CN115657479A - Improved dynamic MOPSO-based control surface fault tolerance method for flying-wing unmanned aerial vehicle - Google Patents
Improved dynamic MOPSO-based control surface fault tolerance method for flying-wing unmanned aerial vehicle Download PDFInfo
- Publication number
- CN115657479A CN115657479A CN202211306318.3A CN202211306318A CN115657479A CN 115657479 A CN115657479 A CN 115657479A CN 202211306318 A CN202211306318 A CN 202211306318A CN 115657479 A CN115657479 A CN 115657479A
- Authority
- CN
- China
- Prior art keywords
- control
- control surface
- fault
- deflection
- individuals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention discloses a fault tolerance method for a control surface of a flying wing unmanned aerial vehicle based on improved dynamic MOPSO, and belongs to the technical field of calculation, calculation or counting. The invention provides a fault tolerance method based on improved dynamic multi-target particle swarm control redistribution aiming at the comprehensive characteristics of redundancy, limited control surface, strong coupling, strong nonlinearity and the like of the control surface of an advanced layout flying wing unmanned aerial vehicle, effectively processes the control surface characteristics of the flying wing unmanned aerial vehicle, reduces the influence of the control surface fault during flight control, and ensures the stability and tracking precision of a control system.
Description
Technical Field
The invention relates to an advanced layout unmanned aerial vehicle fault tolerance technology, and particularly discloses a control surface fault tolerance method of a flying wing unmanned aerial vehicle based on an improved dynamic multi-target PSO (Particle Swarm Optimization), belonging to the technical field of calculation, calculation or counting.
Background
With the increasingly greater limitation on the function of the unmanned aerial vehicle with the traditional layout in the modern complex war environment, the flying wing layout becomes the first-choice aerodynamic layout of the current high-performance unmanned aerial vehicle due to the advantages of the integrated design of removing vertical tails, strong stealth performance, good aerodynamic efficiency, high lift-drag ratio and the like, and meanwhile, the configuration of multiple groups of redundant control surfaces of the flying wing layout can greatly improve the reliability and the control capability of the unmanned aerial vehicle system; however, the characteristic of compact layout of the control surfaces corresponds to the redundant operation capability, which causes the problems of serious aerodynamic cross coupling effect between the control surfaces, strong deflection nonlinearity of the control surfaces, high failure probability of an actuator and the like, and once a failure occurs and effective measures are not taken timely for detection, positioning, isolation and removal, the whole flight control system can be disabled and paralyzed, so that irreparable loss is caused. The method for linear control distribution and fault tolerance is difficult to meet the requirement of control tracking precision, so that the fault tolerance control problem of the flying wing layout unmanned aerial vehicle needs to comprehensively consider the problems of strong coupling of control surfaces, strong nonlinearity, control scalar constraint, reduction of fault probability and multi-target comprehensive optimization of fault tolerance tracking real-time performance and precision.
Fault tolerant control research has developed to date and is mainly divided into two categories, namely: active fault-tolerant control based on fault information and passive fault-tolerant control based on a strong robustness control system, and methods generally applied to the active fault-tolerant control of the faults of the flight control system generally comprise an adaptive observer, control redistribution, predictive control, sliding mode control, a neural network and the like.
Control reassignment becomes the main approach to solving the redundant control surface reconfiguration of a multi-control surface aircraft, which divides the flight control law into two relatively independent modules: the first is a basic control law module; and the second is a control reallocation module. When the control surface has a fault, the system reconstruction can be realized by using the control distribution module without readjusting a complex flight control law, so that the design difficulty of the traditional reconstruction technology is reduced. The multi-control surface configuration of the flying wing layout unmanned aerial vehicle improves the control difficulty, and provides more possibilities for realizing fault-tolerant control under the condition of actuator failure by utilizing redundant control surfaces through control redistribution.
The reconstruction method based on control distribution is divided into a linear distribution method and a nonlinear distribution method in terms of distribution principle. The linear distribution method mainly comprises a pseudo-inverse method, a chain distribution method, a direct distribution method, a linear programming method and the like; the nonlinear distribution method mainly comprises an intercept correction method, a piecewise linear programming method, a nonlinear programming method, intelligent control distribution and the like. In the field of aircraft control, scholars at home and abroad have achieved more achievements in engineering application and theoretical research of control distribution reconstruction technology.
