is quantified by adjusting the performance index through the coupling channel during the evaluation phase. The specific implementation method of the coupling channel adjustment is as follows: Define cross-cluster coupling bias term The formula is: ; Where C is the set of cross-cluster coupling pairs (e.g., , (This represents the coupling between intermediate and slow-changing clusters). The coupling weights (determined by the physical modal correlation, satisfying...) =1); For clusters The actual performance indicators include integral squared error (ISE), integral absolute error (IAE), time multiplication and integral absolute error (ITAE), root mean square value of control surface deflection RMS(δ), phase margin (PM), gain margin (GM), etc. For cluster j pairs of clusters The predicted performance values ​​are calculated from the transfer function of the coupled channel or an empirical model. For example, the impact of attitude jitter in medium-variable clusters on altitude overshoot in slow-variable clusters can be predicted using the attitude-altitude transfer function G(s), i.e.: ; in, This serves as a baseline for the high performance of slow-varying clusters. The attitude angle error is k, and the proportionality coefficient is k. For the evaluation period; Original performance indicators (Such as ISE for a single cluster, CD for multi-scenario comprehensive indicators, etc.) and coupling deviation term Weighted fusion yields the corrected performance metrics: ; Wherein, λ is the coupling correction coefficient (typically λ∈[0.1,1.0], determined by simulation verification or domain knowledge, used to balance the weights of "single cluster performance" and "cross-cluster coupling effect"); The intelligent optimization engine employs any one of particle swarm optimization, genetic algorithm, differential evolution, or Bayesian optimization, i.e., any one of these serves as the basic intelligent optimization framework, coupled with a periodic elite individual migration mechanism among the three clusters. The corresponding carriers for the candidate parameter vectors under each algorithm are defined as follows: When particle swarm optimization is used, the candidate parameter vector is the particle position vector; When using a genetic algorithm or differential evolution, the candidate parameter vector is the population individual vector; When Bayesian optimization is used, the candidate parameter vector is the evaluation point recommended by the acquisition function; The process involves executing a periodic elite individual migration mechanism among the three clusters, with the intermediate cluster simultaneously receiving optimal solution information from both the fast and slow clusters, to obtain a candidate parameter vector. Specifically, the intelligent optimization engine adopts a hierarchical collaborative optimization strategy, denoted as generation t (the current optimization iteration number). Individuals in the populations that exhibit fast, medium, and slow mutation patterns, respectively. For each cluster to be the current optimal solution, the migration weights of the three clusters decrease sequentially, specifically as follows: Fast-changing cluster migration weights... Greater than the migration weight of the intermediate cluster Medium-variable cluster migration weights Greater than the slow-changing cluster migration weight That is, satisfying The migration weights of the fast-changing clusters The maximum value is taken to reflect the physical causal relationship between high-frequency dynamics and mid-frequency attitude response and low-frequency trajectory tracking. Each generation performs a three-cluster information migration, with the following migration rules: ; in, The next generation of a fast-changing cluster is generated by a weighted combination of its current generation individuals and the optimal solution. The next generation of a medium-changing cluster is generated by a weighted combination of its current generation individuals and the optimal solutions of both the fast-changing and slow-changing clusters. The next generation of a slow-changing cluster is generated by a weighted combination of its current generation individuals and the optimal solution. When receiving elite individual information, the medium-changing cluster receives information from the fast-changing cluster with different weights. Greater than the weight of receiving information from the slow-changing cluster Typical values ​​are: , , , , Through the above mechanism, the excellent solutions of the fast-variant cluster influence the medium-variant cluster with a large weight, while the slow-variant cluster mainly maintains its own evolution. Under the premise of ensuring high-frequency dynamic stability, this effectively guides the low-frequency trajectory and generates a new generation of candidate parameter vectors. ; The parameter cluster management and security injection module is configured to: perform boundary compliance checks on the candidate parameter vectors; select appropriate constraint processing paths for non-compliant parameters based on their attributes to obtain parameter vectors that pass the boundary checks and violation metrics; dynamically determine the smooth transition time of each parameter based on the parameter vectors used in the current simulation, the parameter vectors that pass the boundary checks, and the real-time state data of the current simulation, and generate corresponding smooth transition sequences; specifically, perform boundary violation detection on the candidate parameter vectors; for non-compliant parameters, select corresponding constraint processing paths based on the predefined attributes of the non-compliant parameters to obtain parameter vectors that pass the boundary checks and violation metrics; calculate dynamic activity indicators and relative parameter changes based on the parameter vectors used in the current simulation, the parameter vectors that pass the boundary checks, and the real-time state data of the current simulation, and dynamically determine the smooth transition time of each parameter according to segmented decision logic; and generate smooth transition sequences using a transition curve generation unit based on the smooth transition times. In this embodiment, the parameter cluster management and security injection module receives candidate parameters from the optimization engine. Parameters used in the current simulation Simultaneously read the preset lower bound vector. Upper bound vector Weighting matrix of flight quality standards Population size Feasible solution proportion threshold (The default value is 0.2, which can be adjusted according to the scenario). Performing core safety and smoothing processing ensures that the parameters of the subsequent input aircraft simulation model are both physically feasible and can be smoothly transitioned, which is the key to ensuring the continuity of the automated process. right Perform element-by-element boundary checks for any parameter. ,like or Calculate the number of violations: ; in, For the first The boundary violation metric value of each control parameter is used to quantify the severity of parameter out-of-bounds behavior, ensuring that deviations below the lower bound or above the upper bound are positively accumulated, providing an accurate basis for subsequent adaptive penalty calculation; Two parallel constraint processing paths are provided based on predefined parameter attributes. The constraint processing paths include: For ordinary parameters, the projection method is used to trim non-compliant parameters to the feasible region, resulting in corrected parameters. : To ensure the physical feasibility of the parameter values ​​injected into the simulation model, and to guide the optimization algorithm away from the infeasible region in subsequent iterations, an adaptive penalty cost is generated. This adaptive penalty cost increases with the number of iterations and carries flight quality weights. Specifically, the penalty intensity of the adaptive penalty cost is equal to the sum of the base penalty coefficient and the term that increases linearly with the number of iterations, multiplied by the square of the violation amount, making the penalty for boundary violations stronger in the later stages of optimization than in the early stages. The penalty intensity is multiplied by a differentiated weight set based on flight quality standards, making the penalty for violations of key flight parameters higher than that of ordinary parameters. When the proportion of feasible solutions is lower than a preset threshold, the penalty intensity is automatically increased or an additional amplification factor is introduced to force the intelligent optimization engine away from the infeasible region. The formula for adaptive penalty