Solar fixed-wing unmanned aerial vehicle pid parameter and filter configuration collaborative optimization method

Abstract

The application discloses a solar fixed-wing unmanned aerial vehicle PID parameter and filter configuration cooperative optimization method, and belongs to the technical field of unmanned aerial vehicle flight control, which comprises the following steps: establishing a nonlinear flight mechanics model of a solar fixed-wing unmanned aerial vehicle; determining the pull bias range of overall parameters; generating a probability coverage verification case set by adopting Latin hypercube sampling; positioning corner point combinations in a multi-dimensional parameter space through extreme value optimization based on corner point strong injection of a cost function; merging the probability coverage verification case set and the corner point combinations; cooperatively optimizing PID parameters and filter cutoff frequencies with the aim of minimizing the comprehensive cost function of the mixed verification case set; and performing robustness verification on the optimized PID parameters and filter cutoff frequencies. The method significantly improves the sufficiency and engineering reliability of verification, can obtain globally optimal control parameter configuration, and effectively reduces actual flight test risks.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) flight control technology, specifically relating to a collaborative optimization method for PID parameters and filter configuration of a solar-powered fixed-wing UAV. Background Technology

[0002] As a novel long-endurance UAV platform, solar-powered fixed-wing UAVs face significantly different challenges in flight control system design compared to traditional internal combustion engine or pure electric UAVs. First, the surface roughness, ply thickness, and encapsulation process of the solar panels covering the wing surface alter the wing's aerodynamic characteristics, including lift coefficient, drag coefficient, and zero-lift angle of attack. These changes directly affect the UAV's longitudinal and lateral aerodynamic response. Second, different payload configurations (such as different camera and sensor mounting schemes) lead to changes in the UAV's total mass and center of gravity, thus affecting its moment of inertia and trim characteristics. Furthermore, the elastic deformation of the wing under aerodynamic loads alters the effective arc and aerodynamic distribution, and the temperature effect of the solar panels introduces additional parameter uncertainties. The combined effect of these factors necessitates that the flight control law of solar-powered fixed-wing UAVs possess sufficient robustness to cope with the challenges posed by multi-source parameter uncertainties.

[0003] Currently, the tuning of PID control parameters for UAVs mainly relies on the following methods: The first is empirical tuning based on trial and error, such as the Ziegler-Nichols method and its improvements. This type of method relies heavily on the professional experience of engineers and is difficult to systematically handle multi-parameter coupled scenarios. The second is automatic tuning based on optimization algorithms, such as genetic algorithms and particle swarm optimization. This type of method usually only optimizes for nominal points and does not fully consider parameter deviation scenarios. The third is robustness verification based on Monte Carlo methods. However, existing methods usually use pure random sampling or Latin hypercube sampling (LHS) to generate verification cases. Existing robust control parameter design methods have the following main shortcomings: First, although Latin hypercube sampling (LHS) can probabilistically cover the parameter space uniformly, it is essentially a "probabilistic coverage" strategy, which is likely to miss corner points in the multi-dimensional parameter space (i.e., combinations of multiple extreme errors), resulting in significant corner point omission problems. In real flight tests, it is precisely this combination of multiple extreme errors that leads to crashes. For example, when a UAV simultaneously exhibits a combination of upper limits for mass deviation, rearward shift of the center of gravity, and decrease in lift coefficient, its longitudinal static stability deteriorates drastically, and pure LHS sampling has a very low probability of accurately hitting this corner combination. Secondly, existing methods typically separate PID parameter tuning from filter design, failing to consider the synergistic optimization of the two. Thirdly, existing methods lack a systematic closed-loop mechanism from simulation verification to hardware-in-the-loop verification, meaning that parameters that pass verification in simulation may still exhibit problems in actual flight. Summary of the Invention

[0004] The purpose of this invention is to provide a collaborative optimization method for PID parameters and filter configuration of solar-powered fixed-wing UAVs, in order to solve the problems of existing UAV PID parameter tuning methods, such as corner point omissions in robustness verification methods, separation of PID parameter tuning and filter design, and lack of a systematic closed-loop mechanism.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention relates to a method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing unmanned aerial vehicles (UAVs), comprising the following steps: S1. Establish a nonlinear flight dynamics model for a solar-powered fixed-wing UAV; S2. Determine the range of deviation for the overall parameters; S3. Within the skew range, use Latin hypercube sampling to generate a probability coverage verification test case set; S4. Corner point strong injection based on cost function, and corner point combination in multidimensional parameter space is located by extreme value optimization, wherein the corner point combination is used to represent the most dangerous parameter combination; S5. Merge the probability coverage verification test case set with the corner point combination to form a hybrid verification test case set; S6. With the goal of minimizing the comprehensive cost function under the mixed verification test case set, the PID parameters and filter cutoff frequency are optimized in a coordinated manner. S7. Perform robustness verification on the optimized PID parameters and filter cutoff frequency. If the verification fails, return to S6 to continue optimization. If the verification passes, output the final PID parameters and filter cutoff frequency.

[0006] Preferably, in S1, establishing a nonlinear flight dynamics model for a solar-powered fixed-wing UAV means: based on the geometric parameters, mass characteristics, and aerodynamic data of the solar-powered fixed-wing UAV, establishing a six-degree-of-freedom nonlinear flight dynamics model, which is expressed as: , in, State variables x Regarding time t Find the first derivative. t Indicates time, x State variables, state variables x Includes body axis velocity ( u b , v b , w b ), angular velocity ( p , q ,r ), attitude angle ( , , ) and the velocity of the Earth's axis ( V x , V y , V z ); u b , v b , w b These represent the forward, lateral, and vertical velocities along the body axis, respectively. p , q , r These represent the roll rate, pitch rate, and yaw rate in the body axis system, respectively. , , These represent the roll angle, pitch angle, and yaw angle, respectively. V x , V y , V z These represent the velocities along the x-axis, y-axis, and z-axis, respectively. u Indicates the control variable, control variable u Including throttle opening δt Elevator deflection δe Aileron deflection δ a and rudder deflection δr ; q flex For the generalized coordinate vector of the wing's elastic deformation, T panel For the temperature of the solar panel, W wind This is the atmospheric wind speed vector.

[0007] Preferably, in S2, the range of overall parameters includes mass, center of gravity position, moment of inertia, lift coefficient slope, zero lift drag coefficient, pitch moment coefficient, cruise speed, and solar panel temperature.

