Method for optimizing parameters of angle of approach rate controller based on improved pigeon optimization

Abstract

The application discloses a method for optimizing a landing path angle rate controller parameter based on improved pigeon swarm optimization, and comprises the following steps: step one, determining parameters to be optimized of a three-channel active disturbance rejection controller; step two, initializing a population by adopting a hybrid strategy of 90% Latin hypercube sampling+10% random sampling; step three, improving a map and compass operator, introducing adaptive weights and Gaussian disturbance, and dynamically adjusting parameters in stages; step four, proposing a hierarchical collaborative evolution landmark operator; and step five, outputting a controller parameter combination with optimal fitness after iteration is completed. The method does not need to rely on artificial experience to try parameters, solves the problems of low efficiency and difficulty in searching for an optimal solution in traditional manual parameter adjustment, avoids convergence deviation caused by uneven initial population distribution by adopting a hybrid initialization strategy, and improves the stability of an optimization result. The hierarchical collaborative evolution landmark operator realizes the cooperation of elite guidance, fine development and diversity maintenance, improves parameter optimization precision and operation efficiency, and improves the overall performance of a shipboard aircraft landing control system.

Description

Technical Field

[0001] This invention relates to the field of carrier-based aircraft control and its intelligent optimization technology, specifically to a method for optimizing the parameters of a landing trajectory angular rate controller based on improved pigeon flock optimization. Background Technology

[0002] Aircraft carrier battle groups, as one of the symbols of a major power's military strength, hold a pivotal strategic position. However, the landing of carrier-based aircraft on aircraft carriers is considered one of the most difficult tasks in carrier operations, limiting the carrier's combat effectiveness. To solve this problem, the development of a landing control system has become a focus of attention for many researchers.

[0003] This invention is based on a carrier-based aircraft landing trajectory angular rate controller designed with active disturbance rejection technology, and its specific structural block diagram is as follows. Figure 1 As shown, this control system has a large number of parameters to be adjusted, which affect the performance of the entire control system. Traditional parameter tuning methods are inefficient and it is difficult to find the optimal parameters. Therefore, in recent years, many scholars have adopted intelligent swarm optimization methods to optimize the parameters of the control system.

[0004] The Pigeon-inspired Optimization (PIO) algorithm was first proposed by Professor Duan Haibin in 2014 and applied to the path planning problem of aerial robots, effectively improving the planning results.

[0005] To ensure higher performance of the carrier-based aircraft direct lift trajectory angular rate control system, this invention proposes a parameter optimization method for the carrier landing trajectory angular rate controller based on improved pigeon flock optimization.

[0006] To address the aforementioned problems, this invention proposes a method for optimizing the parameters of the carrier landing trajectory angular rate controller based on improved pigeon flock optimization. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method for optimizing the parameters of the carrier landing trajectory angular rate controller based on improved pigeon flock optimization, thereby solving the aforementioned problems.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: a method for optimizing the parameters of a carrier landing trajectory angular rate controller based on improved pigeon flock optimization, comprising the following steps: S1: determining the control system structure to be optimized and determining the parameters to be adjusted in the system.

[0009] S2: Perform initialization operations; This invention employs a hybrid initialization strategy of 90% Latin hypercube sampling and 10% random sampling to achieve uniform coverage of the solution space while ensuring population randomness. The specific implementation steps of the hybrid initialization are as follows: Initialize the population size to N Optimize the problem dimensions as D The upper and lower limits of the solution space are lb , ub .

[0010] No. i The position vector and velocity vector of a pigeon are defined as follows: (1) In the formula, i =1,2,…, N This is the identifier for an individual within the population. x id , v id The first i Only one pigeon in the 1st d Position and velocity within the 3D search space.

[0011] The population was divided into the following categories: 90% of individuals were sampled using LHS (Low-Hyper-Side Spectroscopy) and the remaining 10% were sampled randomly. (2) Then, the two types of individuals were initialized using different methods, as shown in the following formula: (3) In the formula, lhsdesign(·) is the Latin hypercube sampling function.

[0012] S3: The flock of pigeons updates its status based on the map and compass operators; The map and compass operators simulate how pigeons navigate using the sun and the Earth's magnetic field. Pigeons sense the Earth's magnetic field through their own magnetoreceptors and create a virtual map in their brains. They use the sun as a compass to adjust their flight direction. As they approach their destination, their reliance on sunlight and the Earth's magnetic field gradually decreases.