cost is: ; in, Basic penalty coefficient; A dynamically adjusted growth factor; This represents the maximum number of iterations. The first in the flight quality standard weight matrix The differential penalty weights corresponding to the parameters; This represents the current iteration number; To achieve an adaptive penalty coefficient, which increases with the number of iterations, the penalty is weaker in the early iterations, allowing the algorithm to explore boundary regions, while the penalty is stronger in the later iterations, forcing the algorithm to converge to the feasible region. This design makes the penalty for boundary violations stronger in the later stages of optimization and imposes higher weight on violations of key flight parameters, forcing the search to focus on areas within the feasible region that meet flight quality requirements. In this embodiment... Furthermore, when the proportion of feasible solutions is detected to be too low, the penalty coefficient 'a' is automatically increased or an additional amplification factor is introduced to force the optimization engine away from the infeasible region; specifically, the proportion of feasible solutions is calculated as follows: ; in, To satisfy all boundary constraints in the current population The number of individuals, For population size, when If the proportion of feasible solutions is deemed too low, a feasibility recovery mode is triggered, automatically increasing the penalty intensity or introducing an additional amplification factor to force the intelligent optimization engine away from the infeasible region; (automatically increasing the growth factor) Growth follows a step-by-step rule: initial value Every time In the next iteration, if ,but promote Until the maximum growth limit is reached. (Avoid excessive punishment); if ,but Reset to ; An additional amplification factor γ (default γ=1.5, adjustable according to the scenario) is introduced to add an additional penalty term on top of the original adaptive penalty term, resulting in a total penalty term. for: ; The first item is the original adaptive penalty item, and the second item is an additional penalty item that applies extra penalties to the absolute value of the violation, forcing the optimization engine to move away from the infeasible region. For predefined critical sensitive parameters (critical parameters that may cause simulation crashes even with slight exceedances, such as angular rate damping gain and servo gain fast-changing cluster parameters; these parameters directly affect the actuators and high-frequency dynamic loops, and even a small exceedance of the parameter limits will cause severe control surface jitter and aircraft attitude divergence), immediately terminate the current evaluation and return the preset maximum penalty cost to the intelligent optimization engine; the formula for the maximum penalty cost is: ; in, As an amplification factor (an amplification factor for the penalty of critical parameter out-of-bounds, used to amplify the cost of violations of sensitive parameters and significantly reduce the probability of such parameters being selected), this path, together with methods such as the Lagrange multiplier method, enhances the flexibility of optimization. The preset baseline maximum penalty value; The outputs of both paths are parameter vectors that have passed the boundary check. and the number of violations (used to calculate penalties); To prevent implicit coupling risks even when parameter combinations are within the boundaries, the parameter cluster management and safety injection module also includes an intra-cluster association rule review unit. This unit is configured to: perform a dynamic rationality review on the parameter vectors that pass the boundary check; the review includes one or more of the following constraints: proportional-integral association constraints, lateral gain ratio constraints, differential gain upper limit constraints, stability margin constraints, integral saturation prevention constraints, cross-coupling suppression constraints, and energy dissipation constraints; when a rule is violated, a gradual buffer correction is performed; if the correction fails, a preset maximum penalty cost is returned; specifically, based on flight control law design experience and stability theory, the parameters that pass the boundary check are reviewed... A secondary dynamics rationality review is conducted. The thresholds for each item in the rule base are quantitatively determined based on flight dynamics and control theory. The specific derivation basis is as follows: Damping ratio constraint (threshold) Based on the dominant pole analysis of a second-order system, the closed-loop damping ratio of the attitude control loop is required to satisfy... Combining relational formulas The threshold calculation formula is derived as follows: In the formula The minimum expected undamped natural frequency of the system is determined by the aircraft's maneuverability level. The minimum allowable damping ratio; Lateral coupling constraints (threshold) Based on the lateral small perturbation equation, singular value decomposition and root locus analysis are used to identify the critical gain ratio that causes Dutch roll mode deterioration and excessive coupling of the lateral / lateral channels. The critical gain ratio is defined as follows: ; In the formula, This is the stability margin gain from the rudder to the sideslip channel; Stability margin constraint (threshold) Frequency domain verification is performed based on the Nyquist criterion or state-space analysis method, using the open-loop transfer function. Calculate phase margin, standard design requirements ; The complete intra-cluster association rule base contains seven constraint rules: Proportional-integral correlation rule: requirements This ensures that the dominant pole of the system has reasonable damping, and avoids problems such as oscillation and integral saturation in the system. Lateral gain ratio constraint: requirements To prevent the lateral channel parameters from over-exciting the heading mode and to suppress the lateral coupling effect; Upper limit rule for differential gain: In the formula, This rule, which specifies the maximum permissible rate of change of angle of attack, is used to avoid amplifying high-frequency noise in the differential element and to prevent actuator saturation. The desired damping ratio; The desired natural frequency; Stability margin guarantee rule: requires phase margin Gain margin , To minimize the magnitude margin, ensure the robust stability of the closed-loop system, and avoid critical unstable operating conditions; Integral saturation prevention rules: ,in, To determine the maximum allowable adjustment time, a threshold is designed based on the system's maximum adjustment time to suppress integral saturation (integral wind disturbance) phenomenon; Cross-coupling suppression rules: For pitch-roll and other coupled channels, the following requirements are made: The threshold is based on the fact that the real part of the coupled mode eigenvalue is less than the safety negative threshold. By reverse deduction, This is the coupling channel gain; Energy dissipation constraint: requires the sum of the real parts of all closed-loop poles of the system. ( This is the system's energy dissipation safety threshold. The sum of the real parts of the closed-loop poles must be less than this negative value to ensure that the system continuously dissipates energy without oscillation and accumulation, and to prevent abnormal energy accumulation from causing dynamic instability. Buffer recovery mechanism: When the parameter combination violates any of the above association rules, the system will not directly terminate the process, but will instead activate the parameter smooth buffer recovery mechanism. Deviation metric: Calculates the degree of rule violation. , In the formula, Let r be the rule function corresponding to the r-th rule. This corresponds to the constraint threshold. Taking the proportional-integral correlation rule as an example: Assume the current attitude proportional gain Integral gain , ,satisfy The constraints were met, and the review was approved. Gradual correction: Introducing a range of values Damping factor Generate the corrected parameter vector: The parameters are gradually corrected in reverse along the gradient of the rule violation degree, and the correction step size is controlled by η to avoid sudden parameter changes. Continue iterative corrections until one of the following termination conditions is met: rule violation rate ( The preset tolerance measures the "acceptable degree of rule violation"; the maximum buffer step size Nmax is reached (Nmax is a positive integer and needs to be preset, for example, Nmax=100 to prevent infinite iteration).