[0008] Preferably, the specific steps of corner injection based on the cost function in S4, which locates the most dangerous parameter combination in the multidimensional parameter space through extreme value optimization, are as follows: S4.1. Construct the vertical cost function, which is defined as follows: Cost_lon=W1×R_height+W2×R_pitch, Where Cost_lon represents the longitudinal cost, R_height is the height drop risk indicator, R_pitch is the pitch divergence risk indicator, and W1 and W2 are both weighting coefficients. S4.2. Construct the transverse cost function, which is defined as follows: Cost_lat=W3×R_roll+W4×R_yaw+W5×R_dutch, Where Cost_lat represents the lateral cost, R_roll is the roll divergence risk index, R_yaw is the yaw divergence risk index, R_dutch is the Dutch roll coupling risk index, and W3, W4 and W5 are all weighting coefficients. S4.3. Use sequential quadratic programming or genetic algorithm to maximize the longitudinal cost function and the lateral cost function, locate the corner point combination in the multidimensional parameter space, and call the nonlinear flight mechanics model to calculate the trim state and linearization model in each iteration, and extract risk indicators.

[0009] Preferably, in step S5, when merging the probability coverage verification test case set and the corner point combination, the number of corner point combinations is 3 to 5% of the total number of verification test cases in the mixed verification test case set.

[0010] Preferably, in step S6, the comprehensive cost function is constructed as follows: First, we construct a mathematical model of a PID controller with a low-pass filter, expressed using a transfer function: , in, G c ( s () is the filtered PID control law. s For the Laplace domain, G LPF ( s Let be the transfer function of a first-order low-pass filter, given by the low-pass filter cutoff frequency. fc The decision is made using the following formula: ; Secondly, the proportional gain of each control channel Kp Integral gain Ki Differential gain Kd and the corresponding low-pass filter cutoff frequency. fc For the variables to be optimized together, construct the objective function for the optimization together, expressed as: , in, J Indicates the overall cost. X Represents the vector of variables for collaborative optimization. X =[ Kp , Ki , Kd , fc ], Ω represents a mixed set of verification test cases. and They were respectively in the second i Under the condition of parameter bias in a single verification test case, the current variable is used. X Constructed Filtered PID Control Law G c ( s The longitudinal and lateral costs calculated after the loop is closed.

[0011] Preferably, in step S6, the specific steps for collaboratively optimizing the PID parameters and filter cutoff frequency with the objective of minimizing the comprehensive cost function under the mixed verification test case set are as follows: S6.1. Randomly generate an initial population, where each individual in the initial population contains a set of variables. X ; S6.2. For each individual variable in the population X Calculate the filtered PID control law Gc ( s This process involves constructing a PID controller for the current iteration; iterating through each validation case in the mixed validation case set, calling the nonlinear flight mechanics model for balancing and linearization, and obtaining the open-loop transfer function. L ( s The system calculates the phase margin and gain margin based on the open-loop transfer function, and determines whether the phase margin and gain margin constraints are met. It also determines whether the balancing constraints and the low-pass filter cutoff frequency are satisfied. fc If the frequency band is within the specified range, calculate the longitudinal cost Cost_lon_i and lateral cost Cost_lat_i for the corresponding verification case, extract risk indicators, and then calculate the comprehensive cost function value for the verification case. J Otherwise, impose a very high penalty on the individual. S6.3. Generate a new generation of variables based on the velocity and position update formulas of the MOPSO algorithm or the gradient and Hessian matrix update formulas of the SLSQP algorithm. Kp , Ki , Kd , fc Population; S6.4. Determine whether the maximum number of iterations has been reached or the objective function has converged. If yes, output the Pareto optimal solution set or the global optimal solution to obtain the final PID parameters and filter cutoff frequency; otherwise, return to S6.2 to continue iterating.

[0012] Preferably, in S2, when determining the skew range of the overall parameters, a correlation matrix is ​​used to describe the coupling relationship between the parameters; in S3, when generating the probability coverage verification test case set through Latin hypercube sampling, the correlation of the parameters is maintained through correlation sampling technology; in S4, coupling constraints are used as one of the optimization constraints.

[0013] Preferably, in S7, large-scale Monte Carlo random sampling is used to robustly verify the optimized PID parameters and filter cutoff frequency, including verifying the following indicators: balancing success rate, minimum phase margin, minimum gain margin, and maximum cost function of all verification cases.

[0014] Preferably, after outputting the final PID parameters and filter cutoff frequency in step S7, the method further includes: step S8 deploying the PID parameters and filter cutoff frequency into the hardware and verifying them through a loop simulation system; if a problem is found during verification, the process returns to step S6.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The collaborative optimization method for PID parameters and filter configuration of solar-powered fixed-wing UAVs involved in this invention first determines the bias range of the overall parameters when collecting verification test cases. Then, within the bias range, a probability coverage verification test case set is generated using Latin hypercube sampling. Next, corner point strong injection is performed based on the cost function, and corner point combinations in the multidimensional parameter space are actively located through extreme value optimization. Finally, the probability coverage verification test case set and the corner point combinations are merged to form a hybrid verification test case set. Furthermore, a hybrid verification strategy of "LHS probability coverage + physical boundary corner point strong injection" is proposed, which completely solves the problem that pure LHS sampling cannot cover the corner points of the multidimensional parameter space, and significantly improves the sufficiency of verification and engineering reliability.

[0016] 2. The collaborative optimization method for PID parameters and filter configuration of solar-powered fixed-wing UAVs involved in this invention constructs longitudinal and lateral cost functions during the process of locating the most dangerous parameter combinations in a multidimensional parameter space through extreme value optimization. It then performs a maximization search on the longitudinal and lateral cost functions to locate corner combinations in the multidimensional parameter space. By defining corner analysis methods for longitudinal and lateral cost functions respectively, flight safety risks are quantified into optimizable mathematical indicators, making the corner search process have clear physical meaning and engineering interpretability.

[0017] 3. The collaborative optimization method for PID parameters and filter configuration of solar-powered fixed-wing UAVs involved in this invention aims to minimize the comprehensive cost function under a mixed verification test case set. It collaboratively optimizes PID parameters and filter cutoff frequencies, avoiding the suboptimal solution problem of the traditional two-step method, and can obtain the globally optimal control parameter configuration.