[0013] This invention improves the velocity update formula for map and compass operators by incorporating adaptive weights and Gaussian perturbations. It dynamically adjusts the globally optimal guiding weights and random perturbation intensity based on the iteration process, balancing the algorithm's convergence speed with population diversity. The total number of iterations in the global search phase is defined as... T 1, no. N c iteration (1≤ N c ≤ T 1) The improved speed and position update formulas are as follows: (4) In the formula, , They are the first i The individual N c The speed and position of the next iteration; X gbest This represents the globally optimal individual position in the current population. R For map and compass operators; rand is a uniformly distributed random number in the interval [0,1], and randn(·) is a standard normally distributed random number; w 1 represents the globally optimal guiding weight. w 2 represents the random perturbation weight. S perturb The intensity of the random perturbation is determined, and all three factors are adaptively adjusted as the iteration progresses, according to the following rules: The first 1 / 3 of the iteration ( N c ≤ T 1 / 3): The core objective of the algorithm is to converge quickly to the potential optimal region. Therefore, random perturbations are turned off, and only the globally optimal guidance is retained, i.e.: (5) After the second / third stage of iteration ( T 1 / 3< N c ≤ T 1) The core objective of the algorithm is to maintain population diversity while achieving convergence, avoiding premature convergence. The weights and perturbation strength are adjusted linearly and smoothly as the iteration progresses. (6) In the formula, τ This represents the normalized progress for the second / third stage after iteration, with a value range of [0,1].

[0014] For dimensions whose positions exceed the upper and lower bounds of the solution space after the update, constrain their positions to the boundary values ​​and set the velocity of that dimension to zero. The map and compass operator phase is complete, and then proceed to step four.

[0015] S4: The flock of pigeons updates its status based on the landmark operator; This invention proposes an improved landmark operator strategy based on hierarchical co-evolution. The core improvements include: eliminating the fixed-proportion population elimination rule, keeping the population size constant throughout the local search phase; dividing the population into three layers based on individual fitness ranking, designing differentiated position update strategies for each layer, and realizing a co-evolution mechanism of "elite layer guiding convergence, intermediate layer refining development, and bottom layer maintaining population diversity"; and introducing a pattern search algorithm to improve local optimization accuracy, adapting to the high-precision tuning requirements of landing ADRC parameters.

[0016] Define the total number of iterations in the local search phase as T 2, no. N citeration (1≤ N c ≤ T The specific implementation steps for 2) are as follows: S41: Stratify the population; First, sort all individuals in the population in ascending order (for a minimization optimization problem), and set the population size to . N The population is divided into three layers in a 1:2:7 ratio: the elite layer, the local development layer, and the global exploration layer. The elite layer consists of individuals with the top 10% fitness, whose core role is to guide population convergence and provide a basic direction for local optimization. The local development layer consists of individuals with the top 10%-30% fitness, whose core role is to complete refined searches. The global exploration layer consists of individuals with the bottom 70% fitness, whose core role is to randomly explore within the global solution space and maintain population diversity.

[0017] Meanwhile, to reduce repeated calls to the fitness function and improve algorithm efficiency, the fitness function is only called in the first and second iterations of the local search phase. T In the 2 / 2 iteration, the fitness of the entire population is sorted, and the stratification rules remain unchanged at other times.

[0018] S42: Update the position of the elite layer; The core objective of the elite layer is to achieve high-precision local convergence. Therefore, only the weighted centroids of the top 10% of elite individuals are used as guides to replace the centroids of the entire population in the basic PIO, thus avoiding interference from inferior individuals on the convergence direction.

[0019] First, calculate the weighted centroid of the elite layer. Using the inverse of individual fitness as the weight, it is ensured that individuals with better fitness contribute more to the centroid: (7) In the formula, F ( X i ) is the first i The fitness value of an elite individual. e 8 To avoid the minimum value where the denominator is 0.

[0020] Update the position of elite individuals: (8) In the formula, S current To adapt to the amplitude of random perturbations, the amplitude decays linearly with the iteration process, balancing early exploration and later development: (9) In the formula, Sinit =( ub lb ) / 20 represents the initial disturbance amplitude. S end =( ub lb ) / 100 represents the perturbation magnitude at the end of the iteration.