However, because the flying wing layout unmanned aerial vehicle has strong control plane deflection nonlinearity and control plane cross coupling nonlinearity, the conventional fault-tolerant control strategy based on control redistribution is difficult to be applied to the flying wing unmanned aerial vehicle, and the fault-tolerant control research result aiming at the control plane fault of the flying wing aircraft is less at present.
Aiming at the control surface characteristics of the flying wing unmanned aerial vehicle, an intelligent multi-objective optimization algorithm is introduced to solve the problem of multi-control-surface control redistribution of strong nonlinear control surface characteristics, and compared with the traditional algorithm, the intelligent multi-objective optimization algorithm has the characteristics of self-organization, self-adaptation, self-learning and the like, and further has lower dependence degree on the accuracy of introduced fault information. And redistributing the target characteristics according to the nonlinear control, and selecting appropriate individual and global optimal solutions to make the population converge to an expected optimal solution. Compared with other swarm intelligent optimization algorithms, the particle swarm optimization algorithm has higher convergence speed and is more suitable for the problem of dynamic control reallocation with certain requirements on real-time performance.
At present, the traditional single-target PSO only can consider the single problem of virtual instruction tracking, but cannot further comprehensively consider the requirements of energy consumption, smooth control and the like; the PSO converts the problem into a single-target optimization algorithm, reduces the complexity of the problem to reduce time consumption, but only can compromise and select an optimal solution among targets, and the solution is difficult to adapt to the condition of any instruction change; in the traditional multi-target PSO, all performance targets are comprehensively considered, the optimal solution is determined from each target optimal solution set according to the requirement, and the real-time performance of the algorithm is sacrificed to a certain extent. In the problem of fault-tolerant control, the targets of control redistribution pairing instruction tracking, smooth control and the like are difficult to be optionally selected and compromise, the requirement on the precision of instruction tracking is high, meanwhile, the requirement on the real-time performance of fault tolerance is high, and the single-target and multi-target PSO optimization algorithm is difficult to apply to the system.
Therefore, the invention aims to provide a multi-target particle swarm optimization algorithm based on a composite trajectory prediction strategy, which optimizes and improves the basic multi-target PSO and solves the problem of redistribution of the control surfaces of faults considering the nonlinearity of control surface deflection and the cross coupling effect so as to overcome the defects that the conventional linear fault-tolerant strategy cannot be applied to the complex control characteristics of the flying wing unmanned aerial vehicle and the existing intelligent optimization algorithm cannot meet the contradiction between the high-precision tracking and the real-time requirement of the control redistribution problem after the control surface of the flying wing unmanned aerial vehicle is in fault.
Disclosure of Invention
The invention aims to provide a fault-tolerant method for control surfaces of an unmanned aerial vehicle with flying wings based on improved dynamic MOPSO (metal oxide semiconductor optical power station), aiming at faults of actuators of an unmanned aerial vehicle with multiple control surfaces, based on a multi-target particle group control redistribution framework under the control surface faults, a composite track multi-directional prediction model is adopted to reinitialize initial population at the next moment, the iteration times of particle swarms are reduced, the optimization speed of a global optimal solution is improved, and the real-time performance and the reliability of fault tolerance are ensured.
The invention adopts the following technical scheme for realizing the aim of the invention:
considering the influence of the non-linearity of the control surface deflection and the cross coupling non-linearity of the flying wing layout unmanned aerial vehicle, calculating three-channel control moment coefficients of rolling, pitching and yawing by adopting a non-linear dynamic reverse flight control law, and constructing a virtual control instruction; the method comprises the steps of establishing a multi-objective function according to virtual instruction tracking constraints, energy consumption indexes and smooth control indexes, introducing fault estimation information fed back by a fault diagnoser, adopting an improved dynamic multi-objective particle swarm optimization (MOPSO) algorithm to solve control surface deflection instructions of the multi-objective optimization function to the maximum extent under the condition of control surface faults, reducing the fault influence of an actuator, and realizing control surface redistribution control after the faults.