[0018] 4. The collaborative optimization method for PID parameters and filter configuration of solar-powered fixed-wing UAVs involved in this invention also deploys the final PID parameters and filter cutoff frequency into the hardware, verifies them through a loop simulation system, forms a complete closed-loop process, ensures the engineering reliability of the parameters, and effectively reduces the risk of actual flight testing. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 A schematic diagram of the overall parameter bias model for a solar-powered fixed-wing UAV; Figure 3 This is a schematic diagram of a hybrid strategy combining LHS probability coverage and corner strong betting. Figure 4 This is a schematic diagram of the process of finding the extreme value of the longitudinal cost function; Figure 5 This is a schematic diagram of the process of finding the extreme value of the transverse cost function; Figure 6 This is a schematic diagram illustrating the collaborative optimization process of PID parameters and filters. Figure 7 This is a flowchart of the verification process in a hardware-in-the-loop simulation system. Detailed Implementation

[0020] To further understand the content of this invention, the invention will be described in detail with reference to the embodiments. The following embodiments are used to illustrate the invention, but are not intended to limit the scope of the invention.

[0021] This embodiment uses a typical solar-powered fixed-wing UAV as an example, whose basic overall parameters are shown in Table 1. The UAV adopts a conventional fixed-wing layout, with large solar panels mounted on the wings. It uses a pure electric + solar hybrid power system, achieving an endurance of over 4 hours. The flight control system employs a classic PID control architecture, including longitudinal channels (pitch, throttle) and lateral channels (roll, yaw).

[0022] Table 1: Basic Overall Parameters of Solar-Powered Fixed-Wing UAVs

[0023] See attached document Figure 1 As shown, this invention relates to a method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing unmanned aerial vehicles (UAVs), comprising the following steps: S1. Establish a nonlinear flight dynamics model for a solar-powered fixed-wing UAV. Specifically, based on the geometric parameters, mass characteristics, and aerodynamic data of the solar-powered fixed-wing UAV, establish a six-degree-of-freedom nonlinear flight dynamics model. The nonlinear flight dynamics model is expressed as follows: , in, State variables x Regarding time t Find the first derivative. t Indicates time, x State variables, state variables x Includes body axis velocity ( u b , v b , w b ), angular velocity ( p , q , r ), attitude angle ( , , and the velocity of the Earth's axis ( V x , V y , V z ); u b , v b , w b These represent the forward, lateral, and vertical velocities along the body axis, respectively. p , q , r These represent the roll rate, pitch rate, and yaw rate in the body axis system, respectively. , , These represent the roll angle, pitch angle, and yaw angle, respectively. V x , V y , V z These represent the velocities along the x-axis, y-axis, and z-axis, respectively. u Indicates the control variable, control variable u Including throttle opening δt Elevator deflection δe Aileron deflection δ a and rudder deflection δr ; q flex For the generalized coordinate vector of the wing's elastic deformation, T panel For the temperature of the solar panel, W wind This is the atmospheric wind speed vector.

[0024] In addition to establishing the aforementioned nonlinear flight mechanics model, the following models should also be established: (1) To reflect the static elastic effect and temperature-dependent aerodynamic characteristics, the aerodynamic coefficient (in terms of lift coefficient) is... CL and pitching moment coefficient Cm (For example) Expanded as: , in, θ The elastic torsion angle of the wing (static elastic deformation); C Lθ , C mθ This is the coupling derivative of elastic deformation with respect to the aerodynamic coefficient. C L0 ( T panel )and C Lα ( T panel The lift aerodynamic parameters, which are affected by the temperature of the solar panel, can all be obtained through fluid dynamics simulation or interpolation of measured data. C m0 ( T panel )and C mα ( T panel ( ) represents the pitch moment aerodynamic parameter affected by the temperature of the solar panel. α For the angle of attack, and These are the lift and pitch moment coefficients related to the elevator, respectively. This refers to the elevator angle.

[0025] (2) Elastic deformation of the wing under aerodynamic load (based on the elastic torsion angle of the wing) θ (For example) Calculated using the static equilibrium equation: , in, K θ For the wing torsional stiffness, ρ air density, V Airspeed, S For wing area, c For the average aerodynamic chord length, C mα The slope represents the pitching moment coefficient. This formula illustrates the coupling mechanism by which aerodynamic loads cause wing twisting, which in turn alters the effective angle of attack and feeds back to aerodynamic forces / moments.

[0026] (3) The atmospheric model adopts the standard atmospheric model function atmosisa() in MATLAB software. The airspeed of the aircraft is affected by the wind speed Wwind=[uw,vw,ww] TThe airflow velocity under the body axis due to disturbance is expressed as: , in, u air , v air , w air Represent the three axial velocities of the wind axis system, R bg This is the transformation matrix from the Earth axis system to the body axis system. u w , v w , w w Gust and turbulence components generated for atmospheric models.

[0027] The model inputs are the deflection angles of the control surfaces (elevator, aileron, rudder) and the throttle opening, and the outputs are the UAV's state response, including angle of attack, angle of flight, roll angle, and kinematic velocity. The model should include the aeroelastic effects of wing elastic deformation, the temperature-dependent aerodynamic characteristics of the solar panels, and the influence of the atmospheric model.

[0028] S2. See Appendix Figure 2As shown, the range of overall parameter deviations is determined. This deviation range comprehensively considers factors such as mass deviations caused by manufacturing tolerances, mass changes caused by different load configurations, aerodynamic parameter changes caused by the elastic deformation of the solar panel, center of gravity position shifts, and changes in moment of inertia. Based on the above analysis, the deviation ranges include: mass deviation range, center of gravity position deviation range, moment of inertia deviation range, lift coefficient slope deviation range, zero-lift drag coefficient deviation range, pitching moment coefficient deviation range, cruise speed deviation range, and solar panel temperature deviation range. According to common conservative design principles in engineering practice, manufacturing tolerances, load configuration differences, solar panel process variations, and environmental changes are also considered. The following factors influence the deviation range of most solar-powered fixed-wing UAVs: the mass deviation range is ±10% of the nominal design mass; the center of gravity deviation range is ±5% of the average aerodynamic chord length of the standard design center of gravity; the moment of inertia deviation range is ±15% of the standard design moment of inertia; the lift coefficient slope deviation range is ±8% of the standard design value; the zero-lift drag coefficient deviation range is ±10% of the standard design drag coefficient; the pitch moment coefficient slope deviation range is ±10% of the standard design value; the cruise speed deviation range is ±5 m / s of the nominal cruise speed; and the solar panel temperature deviation range is -10 to +40℃, as shown in Table 2. The determination of these overall parameter deviation ranges comprehensively considers factors such as mass deviations caused by manufacturing tolerances, mass variations caused by different load configurations, aerodynamic parameter variations caused by elastic deformation of the solar panel, center of gravity position shifts, and changes in moment of inertia. Simultaneously, a correlation matrix can be used to describe the coupling relationship between parameters, which is established by constructing a Pearson correlation coefficient matrix R. The specific method is as follows: Based on the historical flight data of the UAV, detailed mass characteristic CAD models, or aeroelastic calculation data, extract sample sets of each overall parameter; calculate any two parameters. X i and X j Linear correlation coefficient between ρ ij This forms a correlation coefficient matrix R, where | ρ ij |≤1. Each element in matrix R ρ ij Representing the i The parameter and the first j The Pearson correlation coefficient between the two parameters is calculated by dividing the covariance of the two parameters by the product of their standard deviations. , in, and Two parameters respectively Xi and X j Standard deviation; For example, increasing the mass m of a drone usually leads to an increase in the moment of inertia Ixx, and the two are strongly positively correlated (ρ≈0.8~0.9); while increasing the rearward shift of the center of gravity xcg may lead to a decrease in the absolute value of the pitch moment coefficient slope Cmα, which is negatively correlated.