[0017] Penalty backtracking: If the rule requirement (i.e., Rviol ≥ 1) cannot be met after buffer correction (iteration to the termination condition). If Nmax has already been reached, then the maximum penalty cost is returned. At the same time, the parameter set is marked as infeasible, guiding the optimization engine to actively avoid the parameter area; if all parameter combinations meet the rule requirements after review and correction, the process proceeds to the next stage; if the buffer correction fails or the rule verification fails, the penalty backtracking logic is executed. Based on the parameter vectors used in the current simulation The parameter vector that passes the boundary check (i.e. Based on the current simulation's real-time status data, a dynamic activity index is calculated. and relative change of parameters The smooth transition time of each parameter is dynamically determined according to the segmented decision-making logic; Among them, the dynamic activity index Calculate using one of the following methods: Method 1, based on the weighted fluctuation of key state variables, specifically: using the instantaneous values ​​of key state variables (key state variables include attitude angular velocity and control surface deflection rate) Relative to the benchmark stationary value The deviation divided by the historical maximum fluctuation range of the state variable The weighted L2 norm is used as a dynamic activity indicator, and its formula is as follows: ; Among them, the weights of each state variable Sensitivity analysis determined that the sum of all weights is 1, that is... ; Alternatively, the calculation can be performed based on the sum of the real parts of the eigenvalues ​​of the system state matrix obtained by linearizing the flight simulation model under the current state.