The improved MOPSO algorithm initializes an individual optimal solution set to be an initial population in each control surface deflection variable constraint range, calculates a Pareto optimal solution set according to each particle in the population and selects a global optimal solution, when the change of a dynamic environment or the fault of an actuator is detected, a plurality of groups of particles capable of describing the shape of the optimal solution set are selected from the Pareto optimal solution set to be used as representative individuals, the rest particles in the optimal solution set are divided into different common evolutionary clusters according to distances, the optimal solution set which is possible at the next moment is predicted according to a composite track prediction model based on fault information, and the predicted solution is used as the initial population at the next moment.
Considering the influence of the control surface deflection nonlinearity and the cross coupling nonlinearity, fitting a nonlinear dynamic efficiency model of the control surface as follows:
wherein the virtual control instruction v = [ C = lδ ，C mδ ，C nδ ] T C (delta) is a three-channel control moment coefficient of rolling, pitching and yawing, C is a mapping relation between the control moment coefficient and control surface deflection, and delta = [ delta ] l1 ，δ l2 ，δ l3 ，δ l4 ，δ r1 ，δ r2 ，δ r3 ，δ r4 ] T To control the amount of surface deflection, delta l1 ，δ l2 ，δ l3 ，δ r1 ，δ r2 ，δ r3 Respectively left and right elevon, delta l4 ，δ r4 Are respectively a left resistance rudder and a right resistance rudder,
the roll control moment coefficient and the pitch control moment coefficient of the ith control surface respectively,
a non-linear fitting expression, p, for the yaw channel control moment coefficient of the ith control surface i3 ，p i2 ，p i1 ，p i0 Is a cubic coefficient, a quadratic coefficient, a primary coefficient and a constant term of a nonlinear fitting expression,
a third lifting aileron at the left side and the right side and a resistance rudder at the same side are respectively cross-coupled with a tee jointRoad control moment coefficient;
control surface faults of the flying-wing unmanned aerial vehicle are usually damage, jamming, saturation, floating and the like, delta is a control surface deflection input command, f a (delta, t) is the control surface deflection output after the fault, and the fault relation of the control surface deflection output and the fault relation is parameterized as
Wherein Λ = diag { α) 1 ，α 2 ，α 3 ，α 4 ，α 5 ，α 6 ，α 7 ，α 8 }，Γ＝diag{γ 1 ，γ 2 ，…，γ 8 }，
α i Characterizing the type of failure as i =1, the i-th control surface is completely deactivated (seized or floating), when α is i If =0, the i-th control surface is normal or partially failed. k is a radical of i For residual efficiency under damage fault, gamma i =0 denotes a floating failure of the i-th control surface, 0 < γ i < 1 indicates the remaining operational capacity of the i-th control surface in the event of a partial failure, γ i And =1 indicates that the control surface is operating normally. Delta i (t F ) For control surfaces at t F A stuck position where the moment is completely failed.
Respectively establishing a PSO optimization objective function as
Wherein v is d In order to expect a virtual control command for three channels calculated according to a flight control law, v is a three-channel control moment coefficient calculated by a control surface deflection position distributed by a particle swarm according to a nonlinear dynamic efficiency model of a control surface, delta (t) and delta (t-1) are deflection amounts of the control surface at the time t and the time t-1 respectively, | | \ | | | zero 2 Expressing the two norms of the matrix, and constraining the values of the control variables of each control surface as follows:
wherein the content of the first and second substances,
for controlling the speed of change of surface deflection, delta min ，δ max The minimum and maximum values of deflection of each control surface,
the minimum value and the maximum value of the deflection speed of each control surface.
And performing predictive analysis on the initial population of the particle swarm at the next moment by adopting a compound track multidirectional prediction model, wherein the prediction track is formed by compounding a time sequence prediction model and a linearized control effect inverse model, and the positions of the particles represent the deflection positions of the control surface.