[0029] Table 2: Overall Parameter Variation Range

[0030] S3. See Appendix Figure 3 As shown, within the skew range, Latin Hypercube Sampling (LHS) is used to generate N sets (typically N=200) of parameter combinations as the probabilistic coverage validation test case set. The basic principle of the LHS method is to divide the skew range of each parameter into N equal levels, then randomly sample a value in each level, and finally randomly combine the sampled values ​​of each parameter. This method can ensure that each level interval of each parameter is sampled, thus uniformly covering the entire parameter space in a probabilistic sense. However, it should be noted that LHS is essentially a "probabilistic uniformity" sampling strategy, which cannot guarantee hitting the corner regions of the multidimensional parameter space. For example, in a 10-dimensional parameter space, even if 200 samples are sampled, the probability of hitting a specific corner is still extremely low, and these corners may be the most dangerous parameter combinations. Therefore, it must be used in conjunction with the subsequent corner injection step.

[0031] If a correlation matrix is ​​used to describe the coupling relationship between parameters in step S2, then this step also requires maintaining the correlation of the parameters through correlation sampling techniques. The correlation sampling technique adopted is the Iman-Conover method (rank-based correlation sampling method), and the specific implementation steps are as follows: Standard Latin hypercube sampling (LHS) is used to generate an independent edge distribution sampling matrix Xraw for each parameter; Based on the target correlation coefficient matrix R established in S2, perform Cholesky decomposition to obtain a lower triangular matrix L, such that R = LL. T ; Calculate the rank correlation matrix Rraw of the independent sampling matrix Xraw, and perform Cholesky decomposition on it to obtain the lower triangular matrix M, such that Rraw = MM. T ; Construct the transformation matrix T=M -1 L, apply the transformation matrix T to the sorting matrix of the original samples, rearrange the positions of the elements in each column of Xraw, and obtain the rearranged sampling matrix Xcorr; The rearranged matrix Xcorr maintains the original marginal distribution of each parameter (i.e., uniform coverage of LHS), while the rank correlation between each parameter within it approximates the target correlation coefficient matrix R, thus achieving "preserving the correlation of parameters".

[0032] S4. Since pure LHS sampling cannot cover the corner points of the multidimensional parameter space, this invention further utilizes corner injection based on cost functions to locate corner point combinations in the multidimensional parameter space through extreme value optimization. These corner point combinations represent the most dangerous parameter combinations. The basic idea is: within the parameter bounding box that satisfies the constraints, by constructing cost functions in the longitudinal and lateral directions, a mathematical optimization algorithm is used to actively search for parameter combinations that maximize the cost functions, i.e., "theoretical corner points." These corner points represent the most dangerous scenarios with multiple extreme errors superimposed, which are almost impossible to hit by pure probability sampling. (See attached...) Figure 4 and 5 As shown, the specific steps are as follows: S4.1. Construct the vertical cost function, which is defined as follows: Cost_lon=W1×R_height+W2×R_pitch, Where Cost_lon represents the longitudinal cost; R_height is the altitude loss risk index, defined as the altitude loss rate of a UAV in a level flight trim state under a given parameter combination after being subjected to a unit gust disturbance. This index reflects the UAV's altitude holding capability under longitudinal disturbances. The calculation method is as follows: under a given parameter combination, firstly, the level flight trim state (lift equals gravity, pitch moment equals zero, yaw force equals zero) is calculated using a flight mechanics model. Then, a unit vertical gust disturbance (typically 2 m / s) is applied in the trim state, and a nonlinear time-domain simulation is performed to extract the UAV's altitude response curve and calculate the altitude loss rate as R_height.

[0033] R_pitch is a pitch divergence risk index, defined as the reciprocal of the damping ratio of the short-period longitudinal modes of the UAV under a given parameter combination. This index reflects the stability boundary of the UAV's longitudinal dynamic response. The calculation method for R_pitch is as follows: linearize the flight mechanics model under trim conditions with small perturbations, extract the eigenvalues ​​of the short-period longitudinal modes, and calculate the reciprocal of the damping ratio as R_pitch. When the damping ratio approaches 0, R_pitch tends to infinity, indicating that the longitudinal dynamics are about to become unstable. W1 and W2 are weighting coefficients, with typical values ​​of W1=0.6 and W2=0.4, which can be adjusted according to the safety priority of the actual flight mission.

[0034] S4.2. Construct the transverse cost function, which is defined as follows: Cost_lat=W3×R_roll+W4×R_yaw+W5×R_dutch, Wherein, Cost_lat represents the lateral cost; R_roll is the roll divergence risk index, defined as the reciprocal of the damping ratio of the lateral roll mode, reflecting the stability of the UAV's lateral roll steering; R_yaw is the yaw divergence risk index, defined as the reciprocal of the damping ratio of the directional mode, reflecting the directional stability of the UAV; R_dutch is the Dutch roll coupling risk index, defined as the reciprocal of the damping ratio of the Dutch roll mode, reflecting the stability of the UAV's three-axis coupled dynamics; W3, W4, and W5 are all weighting coefficients, with typical values ​​of W3=0.4, W4=0.3, and W5=0.3.

[0035] The calculation of the lateral cost function also requires calling the flight mechanics model in each optimization iteration to calculate the trim state and extract the lateral modal eigenvalues.

[0036] S4.3. The sequential quadratic programming (SLSQP) or genetic algorithm (GA) is used to maximize the longitudinal and transverse cost functions, find the parameter combinations that maximize Cost_lon and Cost_lat, and then locate the corner points in the multidimensional parameter space. The SLSQP algorithm is suitable for handling continuous variable optimization problems with boundary constraints and can efficiently find local optima. The GA algorithm is used to verify the globality of the SLSQP results and avoid getting trapped in local optima.