[0021] S43: Update the location of the local development layer; The core objective of the local development layer is to achieve refined local optimization within a relatively optimal region. Therefore, a coordinate pattern search algorithm is introduced to replace the centroid-guided update rule of the basic PIO. This method is a gradient-free direct search optimization method that does not rely on the gradient information of the objective function. It can achieve optimization by comparing fitness values ​​alone, making it perfectly suited for black-box optimization scenarios without explicit gradients, such as ADRC controller parameter tuning.

[0022] For the local development layer i Each individual, based on their current location As the initial starting point for the search X current Define the core parameters for pattern search: optimize the problem dimensions. D The feasible solution space j The upper and lower bounds are lb j , ub j ;No. j The initial search step size is step init,j =( ub j lb j ) / 10, minimum convergence step size step min,j =10 6 Maintain a step-by-step state matrix for each individual as follows: step state ( i , j ), record the i The individual j The current search step size of the dimension.

[0023] Pattern search starts from the initial starting point and explores in both positive and negative directions sequentially according to the dimension. j Two test points are generated using the following formula: (10) In the formula, stepstate ( i , j ) is the first i The individual j The current search step size of the dimension, e j For the first j A 3D unit coordinate vector. To ensure the feasibility of the test points, boundary limiting is applied to each test point: (11) Calculate the number in sequence j The fitness values ​​of two trial points (positive and negative) are used to select the trial point with the best fitness in that dimension. X best,j Its fitness value is F best,j ,satisfy: (12) In the formula F (·) represents the fitness function of the problem to be optimized. Based on the fitness of the optimal exploration point, the search step size is adaptively adjusted and the search starting point is updated: if a better solution is found, the search starting point is updated and the step size is maintained to expand the search range; if no better solution is found, the step size is reduced by a factor of 0.5 to shrink the search granularity. The specific rules are as follows: (13) In the formula F current This is the fitness value at the current search starting point. During the dimensional search process, if the current step size of a certain dimension is less than the minimum step size of that dimension... step min,j If the search fails, skip that dimension; after traversing all dimensions, use the final optimal search starting point as the update position for this individual in this iteration. .

[0024] S44: Update the position of the global exploration layer; The core objective of the global exploration layer is to maintain population diversity and avoid the algorithm getting trapped in local optima. Therefore, it uses purely uniform random sampling within the total solution space to update positions, without any convergence guidance. (14) S45: Apply boundary constraints uniformly to all individuals in the entire population, update the historical best position and global best position of each individual, and record the convergence curve data of this iteration. If the current iteration number is less than [number missing], [then the process continues]. T If 2, then jump to sub-step one and continue iterating.

[0025] S5: Output the individual with the best fitness function value in the pigeon flock, and use its corresponding value as the controller parameter.

[0026] The present invention has the following beneficial effects: The method adopts an improved pigeon flock optimization algorithm to optimize the controller parameters, without relying on manual experience to try and match parameters, and the traversal of the controller parameter search space is more comprehensive. It solves the problems of low efficiency and difficulty in finding the optimal parameters in traditional manual parameter tuning, as well as the insufficient adaptability of basic optimization algorithms to complex parameter spaces.

[0027] A hybrid initialization strategy of 90% Latin hypercube sampling and 10% random sampling is adopted to ensure the randomness of the population while achieving uniform coverage of the solution space, avoiding convergence bias caused by uneven initial population distribution, and improving the stability of the optimization results.

[0028] Adaptive weights and Gaussian perturbations are introduced to improve the map and compass operators. The global guiding weights and perturbation intensity are dynamically adjusted in stages, balancing the algorithm's global search capability and convergence speed, and effectively avoiding premature convergence.

[0029] We propose an improved landmark operator strategy based on hierarchical co-evolution. The population is divided into three layers according to fitness, and differentiated update rules are designed. Fixed-proportion population elimination is eliminated. At the same time, a pattern search algorithm is introduced to improve the accuracy of local optimization. Interval sorting reduces repeated calls to the fitness function. This achieves the synergy of elite guidance, refined development and diversity maintenance, and significantly improves the accuracy of parameter optimization and the efficiency of algorithm operation.