And (4) through clustering analysis, taking the extreme value solution of each objective function in the optimal solution set PS and the center of the PS as an initial representative individual set. PS center at time t
Is defined as
Wherein, PS t PS, | PS at time t t I denotes PS t The number of the intermediate solutions is equal to or greater than the total number of the intermediate solutions,
is PS t The ith solution in (1). And dividing the rest individuals in the PS into different clusters formed by corresponding representative individuals, wherein each representative individual is the center of the corresponding cluster. If the number of the initial representative individuals (namely the number of the current clusters) does not meet the requirement, selecting the individual farthest from the center of the cluster from the cluster with the largest radius as a new representative individual, reselecting the individual closest to the new representative individual, and dividing to form a new cluster. Repeating the steps until the number of representative individuals and clusters meets the requirement. Wherein the Euclidean distance between the ith individual and the jth individual is
After the individuals in all the PS are divided into clusters with K representative individuals as centers, the individuals in each cluster follow the evolution track of the center of the corresponding cluster together, and the same prediction model is adopted to generate the initial population at the next moment.
Selecting the number of representative individuals according to the intensity of environmental change, considering the environmental change caused by the failure of the control surface of the unmanned aerial vehicle, providing a measurement lambda (t) reflecting the severity of the environmental change according to the influence of the failure, and providing N before and after the environmental change P The variation degree of the objective function value of each individual is expressed as:
wherein Δ f j (Ar i )＝[(f j (Ar i ，g)-f i (Ar i ，g-1))/(u j (g)-l j (g))]，
Δf j (Ar i ) Is the jth objective function value u of the ith individual in the external file at the g iteration j (g) And l j (g) Respectively representing the maximum and minimum values, η, of the jth objective function in the g-th iteration 1 ，η 2 ，…，η M Weights respectively corresponding to the variation degrees of the M objective functions; the number of representative near-optimal solutions can be adaptively determined according to the degree of environmental change as defined above
K 1 And K 2 Respectively the upper and lower boundaries of K, and selecting K through test verification 1 = M +1 and K 2 ＝3M。
The time sequence prediction model constructs a plurality of time sequence models based on historical information provided by representative individuals in the first two environments to generate several different possible evolution directions to fully predict the motion of the PS, namely, the optimal allocation solution set of the control plane deflection possibility under the redistribution conditionThe change trajectory of (2); two representative sets of individuals at times t and t-1 are labeled separately
A set of centers for each cluster; for C t Of (2)
The proposed strategy is first in set C t-1 In the search for the closest representative individual, note
Consider it as
The parent of (2) adopts Euclidean distance as the measure of the closeness degree between individuals, and when the control surface does not have faults and dynamic environment changes, the individuals
The direction of evolution is
When partial control surface faults occur, if the environment is not changed, the optimal solution set which is a part of the new initial population is
If the dynamic environment changes, the evolution track is
When a control surface with faults is newly added, an optimal solution set containing fault information of the first two moments is used as a part of a new initial population, namely
According to the requirements of control distribution and fault tolerance, in order to reduce the failure probability of the control surface, when the deflection angle of the resistance rudder is smaller, a linear rudder effect inverse model is adoptedProviding the change direction of the solution set to obtain an auxiliary evolution track of a prediction solution; the objective function may be at point δ t Near linearization as Δ v t ＝BΔδ t Wherein δ t ＝[δ l1 t ，δ l2 t ，δ l3 t ，δ l4 t ，δ r1 t ，δ r2 t ，δ r3 t ，δ r4 t ] T For the optimal allocation result at time t, B is at δ t Control distribution efficiency matrix after nearby linearization:
Δv t ＝[ΔC lδ ，ΔC mδ ，ΔC nδ ] T the torque coefficient increment is controlled for the three channels to be allocated,
to control the moment coefficient C i About control surface delta j Partial differential of (a); according to the deflection relation of the control surface in the fault state, the increment of the control moment coefficient is converted into the following steps:
when the fault state of the control surface at the time t is consistent with the time t-l, the least square solution delta of the formula is adopted t The evolution track as the prediction optimal solution guides the population to evolve near the optimal solution after the environment changes, namely delta t ＝Γ -1 (I-Λ) -1 (B T B) -1 B T (v t -v t-1 ) Wherein Γ = Γ t ＝Γ t-1 ，Λ＝Λt＝Λ t-1 ；