[0037] Taking the Genetic Algorithm (GA) as an example, its search process from data input to corner combination output is as follows: S4.3.1. Encoding and Initialization: The range of deviation of the overall parameters (such as mass, center of gravity, moment of inertia, aerodynamic coefficient, etc.) determined in S2 is used as the boundary constraint, and the coupling constraint is used as one of the optimization constraints. N individuals are randomly generated within the upper and lower limits of each parameter to form the initial population, and each individual represents a combination vector of one overall parameter; S4.3.2. Fitness Assessment: Substitute each individual in the population (i.e., the parameter combination) into the nonlinear flight dynamics model. In the process, the trim state is calculated and linearized with small perturbations, and the corresponding risk indicators (R_height, R_pitch or R_roll, R_yaw, R_dutch) are extracted. Then, the longitudinal cost Cost_lon or the lateral cost Cost_lat is calculated, and its value is used as the fitness of the individual. S4.3.3. Selection Operation: Based on the fitness of individuals, the fitness ratio selection method or tournament selection method is used to prioritize the retention of individuals with larger cost function values ​​(i.e., higher risk) to the next generation; S4.3.4. Crossover and Mutation: Perform arithmetic crossover and polynomial mutation on the selected individuals to generate offspring individuals, and perform boundary checks on the offspring individuals to ensure that the generated parameters are still within the bias range and coupling constraints specified in S2; S4.3.5. Iteration and Output: Repeat steps S4.3.2 to S4.3.4 until the maximum number of iterations is reached or the fitness converges. Output the individual parameter combinations that maximize Cost_lon and Cost_lat, thus locating the most dangerous corner combinations in the longitudinal and lateral directions in the multi-dimensional parameter space. These parameter combinations are the "theoretical corners," representing the worst physical combination of aerodynamic and mass characteristics of the UAV in the parameter bias space, serving as the extreme verification scenario for the subsequent S6 co-optimization of PID parameters and filter cutoff frequency. Each iteration calls the nonlinear flight mechanics model to calculate the trim state and linearized model, and extracts risk indicators.

[0038] The calculation process for the balancing state is as follows: Initial values ​​for trim calculation: The initial values ​​of state variables and control variables are zero by default. The initial value of velocity in the state variables can be given according to the flight mission. Trial calculation constraints: The constraint formula for the aircraft to enter a steady-state level flight state is expressed as follows: , in, Represents the forward velocity on the body axis as a function of time. t Find the first derivative. Represents the lateral velocity on the body axis as a function of time. t Find the first derivative. Represents the vertical velocity on the body axis as a function of time. t Find the first derivative. Represents the roll angular velocity in the body axis system as a function of time. t Find the first derivative. Represents the pitch angular velocity relative to time in the body axis system. t Find the first derivative. Indicates the yaw rate in the body axis system as a function of time. t Find the first derivative; By using the initial values ​​of state and control variables, steady-state constraints, and nonlinear flight dynamics model equations, a sequential quadratic programming algorithm is invoked to obtain the state and control variables of the aircraft at the desired altitude and speed, which are then used as the trim result.

[0039] The linearization model calculation process is as follows: Suppose the equation of motion for a certain total nonlinear motion is: , variables in the formula It is a motion state variable or its derivative, and can be expressed as a reference motion state variable. and deviation of small disturbances The sum of, therefore: ; For any type of movement, the following two equations are always satisfied: , , because Since it is a small perturbation (also known as an increment), the above equation can be expanded using Taylor, and then its second and higher derivatives can be ignored, resulting in the following formula: , Since the first term in the above equation is zero, we can obtain the following linearized small perturbation equation: , coefficients in the formula , , These are all partial derivatives of a certain reference motion point, and are all known constants.

[0040] Using the above method, nonlinear flight dynamics models are analyzed. By performing linearization, a small-perturbation linearized model of the aircraft can be established. The small-perturbation linearized model is expressed as follows: , Where the state vector The output variable represents the change in the state variable x. , Manipulation vector for changes in the velocity of the Earth's axis system , A The state matrix, B For the control matrix, C For the observation matrix, D It is a feedforward matrix.

[0041] Based on the state space matrix obtained after linearization A , B , C , D The eigenvalues ​​of the longitudinal short-period mode, roll mode, heading mode, and Dutch roll mode can be extracted, and the damping ratio can be calculated. This extraction method is a commonly used eigenvalue analysis method in the aerospace industry, which can be performed using functions in MATLAB software. It is not the subject of this invention and will not be elaborated here.

[0042] The risk indicators include the following: altitude loss risk indicator R_height, pitch divergence risk indicator R_pitch, roll divergence risk indicator R_roll, yaw divergence risk indicator R_yaw, and Dutch roll coupling risk indicator R_dutch. The altitude loss risk index R_height is defined as the altitude loss rate of a UAV in a level flight trim state under a given parameter combination after being subjected to a unit gust of wind disturbance. This index reflects the UAV's altitude maintenance capability under longitudinal disturbances. The calculation method is as follows: Under a given parameter combination, firstly, the level flight trim state (lift equals gravity, pitch moment equals zero, yaw force equals zero) is calculated using a flight mechanics model. Then, a unit vertical gust of wind disturbance (typically 2 m / s) is applied under the trim state, and a nonlinear time-domain simulation is performed. The altitude response curve of the UAV is extracted, and the altitude loss rate is calculated as R_height.

[0043] The pitch divergence risk index R_pitch is defined as the reciprocal of the damping ratio of the short-period longitudinal modes of the UAV under a given combination of parameters. This index reflects the stability boundary of the UAV's longitudinal dynamic response. The calculation method is as follows: under trim conditions, the flight mechanics model is linearized with small perturbations, the eigenvalues ​​of the short-period longitudinal modes are extracted, and the reciprocal of the damping ratio is calculated.

[0044] The roll divergence risk index R_roll reflects the stability of the UAV's roll turn. It is calculated by linearizing the flight dynamics model under trim conditions with small disturbances and then calculating the reciprocal of the damping ratio of the roll mode.

[0045] The yaw divergence risk index R_yaw reflects the heading stability of the UAV. It is calculated by linearizing the flight dynamics model under trim conditions with small perturbations and then calculating the reciprocal of the damping ratio of the heading mode.