[0018] Wherein, the relative change of the parameter The calculation method is as follows: for each parameter Combined with the corresponding lower bound Upper Realm Calculate its normalized relative change. : After normalization, the range of change is [0, 1]. The segmented decision-making logic is as follows: When the relative change of parameters does not exceed the first threshold and the dynamic activity index does not exceed the safety threshold (i.e.) ,in, The first threshold, (As a safety threshold), the baseline smoothing time of the cluster to which this parameter belongs is used. The reference smoothing time increases sequentially for fast-changing clusters, medium-changing clusters, and slow-changing clusters. ; When the relative change of the parameter is between the first threshold and the second threshold (i.e.) ,in, (As the second threshold), based on the baseline smoothing time, a nonlinear extension is performed according to the difference between the relative change of the parameter and the first threshold, and the difference between the dynamic activity index and the safety threshold. The first threshold is less than the second threshold. This is the amplification factor for the change. This is the activity amplification factor; When the relative change of the parameter exceeds the second threshold or the dynamic activity index exceeds the warning threshold (i.e.) ,in, (As the warning threshold), a preset maximum safe smoothing time is used. ; In this embodiment, , ; Assuming the angular rate damping gain used in the current simulation Target value The feasible region is determined by this parameter. ,but: ; Current dynamic activity level ,satisfy and The calculation yields: ; Based on the smooth transition time A smooth transition sequence is generated using a transition curve generation unit. The smooth transition sequence generated by the transition curve generation unit is monotonic and bounded: the parameter values ​​in the smooth transition sequence transition monotonically from the parameter values ​​applied in the current simulation to the parameter vector that has passed the boundary check, and the parameter values ​​remain within the boundary of the feasible region throughout the transition. This achieves a smooth, continuous, and differentiable transition of parameters from their current value to the target value. The specific implementation methods of the transition curve generation unit include any one of the following: S-curve mapping, linear transition, exponential transition, first-order low-pass filtering, or polynomial trajectory planning. In this embodiment, an S-curve mapping function is used to generate a smooth sequence, and a smoothing factor is defined. : ; in, The curve shape factor (default 12) controls the steepness of the transition curve; a larger value results in a steeper transition. t is the time elapsed since the start of the transition. Therefore, the parameter values ​​applied to the simulation at time t are: ; In this embodiment, The sequence monotonically increases from 1.2 to 2.8, and lies entirely within the feasible region [0.5, 3.0], achieving a smooth, continuous, and differentiable transition. Automated simulation actuator, configured as follows: based on simulation step size ( The parameter values ​​in the smooth transition sequence are gradually injected into the flight simulation model, driving the flight simulation model to run all test cases in the multi-scenario test set and record simulation data. In this embodiment, the automated simulation actuator loads the parameter values ​​in the smooth transition sequence and sequentially drives the high-fidelity flight simulation model to run multiple scenario test sets. All m test cases are recorded, and the entire process data of each simulation is recorded. The multi-dimensional objective performance evaluation module is configured to: quantitatively evaluate the simulation data and construct an improved multi-scenario robustness index, and generate a comprehensive feedback cost after weighted normalization and feed it back to the intelligent optimization engine; In this embodiment, the simulation data is quantitatively evaluated and an improved multi-scenario robustness index is constructed. After weighted normalization, a comprehensive feedback cost is generated and fed back to the intelligent optimization engine. Specifically: Calculate multiple performance metrics for various scenarios, including dynamic quality metrics (rise time). Adjusting time Overshoot Steady-state error Error integral indicators (integral squared error ISE, integral absolute error IAE, time-integral absolute error ITAE), control effectiveness and robustness indicators (root mean square value of control surface deflection). Multiple parameters including phase margin (PM) and gain margin (GM); For the same key performance indicator (such as ITAE) under different scenarios, calculate its standard deviation σ and mean. The ratio and the coefficient of dispersion (CD), where the coefficient of dispersion is the difference between the maximum and minimum values ​​of the key performance indicator in each scenario divided by the mean. ; The ratio of the standard deviation to the mean The improved multi-scenario robustness index is constructed by weighted summation of the dispersion coefficients CD. , ; in, For the preset weights, satisfy ; In this embodiment, it is assumed that the ITAE index of the current flight control parameter set in the five scenarios are 150, 180, 200, 220, and 250, respectively, with the mean value being... Standard deviation , ,but: ; After normalizing all performance metrics to a uniform interval [0,1] (for metrics where smaller is better, such as...), The normalization method is as follows: For metrics where larger is better (such as PM, GM), the normalization method is: ;in, and (The maximum and minimum values ​​of this indicator in all evaluations of the current iteration) are used to weight and sum the normalized indicators in each scenario to obtain the comprehensive cost of each scenario. , Then, the average cost of each scenario is taken and weighted and summed with the improved multi-scenario robustness index to generate the comprehensive feedback cost. Feedback is sent to the intelligent optimization engine; the overall feedback cost is considered. The formula expression is: ; in, For robust weights across multiple scenarios; Each scenario adopts a phased invocation strategy: in the first N generations of optimization iteration, a first number of representative scenarios are used for quick screening; after the N generations of optimization iteration, all scenarios are activated for fine evaluation, where N is preset by the user (in this embodiment, N=20). The weights of the overall cost for each scenario are adjusted using an adaptive strategy: during the optimization iteration process, the rate of change of each performance indicator with the number of iterations is recorded. When the improvement of any indicator within a preset number of iterations is lower than a preset threshold, the weight of that indicator in the overall cost for each scenario is increased. The specific implementation is as follows: Rate of change calculation: For each performance index j (a total of n indices), record its value in the most recent G generations ("preset generation" G, recommended to be 10). 20, or 5% of the total number of iterations. Normalized value sequence within 10% Where t is the current iteration number; The rate of change was calculated using the slope of the linear fit. (Methods such as mean difference can also be used): ; That is, fitting the linear trend of the time series using the least squares method, the rate of change This reflects the average rate of change of index j over the most recent G generations; the smaller the index, the better. This indicates that the indicator is improving (the normalized value is decreasing); if "the larger the better", it means the indicator is improving. This indicates that the indicator is improving (the normalized value is rising). b. Determine the preset algebraic number G: Defined as the "observation window length," i.e., the upper limit of the number of iterations for changes in statistical indicators. It is recommended to preset it according to the convergence speed of the optimization problem: Fast convergence problems (such as simple control parameter optimization): G=10; Slow convergence problems (such as multivariable, strongly coupled system optimization): G=20 30; it can also be set as a proportion of the total number of iterations (e.g., G=0.1×Ttotal, where Ttotal is the total number of iterations). Improve threshold For each indicator j, a preset "minimum effective improvement amount" is defined. (The setting needs to be combined with the physical meaning and normalization range of the indicator): For indicators that are "the smaller the better" (such as settling time ts, overshoot σ%, integral absolute error IAE, etc.): If (That is, the recent decrease in the normalized value of G is less than) If there are deficiencies (e.g., insufficient improvement), a weight adjustment will be triggered; for indicators where "the larger the better" (e.g., phase margin PM, gain margin GM): if (That is, the recent increase in the normalized value of G is less than) If deficiencies are not addressed, a weight adjustment will be triggered. in, Take 0.01 0.05 (because the normalized index range is [0,1], a small threshold can sensitively capture "no substantial improvement in a long period of time"). Weighting adjustment range and constraints: When indicator j meets the condition of "no improvement in the long term", its weight... Adjustments will be made according to the following rules: ; Where α is the weight adjustment coefficient (taken as 0.1). 0.2, meaning an increase of 10% each time. 20%, to avoid sudden weight changes); This represents the upper limit of the weight of indicator j (taken as 0.5). 0.8 ensures that no single metric excessively dominates the total cost, while other metrics still have room for optimization; initial weights We propose to take 1 / n (where n is the total number of indicators, ensuring uniform initial weights), and satisfy the following conditions: (After adjustment, all weights need to be renormalized to ensure the sum is 1). Weight update timing: Perform a weight check and adjustment once after each G-generation iteration (to avoid frequent adjustments that could cause algorithm oscillations). If adaptive costing is used in this assessment, the final feedback cost will be: ; If this assessment employs a severe penalty, then direct feedback will be provided. ; The results output module is configured to: filter the optimal parameter combination based on the comprehensive feedback cost, and generate an optimization report containing convergence curves and performance analysis; In this embodiment, the intelligent optimization engine determines the fitness of each individual in the population based on the received comprehensive feedback cost, updates the search direction accordingly, and generates candidate parameters for the next generation. The iterative process continues until a preset convergence condition is met: the improvement in the cost function is below a threshold for 10 consecutive generations. or reaching the maximum number of iterations. The output module records the set of flight control parameters with the lowest cost throughout the entire optimization history, and uses it as the global optimal solution. Output, and automatically generate an optimization convergence curve (horizontal axis is iteration number t, vertical axis is...). The report includes performance scores for each scenario (weighted average / original value of normalized indicators), final parameter values, and detailed reports on parameter sensitivity analysis (such as the impact of single-parameter perturbations on cost), for engineers to analyze. The improvement in the cost function is defined as follows: ; Let the first The global optimal cost of generation is (i.e., the minimum of all individuals / scenes in the current iteration) ,exclude (situation), then This indicates a decrease in cost and an improvement in performance; statistically, continuous satisfaction is achieved. The algebraic count is considered convergent when the continuous algebra reaches M, i.e., count ≥ M, where M (presumably the continuous algebra) is 10. 20 (Balance convergence sensitivity and stability to avoid premature / late termination); (Improvement Amount) Threshold: 0.001 0.01 (adjusted according to the magnitude of the cost function to ensure that "minor improvements" can be identified); the maximum number of iterations Nmax is reached, and the iteration is forcibly terminated when the number of iterations t≥Nmax, with Nmax set to 200. 500 (Complex problems require more iterations, while simple problems can have their iteration count reduced appropriately); After the iteration terminates, the results output module records the entire optimization history. The minimum set of flight control parameters is used as the global optimal solution. Output.