According to the time sequence prediction model and the linearization inverse model, when the environment change is detected, the possible individuals in the jth cluster of the new prediction population
Can be selected from corresponding individuals
Is specifically shown as
Wherein
In order to predict the set of evolution directions,
to predict the set of linear inverse models, i =1,2, …, N P ，j＝1，2，…，K，ξ 1 And xi 2 Is a weight of the prediction model and satisfies ξ 1 +ξ 1 =1, gaussian disturbance epsilon t ～(0，σ t ) Is based on the mean value of 0 and the variance σ t For increasing the diversity of the population, sigma t Satisfy the requirements of
Since the cross-coupling nonlinearity caused by the resistance rudder deflection affects the prediction accuracy, the weight ξ 2 Related to the deflection of the resistance rudder at time t, in particular
N may be generated from a predictive model according to the above process P New possible individuals, the rest of the initial population N-N P Individuals are randomly generated in a decision space;
the particle velocity and position are updated according to the following expression:
in the formula, v i Denotes the particle velocity, x i As the position of the current particle, p ibest Represents the optimal position of the ith particle, g ibest Represents the global optimal position of the ith particle, c 1 ，c 2 Is a learning factor, r 1 ，r 2 Has a range of [0,1]A random number in between. And updating the individual optimal solution sets according to a Pareto domination relation when the particles in the two generations of individual optimal solution sets before and after updating meet the convergence index, performing variation operation on the particles when the convergence index is not met, sorting the updated particles according to the concentration, screening the individual optimal solution sets and the global optimal solution sets, and selecting one global optimal solution as a control plane deflection instruction for controlling redistribution of the fault control surface.
By adopting the technical scheme, the invention has the following beneficial effects:
(1) According to the invention, a fault tolerance strategy based on control redistribution is designed aiming at the redundant characteristics of the control surface of the flying wing unmanned aerial vehicle, the control surface deflection and cross coupling nonlinear characteristics, the precision requirement of tracking control and the real-time requirement of fault tolerance control are comprehensively considered, the influence of the control surface fault on flight control is reduced, and the stable flight of the flying wing unmanned aerial vehicle is realized.
(2) The method improves the traditional dynamic multi-target particle swarm algorithm, combines the requirements of control distribution targets and fault-tolerant control reliability and real-time performance, dynamically generates the initial particle swarm at each moment by improving the generation mode of the initial population at the next moment after the change of the redistribution targets and introducing a multi-direction prediction mechanism of a composite track, accelerates the process of convergence of the population to the expected precision requirement of each target, and meanwhile, the multi-direction prediction mechanism can effectively save the shape and direction characteristics of an optimal solution set, increases the diversity of the population and improves the optimization performance.
Drawings
Fig. 1 is an architecture diagram of a fault tolerant method for control surfaces of a flying wing drone based on an improved dynamic MOPSO according to the present invention.
FIG. 2 is a flow chart of the multi-target particle swarm algorithm based on the composite trajectory prediction strategy.
Fig. 3 is a fault curve for the first elevon on the right-hand side of the invention at 2s onset of a stuck-at fault.
Fig. 4 is a time response plot of the attitude angle of the first elevon at the beginning of 2s in the stuck-at fault condition of the present invention.
Fig. 5 is a time response graph of attitude angular velocity of the first elevon at the right-hand side at 2s onset in a stuck-at fault condition of the present invention.
FIG. 6 is a tracking response diagram of three-channel control moment coefficients in a stuck fault state of the first elevon at the right side after 2 s.
Fig. 7 is a control surface redistribution graph of the present invention in the event of a stuck-at-fault condition of the first elevon at the beginning of 2s on the right side.
Detailed Description
In order to facilitate understanding of those skilled in the art, the present invention will be further described with reference to the accompanying drawings.
The invention provides a fault-tolerant method for control surfaces of a flying wing unmanned aerial vehicle based on an improved dynamic MOPSO (mean-particle swarm optimization), which solves the problem of fault-tolerant control of an aircraft with redundant control surfaces after an actuator fails. Fig. 1 reflects implementation steps of control reallocation after control surface failure of a flying wing drone, and specifically includes the following four steps.