[0046] The Dutch roll coupling risk index R_dutch reflects the stability of the three-axis coupled dynamics of an unmanned aerial vehicle (UAV). It is calculated by linearizing the flight dynamics model under trim conditions with small perturbations, yielding the reciprocal of the damping ratio of the Dutch roll mode.

[0047] The calculation of the above risk indicators is a common practice in this field.

[0048] This step identifies 3–5 longitudinal corner points and 3–5 lateral corner points, for a total of 6–10 corner point combinations. These parameter combinations are known as "theoretical corner points," representing the worst physical combination of aerodynamic and mass characteristics of the UAV within the parameter bias space. These serve as the extreme verification scenario for the subsequent collaborative optimization of PID parameters and filter cutoff frequency in S6.

[0049] When locating corner point combinations for extreme value optimization in S4, and using the correlation matrix to describe the coupling relationship between parameters, the coupling constraint is further incorporated as one of the optimization constraints. Its mathematical expression is: , in, x i , x j For any two population parameters that have a strong coupling relationship (i.e., | ρ ij | Parameters with a value greater than 0.5); For the first k The physical constraint function of the group coupling parameters in the current optimization iteration; This represents the baseline coupling relationship value fitted based on engineering data; ε k This represents the allowable coupling deviation tolerance. In a typical embodiment, for the strong positive coupling constraint between mass and moment of inertia, the specific expression can be written as: , In the formula, m 0、 I xx0 Δ is the nominal design value. I xx For tolerance bandwidth, I xx min , I x max These represent the upper and lower limits of the moment of inertia Ixx, respectively. m This represents mass. This constraint ensures that when the optimization algorithm searches for the most dangerous combination of parameters (such as maximum mass), the moment of inertia must also be within a physically reasonable range of increase, and will not search for spurious corner points that are physically impossible to achieve, such as "large mass + extremely small moment of inertia".

[0050] S5. Merge the probability coverage verification test case set generated in S3 with the corner combination obtained in S4 to form a hybrid verification test case set of "LHS probability coverage + corner strong injection", wherein the number of corner combinations is 3 to 5% of the total number of verification test cases in the hybrid verification test case set.

[0051] The LHS probability coverage validation test case set (200 sets) and the corner injection results (6-10 sets) are merged to form a hybrid validation test case set. The core advantage of this hybrid strategy is that LHS provides probabilistic uniform coverage of the parameter space, ensuring the breadth of validation; while corner injection ensures that the most dangerous extreme combinations are fully tested, ensuring the completeness of validation. The combination of the two avoids both the corner omission problem of pure random sampling and the insufficient coverage problem of pure corner analysis, achieving the optimal balance between validation sufficiency and validation efficiency.

[0052] S6. With the goal of minimizing the comprehensive cost function under the mixed verification test case set, the PID parameters and filter cutoff frequency are optimized in a coordinated manner using the multi-objective particle swarm optimization algorithm (MOPSO) or the sequential quadratic programming method (SLSQP).

[0053] First, a mathematical model of a PID controller with a low-pass filter is constructed. To suppress high-frequency noise introduced by the derivative term, the low-pass filter is connected in series with the derivative channel of the PID controller to form a filtered PID control law. In the Laplace domain (s-domain), the controller's transfer function G... c (s) is represented as: , in, G LPF ( s Let be the transfer function of a first-order low-pass filter, given by the cutoff frequency. fc The decision is made using the following formula: , Optimization variables include the proportional gain of each control channel. Kp Integral gain Ki Differential gain Kd and the corresponding low-pass filter cutoff frequency. fc The optimization objective function is: , in, J Indicates the overall cost. X Denotes the vector of variables for collaborative optimization, X=[ Kp , Ki , Kd , fc ], Ω represents the hybrid verification test case set generated in S5. and They were respectively in the second i Under the condition of parameter bias in a verification test case, the control law is constructed using the current variable X. G c ( s The longitudinal and lateral costs calculated after loop closure. Combined with... Kp , Ki , Kd and fc The calculation formula and optimization process are as follows: (1) If the multi-objective particle swarm optimization algorithm (MOPSO) is used for iterative optimization, the proportional gain will be... Kp Integral gain Ki Differential gain Kd and the low-pass filter cutoff frequency fc This is considered as a unified optimization variable vector. In each iteration, the update of this vector involves three parts: first, preserving the update trend of the previous generation of parameter variables, i.e., the inertial part; second, being influenced by the particle's own historical optimal parameter combination (i.e., the combination with the lowest cost during the historical optimization process). Kp , Ki , Kd and fc Attracted by the combination of parameters, the particle swarm moves towards its individual extremum; finally, attracted by the historical optimal parameter combination of the entire particle swarm, it moves towards the global extremum. The effects of these three parts are added together to calculate the velocity update for this iteration. Then, the velocity update for the current generation is calculated. Kp , Ki , Kd and fc The numerical values ​​are then added to their corresponding velocity update values ​​to generate a new generation of parameter combinations. After each update, boundary checks are performed on the parameters to ensure the newly obtained parameters are accurate. Kp , Ki , Kd and fc Within the physically feasible range and reasonable frequency band, if it exceeds the set boundary, amplitude limiting and truncation will be performed.

[0054] (2) If Sequential Quadratic Programming (SLSQP) is used for gradient optimization, the parameter update is based on the gradient information at the current variable point and the approximate Hessian matrix. In the current generation Kp , Ki , Kd and fc At the parameter combination point, the gradient of the comprehensive cost function is calculated. Under the constraints of balancing success rate, phase margin, and gain margin, a local quadratic programming subproblem is solved to determine the current optimal parameter search direction. After determining the search direction, a line search is performed along this direction to determine a reasonable iteration step size that effectively reduces the cost function. Finally, the current... Kp , Ki , Kd and fc The value is moved by that step length along the search direction, thus obtaining a new generation of parameter variable combinations.

[0055] Both of the above optimization processes are conventional methods in this field. Through the aforementioned iterative update mechanism, the algorithm continuously adjusts and optimizes in multidimensional space. Kp , Ki , Kd and fc The value of is taken until the set maximum number of iterations is reached or the overall cost objective function converges, and finally the optimized PID parameters and filter cutoff frequency are output.