[0019] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related explanations of the methods described above can be found in the corresponding processes in the foregoing system embodiments, and will not be repeated here.

[0020] A second embodiment of the present invention proposes a flight control parameter cluster grouping optimization and constraint projection safety injection method. Based on the aforementioned flight control parameter cluster grouping optimization and constraint projection safety injection system, the method includes the following steps: S1: Initialize configuration, define the flight control parameter set and its feasible domain boundary, and define the multi-scenario test set; S2: Initialize the optimization process by dividing the flight control parameter set into multiple parameter clusters according to the physical modal response speed of the aircraft; and perform collaborative optimization on the multiple parameter clusters by exchanging information between different parameter clusters to generate candidate parameter vectors; S3: Parameter cluster secure injection and intelligent smooth mapping, perform boundary compliance checks on the candidate parameter vectors, select appropriate constraint processing paths for non-compliant parameters based on their attributes, and obtain parameter vectors that pass the boundary checks and violation measures; Based on the parameter vectors used in the current simulation, the parameter vectors that have passed the boundary check, and the real-time state data of the current simulation, the smooth transition time of each parameter is dynamically determined, and a corresponding smooth transition sequence is generated. The parameter values ​​in the smooth transition sequence are injected into the flight simulation model step by step according to the simulation step size; S4: Automated simulation and objective evaluation, driving the flight simulation model to run all test cases in the multi-scenario test set and recording simulation data; The simulation data is quantitatively evaluated and an improved multi-scenario robustness index is constructed. After weighted normalization, a comprehensive feedback cost is generated. S5: Optimize feedback and convergence judgment. Update the population and generate the next generation candidate parameter vector based on the comprehensive feedback cost. The process of updating the population includes: performing the collaborative optimization, updating the parameters by exchanging information between different parameter clusters based on the comprehensive feedback cost; iteratively executing steps S3 to S5 until the preset convergence condition is met. S6: Output the global optimal solution, select the optimal parameter combination based on the comprehensive feedback cost, and generate an optimization report containing the convergence curve and performance analysis.