The method comprises the following steps of firstly, considering the influence of control surface deflection nonlinearity and cross coupling nonlinearity, calculating three-channel control moment coefficients of rolling, pitching and yawing by adopting a nonlinear dynamic inverse flight control law, constructing a virtual control instruction v, and fitting a nonlinear dynamic efficiency model of the control surface as follows:
wherein the virtual control instruction v = [ C = lδ ，C mδ ，C nδ ] T C (delta) is a three-channel control moment coefficient of rolling, pitching and yawing, C is a mapping relation between the control moment coefficient and the deflection of the control surface, and delta is = [ delta ] l1 ，δ l2 ，δ l3 ，δ l4 ，δ r1 ，δ r2 ，δ r3 ，δ r4 ] T To control the amount of deflection of the surface, delta l1 ，δ l2 ，δ l3 ，δ r1 ，δ r2 ，δ r3 Respectively left and right elevon, delta l4 ，δ r4 Are respectively a left resistance rudder and a right resistance rudder,
the roll control moment coefficient and the pitch control moment coefficient of the ith control surface respectively,
a non-linear fitting expression, p, for the yaw channel control moment coefficient of the ith control surface i3 ，p i2 ，p i1 ，p i0 Is a cubic coefficient, a quadratic coefficient, a primary coefficient and a constant term of a nonlinear fitting expression,
three channels are respectively used for controlling moment coefficients by the cross coupling of the left third lifting aileron and the right third lifting aileron on the left side and the right third lifting aileron on the same side with the resistance rudder on the same side;
step two, the control surface faults of the flying wing unmanned aerial vehicle are usually damage, jamming, saturation, floating and the like, delta is a control surface deflection input instruction, f is a control surface deflection input instruction a (delta, t) is the control plane deflection output after the fault, and the fault relation of the control plane deflection output and the fault relation is parameterized as
Wherein Λ = diag { α) 1 ，α 2 ，α 3 ，α 4 ，α 5 ，α 6 ，α 7 ，α 8 }，Γ＝diag{γ 1 ，γ 2 ，…，γ 8 }，
α i Characterizing the type of failure as i =1, the i-th control surface is completely deactivated (seized or floating), when α is i If =0, the i-th control surface is normal or partially failed. k is a radical of i For residual efficiency under damage fault, gamma i =0 denotes a floating failure of the i-th control surface, 0 < γ i < 1 denotes the firstResidual capacity of i control surfaces in the event of partial failure, gamma i And =1 indicates that the control surface is operating normally. Delta. For the preparation of a coating i (t F ) For control surfaces at t F A stuck position that is completely failed at that moment.
Step three, respectively establishing a PSO optimization objective function as
Wherein v is d V is a three-channel expected virtual control command calculated according to a flight control law, v is a three-channel control moment coefficient calculated by a control surface deflection position distributed by a particle swarm according to a control surface nonlinear dynamic efficiency model, delta (t) and delta (t-1) are deflection amounts of control surfaces at the time t and the time t-1 respectively, | | 2 Expressing the two norms of the matrix, and the value constraint of each control surface manipulated variable is as follows:
wherein the content of the first and second substances,
for controlling the speed of change of surface deflection, delta min ，δ max The minimum and maximum values of deflection of each control surface,
the minimum value and the maximum value of the deflection speed of each control surface.
Fourthly, performing predictive analysis on the initial population of the particle swarm at the next moment by adopting a composite track multidirectional prediction model, wherein the prediction track is formed by compounding a time sequence prediction model and a linearized control effect inverse model, and the positions of the particles represent the deflection positions of the control surface;
and (4) through clustering analysis, taking the extreme value solution of each objective function in the optimal solution set PS and the center of the PS as an initial representative individual set. PS center at time t
Is defined as
Wherein, PS t Is PS, | PS at time t t I denotes PS t The number of the intermediate solutions is equal to or greater than the total number of the intermediate solutions,
is PS t The ith solution in (1). And dividing the rest individuals in the PS into different clusters formed by corresponding representative individuals, wherein each representative individual is the center of the corresponding cluster. If the number of the initial representative individuals (namely the number of the current clusters) does not meet the requirement, selecting the individuals farthest from the center of the clusters from the clusters with the largest radius as new representative individuals, reselecting the individuals closest to the new representative individuals, and dividing into new clusters. Repeating the steps until the number of representative individuals and clusters meets the requirement. Wherein the Euclidean distance between the ith individual and the jth individual is
After the individuals in all the PS are divided into clusters taking K representative individuals as centers, the individuals in each cluster follow the evolution track of the corresponding cluster center together, and the same prediction model is adopted to generate the initial population at the next moment.