[0056] See attached document Figure 6 As shown, taking the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm as an example, the specific optimization process and calculation formulas are as follows: S6.1. Particle Encoding and Population Initialization: The optimization variables of each control channel are combined and encoded into a single particle, i.e., the position vector of each particle is represented as... Under the constraints of satisfying the physical feasibility and reasonable range, an initial particle swarm is randomly generated, and the velocity vectors of the particles are initialized; S6.2. Fitness Assessment and Constraint Verification: For each particle in the population (i.e., a set of PID and filter parameter configurations), it is substituted into the closed-loop control system. All verification cases in the mixed verification case set are traversed, and the open-loop transfer function L(s) of all verification cases is calculated. Based on the open-loop transfer function, the phase margin and gain margin are calculated. Then, the constraints are verified: the trim state, phase margin, and gain margin under each verification case are calculated. If the constraints of "each verification case can be trimmed, phase margin ≥ 45°, and gain margin ≥ 6dB" are not met, a very large penalty value is imposed on the particle. For particles that meet the constraints, the longitudinal cost Cost_lon_i and the lateral cost Cost_lat_i of each verification case are calculated, and then the comprehensive cost function value of the particle is calculated. J As its fitness; S6.3. Based on the dominance relationships between particles, perform non-dominated sorting, store the non-dominated solutions in the current population in an external archive (Pareto solution set), and maintain the diversity of the external archive using an adaptive grid method or crowding distance; update velocity and position: select a global extremum from the external archive to guide the particle, combine it with the individual extremum, adjust the particle flight velocity according to the MOPSO velocity update formula, and then update the particle position (i.e., update...). Kp , Ki , Kd and fc The value), ensuring that the updated position remains within the set boundary range; S6.4. Determine whether the maximum number of iterations has been reached or the objective function has converged. If yes, output the Pareto optimal solution set or the global optimal solution to obtain the final PID parameters and filter cutoff frequency; otherwise, return to S6.2 to continue iterating.

[0057] By fc and Kp , Ki , Kd Synchronization is used as an optimization variable. In each iteration of calculating phase margin and gain margin, the phase lag and amplitude attenuation caused by the low-pass filter are taken into account in real time in the open-loop frequency response. This collaborative optimization mechanism avoids the problems of excessive compression of system bandwidth or artificially high stability margin caused by the traditional "tuning PID parameters first and then designing the filter", and ensures the globally optimal configuration of control gain and noise suppression capability.

[0058] The optimization constraints involved in S6.2 include: each verification test case can be balanced, phase margin ≥ 45°, gain margin ≥ 6dB, PID parameters within the physically feasible range, and filter cutoff frequency within a reasonable range.

[0059] The balancing constraint is expressed as: , Phase margin PM ≥ 45°, gain margin GM ≥ 6dB; Let the open-loop transfer function of the closed-loop system be... L ( s )= Gc ( s ) P ( s ), P ( s ) represents the transfer function of the controlled object, derived from a small-perturbation linearized model (state-space equations). It is derived from this and can be represented as: , This is then used to calculate phase margin and gain margin: in, C For the observation matrix, I It is the identity matrix. A The state matrix, B For the control matrix, D It is a feedforward matrix; , in, ω c For amplitude crossover frequency, ω g For phase crossing frequency, j It is the imaginary unit.

[0060] PID parameters are within a physically feasible range: the lower limit is usually greater than 0, while the upper limit is limited by the saturation deflection limit of the actuator and the maximum output capability of the physical device.

[0061] The filter cutoff frequency should be within a reasonable range: the lower limit is typically required to be greater than the effective bandwidth of the controlled object (to prevent excessive attenuation of the effective control signal), while the upper limit is limited by the sampling frequency of the flight control system. The lower frequency limit is given by the UAV designer during the design phase, while the upper limit needs to be determined based on the actual sampling frequency of the flight control system used.

[0062] The core advantage of this collaborative optimization strategy lies in its simultaneous optimization of the filter cutoff frequency and PID parameters, avoiding the suboptimal solution problem caused by the two-step method of "tuning the PID first and then designing the filter". After optimization, the robust PID parameters and filter configurations are shown in Table 3.

[0063] Table 3: Robust PID Parameter Tuning Results

[0064] S7. Large-scale Monte Carlo random sampling (typically M=2000 groups) is used to perform ultimate robustness verification on the optimized PID parameters and filter cutoff frequency. Verification metrics include: balancing success rate, minimum phase margin, minimum gain margin, and the maximum cost function value for all verification cases. Detailed verification steps are attached. Figure 7 As shown: S7.1. Random Validation Test Case Generation: Within the skewed range of the overall parameters determined in S2, M sets (typically M=2000 sets) of uncorrelated combinations of overall parameters are generated using the Monte Carlo random sampling method to form a large-scale validation test case set; if a correlation matrix between parameters is defined in S2, then a correlation sampling technique is required to generate samples that conform to the coupling relationship. S7.2. Closed-loop model construction and trim calculation: Fix the PID parameters and filter cutoff frequency obtained from S6 optimization into the controller, substitute M sets of verification cases into the nonlinear flight mechanics model in sequence, construct a closed-loop control system, and call the trim algorithm to calculate the trim solution of each verification case under the desired flight state, and record the number of verification cases that are successfully trimmed. S7.3. Closed-loop feature extraction and cost calculation: For the successful trim verification case, the closed-loop system is linearized with small disturbances, the open-loop transfer function of the system is extracted, and the phase margin and gain margin are calculated; at the same time, a unit gust disturbance is applied to perform nonlinear time-domain simulation, and the longitudinal cost Cost_lon_i and lateral cost Cost_lat_i corresponding to the verification case are calculated. S7.4. Statistics and Judgment: Summarize the calculation results of M sets of verification test cases, and calculate the balancing success rate, the minimum phase margin, the minimum gain margin among all successful balancing verification test cases, and the maximum cost function of all verification test cases; determine whether the above statistical indicators meet the preset safety constraints (such as balancing success rate 100%, phase margin ≥ 45°, gain margin ≥ 6dB, cost function below the threshold). If all are met, the verification passes, and the final PID parameters and filter cutoff frequency are output; if any indicator is not met, the verification fails, and the process returns to S6 for further optimization. The verification results are shown in Table 4.

[0065] Table 4: Monte Carlo Robust Validation Results

[0066] S8. Deploy the PID parameters and filter cutoff frequency into the hardware and verify them through a hardware-in-the-loop simulation system. The hardware-in-the-loop simulation system includes a flight control computer, server, actuators, sensor simulators, etc. Execute typical flight tasks such as trim checks, step response, disturbance suppression, and waypoint tracking in the hardware-in-the-loop environment to verify the performance of the parameters on the actual hardware. If problems are found during verification, return to S6 for iterative correction, forming a complete closed-loop process of "simulation → optimization → verification → hardware-in-the-loop → iteration".