[0021] It should be noted that the flight control parameter cluster grouping optimization and constraint projection safety injection method provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0022] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0023] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0024] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0025] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A flight control parameter cluster grouping optimization and constraint projection safety injection system, characterized in that, The system includes: The parameter and scenario configuration module is configured to: define the flight control parameter set and its feasible domain boundary, and define a multi-scenario test set; The intelligent optimization engine is configured to: divide the flight control parameter set into multiple parameter clusters according to the physical modal response speed of the aircraft; and perform collaborative optimization on the multiple parameter clusters to generate candidate parameter vectors by exchanging information between different parameter clusters; The parameter cluster management and security injection module is configured to: perform boundary compliance checks on the candidate parameter vectors; select appropriate constraint processing paths for non-compliant parameters based on their attributes; obtain parameter vectors that pass the boundary checks and violation metrics; and dynamically determine the smooth transition time of each parameter based on the parameter vectors used in the current simulation, the parameter vectors that pass the boundary checks, and the real-time state data of the current simulation, and generate corresponding smooth transition sequences. An automated simulation executor is configured to: inject parameter values ​​from the smooth transition sequence into the flight simulation model step by step according to the simulation step size, drive the flight simulation model to run all test cases in the multi-scenario test set, and record simulation data; The multi-dimensional objective performance evaluation module is configured to: quantitatively evaluate the simulation data and construct an improved multi-scenario robustness index, and generate a comprehensive feedback cost after weighted normalization and feed it back to the intelligent optimization engine; The results output module is configured to: filter the optimal parameter combination based on the comprehensive feedback cost, and generate an optimization report containing convergence curves and performance analysis.

2. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 1, characterized in that, The candidate parameter vectors are subjected to boundary compliance checks. For non-compliant parameters, appropriate constraint processing paths are selected based on their attributes to obtain parameter vectors that pass the boundary checks and violation metrics. Based on the parameter vectors used in the current simulation, the parameter vectors that pass the boundary checks, and the real-time state data of the current simulation, the smooth transition time of each parameter is dynamically determined, and a corresponding smooth transition sequence is generated, specifically as follows: Boundary violation detection is performed on the candidate parameter vector. For the violation parameter, the corresponding constraint processing path is selected according to the predefined attribute of the violation parameter to obtain the parameter vector that passes the boundary check and the violation amount. Based on the parameter vector used in the current simulation, the parameter vector that passes the boundary check, and the real-time state data of the current simulation, the dynamic activity index and the relative change of the parameters are calculated, and the smooth transition time of each parameter is dynamically determined according to the segmented decision logic. Based on the smooth transition time, a smooth transition sequence is generated using a transition curve generation unit; The constraint processing path includes: For ordinary parameters, the projection method is used to trim the non-compliant parameters into the feasible region and generate an adaptive penalty cost. The adaptive penalty cost increases with the number of iterations and is weighted by flight quality. For predefined critical sensitive parameters, immediately terminate the evaluation and return a preset maximum penalty to the intelligent optimization engine.

3. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 2, characterized in that, The parameter cluster management and security injection module also includes an intra-cluster association rule review unit, configured to: perform a dynamic rationality review on the parameter vectors that have passed the boundary check, the review content including one or more of the following: proportional-integral association constraints, lateral gain ratio constraints, differential gain upper limit constraints, stability margin constraints, integral saturation prevention constraints, cross-coupling suppression constraints, and energy dissipation constraints; when a rule is violated, a gradual buffer correction is performed, and if the correction fails, a preset maximum penalty cost is returned.

4. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 2, characterized in that, The segmented decision-making logic is as follows: When the relative change of the parameter does not exceed the first threshold and the dynamic activity index does not exceed the safety threshold, the benchmark smoothing time of the cluster to which the parameter belongs is adopted. The benchmark smoothing time is increased sequentially for fast-changing clusters, medium-changing clusters, and slow-changing clusters. When the relative change of the parameter is between the first threshold and the second threshold, based on the baseline smoothing time, a nonlinear extension is performed according to the difference between the relative change of the parameter and the first threshold and the difference between the dynamic activity index and the safety threshold, wherein the first threshold is less than the second threshold. When the relative change of the parameter exceeds the second threshold or the dynamic activity index exceeds the warning threshold, a preset maximum safe smoothing time is adopted.

5. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 2, characterized in that, The dynamic activity index is calculated in one of the following ways: Method 1, based on the weighted volatility of key state variables, specifically: the weighted L2 norm of the state variable after dividing the deviation of the instantaneous value of the key state variable from the benchmark stationary value by the historical maximum volatility range of the state variable is used as the dynamic activity index, wherein the weight of each state variable is determined by sensitivity analysis and the sum of the weights is 1. Alternatively, the calculation can be performed based on the sum of the real parts of the eigenvalues ​​of the system state matrix obtained by linearizing the flight simulation model under the current state.

6. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 1, characterized in that, The simulation data is quantitatively evaluated, and an improved multi-scenario robustness index is constructed. After weighted normalization, a comprehensive feedback cost is generated and fed back to the intelligent optimization engine. Specifically: Calculate multiple performance indicators for each scenario, including various dynamic quality indicators, error integral indicators, control effectiveness and robustness indicators; For the same key performance indicator of the same parameter set in different scenarios, calculate the ratio of its standard deviation to the mean and the coefficient of dispersion. The coefficient of dispersion is the difference between the maximum and minimum values ​​of the key performance indicator in each scenario divided by the mean. An improved multi-scenario robustness index is constructed by weighting and summing the ratio of the standard deviation to the mean and the dispersion coefficient. After normalizing all performance indicators to a unified range, the normalized indicators in each scenario are weighted and summed to obtain the comprehensive cost of each scenario. Then, the average value of the comprehensive cost of each scenario is taken and weighted and summed with the improved multi-scenario robustness indicator to generate a comprehensive feedback cost that is fed back to the intelligent optimization engine. Each scenario adopts a phased invocation strategy: in the first N generations of optimization iterations, a first number of representative scenarios are used for rapid screening; after the N generations of optimization iterations, all scenarios are activated for fine evaluation, where N is preset by the user. The weights of the comprehensive costs of each scenario adopt an adaptive adjustment strategy: during the optimization iteration process, the rate of change of each performance index with the number of iterations is recorded. When the improvement of any index within a preset number of iterations is lower than a preset threshold, the weight of that index in the comprehensive costs of each scenario is increased.

7. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 1, characterized in that, The flight control parameter set is divided into multiple parameter clusters based on the aircraft's physical modal response velocity; and collaborative optimization is performed on these multiple parameter clusters, generating candidate parameter vectors through information exchange between different parameter clusters, specifically: The flight control parameter set is divided into fast-changing cluster, medium-changing cluster and slow-changing cluster according to the physical modal response speed of the aircraft, and the migration weight of the three clusters decreases step by step. A periodic elite individual migration mechanism between the three clusters is executed, and the medium-changing cluster simultaneously receives the optimal solution information from the fast-changing cluster and the slow-changing cluster to obtain the candidate parameter vector. The migration weights of the three clusters decrease sequentially as follows: the migration weight of the fast-changing cluster is greater than that of the medium-changing cluster, and the migration weight of the medium-changing cluster is greater than that of the slow-changing cluster; the migration weight of the fast-changing cluster is the largest, so as to reflect the physical causal relationship between high-frequency dynamics and medium-frequency attitude response and low-frequency trajectory tracking. When the medium-variable cluster receives information about elite individuals, the weight of information received from the fast-variable cluster is greater than the weight of information received from the slow-variable cluster.

8. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 2, characterized in that, The smooth transition sequence generated by the transition curve generation unit has monotonicity and boundedness: the parameter values ​​in the smooth transition sequence transition monotonically from the parameter values ​​applied in the current simulation to the parameter vector that has passed the boundary check, and the parameter values ​​are always within the boundary of the feasible region throughout the transition, realizing a smooth, continuous, and differentiable transition of parameters from the current value to the target value; The specific implementation methods of the transition curve generation unit include any one of S-curve mapping, linear transition, exponential transition, first-order low-pass filtering, or polynomial trajectory planning.

9. The flight control parameter cluster grouping optimization and constraint projection safety injection system according to claim 2, characterized in that, The adaptive penalty cost is specifically defined as follows: the penalty intensity of the adaptive penalty cost is equal to the sum of the basic penalty coefficient and the term that increases linearly with the number of iterations, multiplied by the square of the violation amount, so that the penalty for boundary violations in the later stage of optimization is stronger than that in the early stage of optimization; the penalty intensity is multiplied by the differentiated weight set based on the flight quality standard, so that the penalty for violations of key flight parameters is higher than that for ordinary parameters; when the proportion of feasible solutions is lower than a preset threshold, the penalty intensity is automatically increased or an additional amplification factor is introduced to force the intelligent optimization engine to move away from the infeasible region; The intelligent optimization engine is implemented using any one of particle swarm optimization, genetic algorithm, differential evolution, or Bayesian optimization: when particle swarm optimization is used, the candidate parameter vector is the particle position vector; when genetic algorithm or differential evolution is used, the candidate parameter vector is the population individual vector; when Bayesian optimization is used, the candidate parameter vector is the evaluation point recommended by the acquisition function.

10. A method for optimizing flight control parameter clusters and constrained projection secure injection, based on the flight control parameter cluster optimization and constrained projection secure injection system according to any one of claims 1-9, characterized in that, The method includes the following steps: S1: Initialize configuration, define the flight control parameter set and its feasible domain boundary, and define the multi-scenario test set; S2: Initialize the optimization process by dividing the flight control parameter set into multiple parameter clusters according to the physical modal response speed of the aircraft; and perform collaborative optimization on the multiple parameter clusters by exchanging information between different parameter clusters to generate candidate parameter vectors; S3: Parameter cluster secure injection and intelligent smooth mapping, perform boundary compliance checks on the candidate parameter vectors, select appropriate constraint processing paths for non-compliant parameters based on their attributes, and obtain parameter vectors that pass the boundary checks and violation measures; Based on the parameter vectors used in the current simulation, the parameter vectors that have passed the boundary check, and the real-time state data of the current simulation, the smooth transition time of each parameter is dynamically determined, and a corresponding smooth transition sequence is generated. The parameter values ​​in the smooth transition sequence are injected into the flight simulation model step by step according to the simulation step size; S4: Automated simulation and objective evaluation, driving the flight simulation model to run all test cases in the multi-scenario test set and recording simulation data; The simulation data is quantitatively evaluated and an improved multi-scenario robustness index is constructed. After weighted normalization, a comprehensive feedback cost is generated. S5: Optimize feedback and convergence judgment. Update the population and generate the next generation candidate parameter vector based on the comprehensive feedback cost. The process of updating the population includes: performing the collaborative optimization, updating the parameters by exchanging information between different parameter clusters based on the comprehensive feedback cost; iteratively executing steps S3 to S5 until the preset convergence condition is met. S6: Output the global optimal solution, select the optimal parameter combination based on the comprehensive feedback cost, and generate an optimization report containing the convergence curve and performance analysis.

Images