Step 4.1, the fault tolerance method for the control surface of the flying wing unmanned aerial vehicle based on the improved dynamic MOPSO is characterized in that the number of representative individuals is selected according to the intensity of environmental change, the environmental change caused by the failure of the control surface of the unmanned aerial vehicle is considered, a measurement lambda (t) reflecting the severity of the environmental change is provided according to the influence of the failure, and N is the number before and after the environmental change P The variation degree of the objective function value of each individual is expressed as:
wherein Δ f j (Ar i )＝[(f j (Ar i ，g)-f j (Ar i ，g-1))/(u j (g)-l j (g))]，
Δf j (Ar i ) Is the jth objective function value u of the ith individual in the external file at the g iteration j (g) And l i (g) Respectively representing the maximum and minimum values, η, of the jth objective function in the g-th iteration 1 ，η 2 ，…，η M Weights corresponding to the variation degrees of the M objective functions respectively; the number of representative near-optimal solutions can be adaptively determined according to the degree of environmental change as defined above
K 1 And K 2 Respectively the upper and lower boundaries of K, and selecting K through test verification 1 = M +1 and K 2 ＝3M。
And 4.2, constructing a predicted evolution direction of the initial population according to the selected representative individuals.
Step 4.2.1, the time sequence prediction model constructs a plurality of time sequence models based on historical information provided by representative individuals in the former two environments to generate several different possible evolution directions, and fully predicts the motion of the PS, namely the change track of the optimal distribution solution set of the control plane deflection possible under the redistribution condition; two representative sets of individuals at times t and t-1 are labeled separately
And
a set of centers for each cluster; for C t Of (2)
The proposed strategy is first in set C t-1 In the search for the closest representative individual, note
Consider it as
The parent of (2) adopts Euclidean distance as the measure of the closeness degree between individuals, and when the control surface does not have faults and dynamic environment changes, the individuals
The direction of evolution is
When partial control surface faults occur, if the environment is not changed, the optimal solution set which is a part of the new initial population is
If the dynamic environment changes, the evolution track is
When a control surface with faults is newly added, an optimal solution set containing fault information of the first two moments is used as a part of a new initial population, namely
Step 4.2.2, according to the requirements of control distribution and fault tolerance, in order to reduce the failure probability of the control surface, when the deflection angle of the resistance rudder is small, a linear rudder effect inverse model is adopted to provide the change direction of a solution set, and an auxiliary evolution track of a prediction solution is obtained; the objective function may be at point δ t Near linearization as Δ v t ＝BΔδ t Wherein δ t = [ δ ] l1 t ，δ l2 t ，δ l3 t ，δ l4 t ，δ r1 t ，δ r2 t ，δ r3 t ，δ r4 t ] T For the optimal allocation result at time t, B is at δ t Control distribution efficiency matrix after nearby linearization:
Δv t ＝[ΔC lδ ，ΔC mδ ，ΔC nδ ] T the torque coefficient increment is controlled for the three channels to be allocated,
for controlling the moment coefficient Ci with respect to the control surface delta j Partial differential of (a); according to the deflection relation of the control surface in the fault state, the increment of the control moment coefficient is converted into the following steps:
when the fault state of the control surface at the time t is consistent with the time t-1, the least square solution delta of the formula is adopted t The evolution track as the prediction optimal solution guides the population to evolve near the optimal solution after the environment changes, namely delta t ＝Γ -1 (I-Λ) -1 (B T B) -1 B T (v t -v t-1 ) Wherein Γ = Γ t ＝Γ t-1 ，Λ＝Λ t ＝Λ t-1 ；
And 4.3, constructing a compound track multidirectional evolution model according to the predicted evolution direction of the initial population. According to the time sequence prediction model and the linearization inverse model, when the environment change is detected, the possible individuals in the jth cluster of the population are newly predicted
Can be selected from corresponding individuals
Is specifically shown as
Wherein
In order to predict the set of evolution directions,
to predict the set of linear inverse models, i =1,2, …, N P ，j＝1，2，…，K，ξ 1 And xi 2 Is the weight of the prediction model and satisfies xi 1 +ξ 1 =1, gaussian disturbance ε t ～(0，σ t ) Is based on the mean value of 0 and the variance σ t For increasing the diversity of the population, sigma t Satisfy the requirement of