[0030] The optimized controller has higher tracking accuracy and faster response under undisturbed conditions, and is more robust in complex disturbance environments such as the wake turbulence, effectively improving the overall performance and safety of the carrier-based aircraft landing control system. Attached Figure Description

[0031] Figure 1 The block diagram of the carrier-based aircraft landing trajectory angular rate controller based on active disturbance rejection for parameter optimization required by the present invention is shown. Figure 2 The image shows the trajectory angular rate response curves of the PID control system, the ADRC control system optimized using PIO, and the ADRC control system optimized by EPIO proposed in this invention during a disturbance-free simulation experiment. Figure 3 The flight path angle, pitch angle, and angle of attack curves of the PID control system, the ADRC control system optimized by PIO, and the ADRC control system optimized by EPIO proposed in this invention are shown in the uninterrupted simulation experiment. Figure 4 The actuator curves are shown for the PID control system, the ADRC control system optimized using PIO, and the ADRC control system optimized by EPIO proposed in this invention in the uninterrupted simulation experiment. Figure 5The speed, pitch rate, and overload curves of the PID control system, the ADRC control system optimized using PIO, and the ADRC control system optimized by EPIO proposed in this invention are shown in the uninterrupted simulation experiment. Figure 6 The trajectory angular rate response curves of the PID control system, the ADRC control system optimized with PIO, and the ADRC control system optimized with EPIO proposed in this invention are shown in the disturbance simulation experiment. Figure 7 The flight path angle, pitch angle, and angle of attack curves of the PID control system, the ADRC control system optimized by PIO, and the ADRC control system optimized by EPIO proposed in this invention are shown in the disturbance simulation experiment. Figure 8 The actuator curves are shown for the PID control system, the ADRC control system optimized using PIO, and the ADRC control system optimized by EPIO proposed in this invention in the disturbance simulation experiment. Figure 9 The speed, pitch rate, and overload curves of the PID control system, the ADRC control system optimized using PIO, and the ADRC control system optimized by EPIO proposed in this invention are presented in the disturbance simulation experiment. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figures 1-9 This invention provides a technical solution: a method for optimizing the parameters of a carrier landing trajectory angular rate controller based on improved pigeon flock optimization, specifically including the following steps: Step S1: Determine the control system structure that needs optimization and identify the parameters to be adjusted in the system. In this embodiment, the control system is as follows: Figure 1 As shown: This control system comprises three independent channels, each with a clearly defined function and working collaboratively: the flap channel is primarily responsible for controlling the aircraft's trajectory angle rate during landing. By receiving trajectory angle rate commands from the pilot, it adjusts the flap deflection angle to alter lift, thereby achieving precise control of the trajectory angle; the horizontal stabilizer channel is responsible for angle-of-attack maintenance, adjusting the horizontal stabilizer deflection to balance the pitching moment generated by other mechanisms, ensuring the angle of attack remains stable within a preset range. To meet the direct lift control requirements for balancing torque, this channel incorporates a direct lift decoupling module, achieving decoupled control of the trajectory angle and angle of attack through the coordinated deflection of the flaps and horizontal stabilizer; the throttle channel is responsible for power compensation, adjusting engine thrust to maintain stable aircraft speed, providing a stable power foundation for trajectory angle and angle of attack control. Each channel employs active disturbance rejection control technology to address model uncertainties and external disturbances during aircraft landing.

[0034] In the above control system, all parameters to be adjusted are shown in the table below.

[0035] Table 1. Control System Parameters to be Adjusted

[0036] Step S2: Perform initialization operations.

[0037] Initialize the population size to N Optimize the problem dimensions as D The upper and lower limits of the solution space are lb , ub .

[0038] No. i The position vector and velocity vector of a pigeon are defined as follows: (15) In the formula, i =1,2,…, N This is the identifier for an individual within the population. x id , v id The first i Only one pigeon in the 1st d Position and velocity within the 3D search space.

[0039] The population was divided into the following categories: 90% of individuals were sampled using LHS (Low-Hyper-Side Spectroscopy) and the remaining 10% were sampled randomly. (16) Then, the two types of individuals were initialized using different methods, as shown in the following formula: (17) In the formula, lhsdesign(·) is the Latin hypercube sampling function.

[0040] Fitness function in this embodiment F (·)= J Defined as a weighted sum of the undisturbed operating condition and the wake disturbance operating condition, specifically implemented as follows: 1) Calculation of single-condition performance indicators Let the total simulation time be T s The tracking error of each control variable is defined as follows: Tracking error of angular rate: (18) in, γ ref ( t () is the track angular rate command. γ ( t ( ) represents the actual angular rate of the flight path. Angle of attack maintenance error: (19) in, α ref ( t () is a fixed reference angle of attack command. α ( t () represents the actual angle of attack.

[0041] Speed ​​holding error: (20) in, V ref ( t () is a fixed reference speed command. V ( t () represents the actual speed.