[0067] The present invention has been described in detail above with reference to the embodiments, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A method for collaborative optimization of PID parameters and filter configuration for a solar-powered fixed-wing UAV, characterized in that, Includes the following steps: S1. Establish a nonlinear flight dynamics model for a solar-powered fixed-wing UAV; S2. Determine the range of deviation for the overall parameters; S3. Within the skew range, use Latin hypercube sampling to generate a probability coverage verification test case set; S4. Corner point strong injection based on cost function, and corner point combination in multidimensional parameter space is located by extreme value optimization, wherein the corner point combination is used to represent the most dangerous parameter combination; S5. Merge the probability coverage verification test case set with the corner point combination to form a hybrid verification test case set; S6. With the goal of minimizing the comprehensive cost function under the mixed verification test case set, the PID parameters and filter cutoff frequency are optimized in a coordinated manner. S7. Perform robustness verification on the optimized PID parameters and filter cutoff frequency. If the verification fails, return to S6 to continue optimization. If the verification passes, output the final PID parameters and filter cutoff frequency.

2. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: In S1, establishing a nonlinear flight dynamics model for a solar-powered fixed-wing UAV means: based on the geometric parameters, mass characteristics, and aerodynamic data of the solar-powered fixed-wing UAV, establishing a six-degree-of-freedom nonlinear flight dynamics model, which is expressed as: ， in, State variables x Regarding time t Find the first derivative. t Indicates time, x State variables, state variables x Includes body axis velocity ( u b , v b , w b ), angular velocity ( p , q , r ), attitude angle ( , , and the velocity of the Earth's axis ( V x , V y , V z ); u b , v b , w b These represent the forward, lateral, and vertical velocities along the body axis, respectively. p , q , r These represent the roll rate, pitch rate, and yaw rate in the body axis system, respectively. , , These represent the roll angle, pitch angle, and yaw angle, respectively. V x , V y , V z They represent x axis, y shaft and z Velocity in the axial direction; u Indicates the control variable, control variable u Including throttle opening δt Elevator deflection δe Aileron deflection δa and rudder deflection δr ; q flex For the generalized coordinate vector of the wing's elastic deformation, T panel For the temperature of the solar panel, W wind This is the atmospheric wind speed vector.

3. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: In S2, the overall parameters include mass, center of gravity position, moment of inertia, lift coefficient slope, zero lift drag coefficient, pitch moment coefficient, cruise speed, and solar panel temperature.

4. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: The specific steps of corner injection based on cost function in S4, which locates the most dangerous parameter combination in the multidimensional parameter space through extreme value optimization, are as follows: S4.

1. Construct the vertical cost function, which is defined as follows: Cost_lon=W1×R_height+W2×R_pitch, Where Cost_lon represents the longitudinal cost, R_height is the height drop risk indicator, R_pitch is the pitch divergence risk indicator, and W1 and W2 are both weighting coefficients. S4.

2. Construct the transverse cost function, which is defined as follows: Cost_lat=W3×R_roll+W4×R_yaw+W5×R_dutch, Where Cost_lat represents the lateral cost, R_roll is the roll divergence risk index, R_yaw is the yaw divergence risk index, R_dutch is the Dutch roll coupling risk index, and W3, W4 and W5 are all weighting coefficients. S4.

3. Use sequential quadratic programming or genetic algorithm to maximize the longitudinal cost function and the lateral cost function, locate the corner point combination in the multidimensional parameter space, and call the nonlinear flight mechanics model to calculate the trim state and linearization model in each iteration, and extract risk indicators.

5. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: In S5, when merging the probability coverage verification test case set and the corner point combination, the number of corner point combinations is 3 to 5% of the total number of verification test cases in the mixed verification test case set.

6. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 4, characterized in that: In S6, the comprehensive cost function is constructed as follows: First, we construct a mathematical model of a PID controller with a low-pass filter, expressed using a transfer function: ， in, G c ( s () is the filtered PID control law. s For the Laplace domain, G LPF ( s Let be the transfer function of a first-order low-pass filter, given by the low-pass filter cutoff frequency. fc The decision is made using the following formula: ； Secondly, the proportional gain of each control channel Kp Integral gain Ki Differential gain Kd and the corresponding low-pass filter cutoff frequency. fc To construct a collaborative optimization objective function for the variables, we need to define collaborative optimization.

7. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 6, characterized in that: In step S6, the specific steps for collaboratively optimizing the PID parameters and filter cutoff frequency with the objective of minimizing the comprehensive cost function under the mixed verification test case set are as follows: S6.

1. Randomly generate initial variables X ; S6.

2. Regarding variables X Calculate the filtered PID control law Gc ( s This process involves constructing a PID controller for the current iteration; iterating through each validation case in the mixed validation case set, calling the nonlinear flight mechanics model for balancing and linearization, and obtaining the open-loop transfer function. L ( s The system calculates the phase margin and gain margin based on the open-loop transfer function, and determines whether the phase margin and gain margin constraints are met. It also determines whether the balancing constraints and the low-pass filter cutoff frequency are satisfied. fc If the frequency band is within the specified range, calculate the longitudinal cost Cost_lon_i and lateral cost Cost_lat_i for the corresponding verification case, extract risk indicators, and then calculate the comprehensive cost function value for the verification case. J ; S6.

3. Based on the gradient and Hessian matrix update formula of the SLSQP algorithm, iteratively generate new variables, i.e., new […]. Kp , Ki , Kd , fc ]; S6.

4. Determine whether the maximum number of iterations has been reached or the objective function has converged. If yes, output the global optimal solution and obtain the final PID parameters and filter cutoff frequency. If not, return to S6.2 to continue iterating.

8. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: When determining the bias range of the overall parameters in S2, a correlation matrix is ​​used to describe the coupling relationship between the parameters; when generating the probability coverage verification test case set through Latin hypercube sampling in S3, the correlation of the parameters is maintained through correlation sampling technology; in S4, coupling constraints are used as one of the optimization constraints.

9. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: In S7, large-scale Monte Carlo random sampling is used to robustly verify the optimized PID parameters and filter cutoff frequency, including verifying the following indicators: balancing success rate, minimum phase margin, minimum gain margin, and maximum cost function of all verification cases.

10. The method for collaborative optimization of PID parameters and filter configuration for solar-powered fixed-wing UAVs according to claim 1, characterized in that: After S7 outputs the final PID parameters and filter cutoff frequency, the process further includes: S8 deploying the PID parameters and filter cutoff frequency into the hardware and verifying them through a loop simulation system. If a problem is found during verification, the process returns to S6.

Images