[0042] Comprehensive performance indicators under single operating conditions J single The weighted sum of the three errors is expressed as: (twenty one) In the formula, w γ , w α , w V These are the weighting coefficients for trajectory angular rate, angle of attack, and velocity, respectively, with the weights set according to the priority of the landing control target.

[0043] 2) Overall fitness function Let the performance index of the undisturbed nominal operating condition be... Jnom The performance indicators for the wake disturbance condition are as follows: J dist Total fitness function J The weighted sum of the dual-condition indicators is expressed as: (twenty two) In the formula, w nom , w dist Weighting coefficients for nominal operating condition and disturbance operating condition, respectively. Step S3: The flock of pigeons updates its status based on the map and compass operator.

[0044] Define the total number of iterations in the global search phase as T 1, no. N c iteration (1≤ N c ≤ T 1) The improved speed and position update formulas are as follows: (twenty three) In the formula, , They are the first i The individual N c The speed and position of the next iteration; X gbest This represents the globally optimal individual position in the current population. R For map and compass operators; rand is a uniformly distributed random number in the interval [0,1], and randn(·) is a standard normally distributed random number; w 1 represents the globally optimal guiding weight. w 2 represents the random perturbation weight. S perturb The intensity of the random perturbation is determined, and all three factors are adaptively adjusted as the iteration progresses, according to the following rules: The first 1 / 3 of the iteration ( N c ≤ T 1 / 3): The core objective of the algorithm is to converge quickly to the potential optimal region. Therefore, random perturbations are turned off, and only the globally optimal guidance is retained, i.e.: (twenty four) After the second / third stage of iteration ( T 1 / 3< N c ≤ T 1) The core objective of the algorithm is to maintain population diversity while achieving convergence, avoiding premature convergence. The weights and perturbation strength are adjusted linearly and smoothly as the iteration progresses. (25) In the formula, τ This represents the normalized progress for the second / third stage after iteration, with a value range of [0,1].

[0045] For dimensions whose positions exceed the upper and lower bounds of the solution space after the update, constrain their positions to the boundary values ​​and set the velocity of that dimension to zero. The map and compass operator phase is complete, and then proceed to step four.

[0046] Step S4: The flock of pigeons updates its status based on the landmark operator.

[0047] Define the total number of iterations in the local search phase as T 2, no. N c iteration (1≤ N c ≤ T The specific implementation steps for 2) are as follows: Step S41: Stratify the population.

[0048] First, sort all individuals in the population in ascending order (for a minimization optimization problem), and set the population size to . N The population is divided into three layers in a 1:2:7 ratio: the elite layer, the local development layer, and the global exploration layer. The elite layer consists of individuals with the top 10% fitness, whose core role is to guide population convergence and provide a basic direction for local optimization. The local development layer consists of individuals with the top 10%-30% fitness, whose core role is to complete refined searches. The global exploration layer consists of individuals with the bottom 70% fitness, whose core role is to randomly explore within the global solution space and maintain population diversity.

[0049] Meanwhile, to reduce repeated calls to the fitness function and improve algorithm efficiency, the fitness function is only called in the first and second iterations of the local search phase. T In the 2 / 2 iteration, the fitness of the entire population is sorted, and the stratification rules remain unchanged at other times.

[0050] Step S42: Update the position of the elite layer.

[0051] First, calculate the weighted centroid of the elite layer. Using the inverse of individual fitness as the weight, it is ensured that individuals with better fitness contribute more to the centroid: (26) In the formula, F ( X i ) is the first i The fitness value of an elite individual. e 8To avoid the minimum value where the denominator is 0.

[0052] Update the position of elite individuals: (27) In the formula, S current To adapt to the amplitude of random perturbations, the amplitude decays linearly with the iteration process, balancing early exploration and later development: (28) In the formula, S init =( ub lb ) / 20 represents the initial disturbance amplitude. S end =( ub lb ) / 100 represents the perturbation magnitude at the end of the iteration.

[0053] Step S43: Update the location of the local development layer.

[0054] For the local development layer i Each individual, based on their current location As the initial starting point for the search X current Define the core parameters for pattern search: optimize the problem dimensions. D The feasible solution space j The upper and lower bounds are lb j , ub j ;No. j The initial search step size is step init,j =( ub j lb j ) / 10, minimum convergence step size step min,j =10 6 Maintain a step-by-step state matrix for each individual as follows: step state ( i , j ), record the i The individual j The current search step size of the dimension.

[0055] Pattern search starts from the initial starting point and explores in both positive and negative directions sequentially according to the dimension. jTwo test points are generated using the following formula: (29) In the formula, step state ( i , j ) is the first i The individual j The current search step size of the dimension, e j For the first j A 3D unit coordinate vector. To ensure the feasibility of the test points, boundary limiting is applied to each test point: (30) Calculate the number in sequence j The fitness values ​​of two trial points (positive and negative) are used to select the trial point with the best fitness in that dimension. X best,j Its fitness value is F best,j ,satisfy: (31) In the formula F (·) represents the fitness function of the problem to be optimized. Based on the fitness of the optimal exploration point, the search step size is adaptively adjusted and the search starting point is updated: if a better solution is found, the search starting point is updated and the step size is maintained to expand the search range; if no better solution is found, the step size is reduced by a factor of 0.5 to shrink the search granularity. The specific rules are as follows: (32) In the formula F current This is the fitness value at the current search starting point. During the dimensional search process, if the current step size of a certain dimension is less than the minimum step size of that dimension... step min,j If the search fails, skip that dimension; after traversing all dimensions, use the final optimal search starting point as the update position for this individual in this iteration. .

[0056] Step S44: Update the position of the global exploration layer.

[0057] Update global exploration layer pigeon flock locations: (33) Step S45: Apply boundary constraints to all individuals in the entire population, update the historical best position and global best position of each individual, and record the convergence curve data for this iteration. If the current iteration number is less than [number missing], [then the process continues]. T If 2, then jump to sub-step one and continue iterating.

[0058] Step 5: Output the individual with the best fitness function value in the pigeon flock, and use its corresponding value as the controller parameter.

[0059] Next, simulation experiments were conducted to verify the performance of the proposed algorithm. The experiments consisted of three sets of control schemes: a direct lift landing control system with PID control, an ADRC direct lift landing control system with optimized parameters based on the basic PIO algorithm, and an ADRC direct lift landing control system with optimized parameters based on the EPIO algorithm proposed in this invention. The following two types of experiments were performed under the same trajectory angular rate command input: The first type was a performance comparison experiment under the ideal, disturbance-free nominal operating condition, verifying and comparing the step response characteristics, trajectory tracking accuracy, and control decoupling effect of the control system in an ideal, disturbance-free environment; the second type was a simulation experiment on the anti-disturbance performance in a complex disturbance environment, introducing typical landing disturbances such as the wake turbulence to verify the disturbance suppression capability and robust stability of the control system.

[0060] The results and analysis of the undisturbed experiment are as follows: Figure 2 Compared to the significant fluctuations and slow convergence of traditional PID control, the basic PIO-optimized ADRC achieves better command tracking, realizing a fast, overshoot-free, and steady-state error-free response to the trajectory angular rate, with dynamic performance significantly superior to PID. The EPIO-optimized ADRC controller designed in this paper further improves tracking accuracy, enabling high-precision tracking of the input trajectory angular rate command. Figure 3 Under PID control, the angle of attack oscillated significantly around the setpoint; the ADRC angle of attack fluctuation amplitude under basic PIO optimization was significantly reduced; while the ADRC angle of attack under EPIO optimization in this paper remained stable around the reference value of 8.1°, with a fluctuation amplitude of no more than 0.1°, demonstrating optimal angle of attack maintenance capability. Furthermore, because the carrier-based aircraft under PID control could not quickly track the track angle rate command, the track angle convergence was slow. Figure 4 Under ADRC control, each actuator can respond quickly to the trajectory angular rate command, the deflection process is smooth and can quickly converge to a steady state value, without large fluctuations or chattering, meeting the physical constraint requirements of the actuators. Figure 5 The PID controller has relatively good dynamic performance in the speed holding stage, but it cannot converge to a stable value quickly. The speed fluctuation amplitude of the ADRC optimized by the basic PIO is about 0.01 m / s. The speed fluctuation amplitude under the ADRC control optimized by EPIO in this paper can also be controlled within 0.01 m / s, which fully meets the technical standards for carrier-based aircraft landing conditions. At the same time, the pitch rate stability is optimal, further verifying the decoupling control effect of the system.

[0061] The results and analysis of the perturbation experiment are as follows: Figure 6Under the influence of stern current disturbances, although the basic PIO-optimized ADRC system exhibited worse disturbance rejection performance than the other two methods, it could maintain the track angular rate near 0 without significant low-frequency fluctuations. The PID-controlled system showed significant fluctuations and could not maintain a position near the setpoint. The EPIO-optimized ADRC controller demonstrated stronger disturbance rejection performance, quickly suppressing the effects of external disturbances and consistently maintaining high-precision tracking of the track angular rate command. Figure 7 Under PID control, the angle of attack fluctuates the most under disturbances, resulting in decreased attitude stability. The ADRC angle of attack fluctuation amplitude optimized by basic PIO is approximately 0.2°, exhibiting better disturbance rejection capability than PID. Furthermore, the ADRC angle of attack fluctuation amplitude optimized by EPIO in this paper can be controlled within 0.05°, maintaining excellent attitude holding capability even under complex disturbance environments. Figure 8 The actuator under PID control exhibited the largest oscillation under disturbance, followed by the ADRC optimized by PIO, while the ADRC optimized by EPIO showed the smallest oscillation. This indicates that the ADRC optimized by EPIO can better manipulate the actuator to quickly bring the control quantity close to the setpoint. Figure 9 In terms of speed, all three controllers can keep the speed close to the set value, which meets the requirements. However, in terms of pitch rate, the PID controller fluctuates significantly under the influence of disturbances, and the ADRC optimized by the basic PIO controller is significantly more stable than the PID controller. The pitch rate and overload fluctuations under the ADRC control optimized by the EPIO controller in this paper are the smallest, and it shows better robust stability in complex landing environments.

[0062] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for optimizing the parameters of a carrier landing trajectory angular rate controller based on improved pigeon flocking optimization, characterized in that: Specifically, the following steps are included: S1: Determine the control system structure that needs to be optimized and determine the parameters to be adjusted in the system; S2: Perform initialization: Use a hybrid initialization strategy of 90% Latin hypercube sampling and 10% random sampling to achieve uniform coverage of the solution space while ensuring the randomness of the population. S3: Pigeon flock updates status based on map and compass operators: For map and compass operators, adaptive weights and Gaussian perturbations are introduced, and parameters are dynamically adjusted in stages to balance the algorithm's convergence speed and population diversity. S4: Pigeon flock updates status based on landmark operator: a hierarchical co-evolutionary landmark operator is proposed, which divides the population into three layers proportionally and adopts differentiated update rules, and introduces pattern search to improve the accuracy of local optimization. S5: Output the individual with the best fitness function value in the pigeon flock, and use its corresponding value as the controller parameter.

2. The method for optimizing landing trajectory angular rate controller parameters based on improved pigeon flock optimization according to claim 1, characterized in that: In step S2, the specific implementation steps of the mixed initialization are as follows: Initialize the population size to N Optimize the problem dimensions as D The upper and lower limits of the solution space are lb , ub No. i The position vector and velocity vector of a pigeon are defined as follows: (1) In the formula, i =1,2,…, N This is the identifier for an individual within the population. x id , v id The first i Only one pigeon in the 1st d Position and velocity within the 3D search space; The population was divided into the following categories: 90% of individuals were sampled using LHS (Low-Hyper-Side Spectroscopy) and the remaining 10% were sampled randomly. (2) Then, the two types of individuals were initialized using different methods, as shown in the following formula: (3) In the formula, lhsdesign(·) is the Latin hypercube sampling function.

3. The method for optimizing landing trajectory angular rate controller parameters based on improved pigeon flock optimization according to claim 2, characterized in that: In step S3, the total number of iterations in the global search phase is defined as follows: T 1, no. N c iteration (1≤ N c ≤ T 1) The improved speed and position update formulas are as follows: (4) In the formula, , They are the first i The individual N c The speed and position of the next iteration; X gbest This represents the globally optimal individual position in the current population. R For map and compass operators; rand is a uniformly distributed random number in the interval [0,1], and randn(·) is a standard normally distributed random number; w 1 represents the globally optimal guiding weight. w 2 represents the random perturbation weight. S perturb The intensity of the random perturbation is determined, and all three factors are adaptively adjusted as the iteration progresses, according to the following rules: The first 1 / 3 of the iteration ( N c ≤ T 1 / 3): The core objective of the algorithm is to converge quickly to the potential optimal region. Therefore, random perturbations are turned off, and only the globally optimal guidance is retained, i.e.: (5) After the second / third stage of iteration ( T 1 / 3< N c ≤ T 1) The core objective of the algorithm is to maintain population diversity while achieving convergence, avoiding premature convergence. The weights and perturbation strength are adjusted linearly and smoothly as the iteration progresses. (6) In the formula, τ This represents the normalized progress for the second / third stage after iteration, with a value range of [0,1]. For dimensions whose positions exceed the upper and lower bounds of the solution space after the position update, constrain their positions to the boundary values ​​and set the velocity of that dimension to zero; the map and compass operator phase is completed.

4. The method for optimizing landing trajectory angular rate controller parameters based on improved pigeon flock optimization according to claim 3, characterized in that: In step S4, the total number of iterations in the local search phase is defined as follows: T 2, no. N c iteration (1≤ N c ≤ T The specific implementation steps for 2) are as follows: Step S41: Stratify the population: First, sort all individuals in the population in ascending order, and set the population size to . N The population is divided into three layers in a ratio of 1:2:7, namely the elite layer, the local development layer, and the global exploration layer. Meanwhile, to reduce repeated calls to the fitness function and improve algorithm efficiency, the fitness function is only called in the first and second iterations of the local search phase. T In the 2 / 2 iteration, the fitness of the entire population is sorted, and the stratification rules remain unchanged at other times; Step S42: Update the position of the elite layer: First, calculate the weighted centroid of the elite layer. Using the inverse of individual fitness as the weight, it is ensured that individuals with better fitness contribute more to the centroid: (7) In the formula, F ( X i ) is the first i The fitness value of an elite individual. e 8 To avoid the minimum value where the denominator is 0; Update the position of elite individuals: (8) In the formula, S current To adapt to the amplitude of random perturbations, the amplitude decays linearly with the iteration process, balancing early exploration and later development: (9) In the formula, S init =( ub lb ) / 20 represents the initial disturbance amplitude. S end =( ub lb ) / 100 represents the perturbation amplitude at the end of the iteration. Step S43: Introduce a coordinate pattern search algorithm to update the position of the local development layer: For the local development layer i Each individual, based on their current location As the initial starting point for the search X current Define the core parameters for pattern search: optimize the problem dimensions. D The feasible solution space j The upper and lower bounds are lb j , ub j ;No. j The initial search step size is step init,j =( ub j lb j ) / 10, minimum convergence step size step min,j =10 6 Maintain a step-by-step state matrix for each individual as follows: step state ( i , j ), record the i The individual j The current search step size of the dimension; Pattern search starts from the initial starting point and explores in both positive and negative directions sequentially according to the dimension. j Two test points are generated using the following formula: (10) In the formula, step state ( i , j ) is the first i The individual j The current search step size of the dimension, e j For the first j A 3D unit coordinate vector; to ensure the feasibility of the test points, boundary limiting processing is applied to each test point: (11) Calculate the number in sequence j The fitness values ​​of two trial points (positive and negative) are used to select the trial point with the best fitness in that dimension. X best,j Its fitness value is F best,j ,satisfy: (12) In the formula F (·) represents the fitness function of the problem to be optimized; based on the fitness of the optimal exploration point, the search step size is adaptively adjusted and the search starting point is updated: if a better solution is found, the search starting point is updated and the step size is maintained to expand the search range; if no better solution is found, the step size is reduced by 0.5 times to shrink the search granularity, with the specific rules as follows: (13) In the formula F current The fitness value is the starting point of the current search; during the dimensional search, if the current step size of a certain dimension is less than the minimum step size of that dimension... step min,j If the search fails, skip that dimension; after traversing all dimensions, use the final optimal search starting point as the update position for this individual in this iteration. ; Step S44: Update the position of the global exploration layer: The core objective of the global exploration layer is to maintain population diversity and avoid the algorithm getting stuck in local optima. It uses purely uniform random sampling within the total solution space to update positions, without any convergence guidance. (14) Step S45: Apply boundary constraints to all individuals in the entire population, update the historical best position and global best position of each individual, and record the convergence curve data for this iteration. If the current iteration number is less than [number missing], [then the process continues]. T If step 2 is reached, proceed to step S41 to continue the iteration.

5. The method for optimizing landing trajectory angular rate controller parameters based on improved pigeon flock optimization according to claim 4, characterized in that: In step S41, the elite layer consists of individuals with the top 10% fitness, whose role is to guide the population convergence and provide a basic direction for local optimization; the local development layer consists of individuals with the top 10%-30% fitness, whose role is to complete the refined search; and the global exploration layer consists of individuals with the bottom 70% fitness, whose role is to randomly explore within the global solution space and maintain population diversity.

Images