Since the cross-coupling nonlinearity caused by the resistance rudder deflection affects the prediction accuracy, the weight ξ 2 Related to the deflection of the resistance rudder at time t, in particular
N may be generated from a predictive model according to the above process P New possible individuals, the rest of the initial population N-N P Individuals are randomly generated in a decision space;
and 4.4, updating the particle speed and the particle position according to the following expression:
in the formula, v i Denotes the particle velocity, x i As the position of the current particle, p ibest Represents the optimal position of the ith particle, g ibest Representing the global optimum position of the ith particle, c 1 ，c 2 Is a learning factor, r 1 ，r 2 Has a range of [0,1]A random number in between. And updating the individual optimal solution sets according to a Pareto domination relation when the particles in the two generations of individual optimal solution sets before and after updating meet the convergence index, performing variation operation on the particles when the convergence index is not met, sorting the updated particles according to the concentration, screening the individual optimal solution sets and the global optimal solution sets, and selecting one global optimal solution as a control plane deflection instruction for controlling redistribution of the fault control surface.
The invention carries out fault-tolerant control simulation on the flying-wing unmanned aerial vehicle, establishes a control redistribution module based on a composite track multidirectional prediction strategy under the condition of a dynamic multi-target particle swarm control surface fault in the simulation, and carries out the simulation process in MATLAB. Fig. 3 to fig. 6 show that when a control surface has a stuck fault in 2s, the fault-tolerant strategy based on improved particle swarm control reallocation can quickly complete reallocation of the remaining control surfaces, and each controlled state quantity is kept in an expected stable state, and the three-channel virtual control moment coefficient command to be allocated is still accurately tracked, so that the real-time performance and the stability are better.
Claims (8)
1. A fault tolerance method for a control surface of a flying wing unmanned aerial vehicle based on an improved dynamic MOPSO is characterized in that the influence of the deflection nonlinearity and cross coupling nonlinearity of the control surface of the flying wing unmanned aerial vehicle is considered, a nonlinear dynamic inverse flight control law is adopted to calculate three-channel control moment coefficients of rolling, pitching and yawing, and a virtual control instruction is constructed; establishing a multi-target function according to the virtual instruction tracking constraint, the energy consumption index and the smooth control index, introducing fault estimation information fed back by a fault diagnoser, adopting an improved dynamic multi-target particle swarm optimization (MOPSO) algorithm to solve the control surface deflection instruction of the multi-target optimization function to the maximum extent under the condition of control surface fault, reducing the fault influence of an actuator and realizing control surface redistribution control after the fault;
the improved MOPSO algorithm initializes an individual optimal solution set to be an initial population in each control surface deflection variable constraint range, calculates a Pareto optimal solution set according to each particle in the population and selects a global optimal solution, when the change of a dynamic environment or the fault of an actuator is detected, a plurality of groups of particles capable of describing the shape of the optimal solution set are selected from the Pareto optimal solution set to be used as representative individuals, the rest particles in the optimal solution set are divided into different common evolutionary clusters according to distances, the optimal solution set which is possible at the next moment is predicted according to a composite track prediction model based on fault information, and the predicted solution is used as the initial population at the next moment.
2. The fault tolerance method for the control surfaces of flying wing drones based on the improved dynamic MOPSO is characterized in that the influence of nonlinear control surface deflection and nonlinear cross coupling is considered, and a nonlinear dynamic efficiency model of the control surfaces is fitted as follows:
wherein the virtual control instruction v = [ C = lδ ，C mδ ，C nδ ] T C (delta) is a three-channel control moment coefficient of rolling, pitching and yawing, C is a mapping relation between the control moment coefficient and control surface deflection, and delta = [ delta ] l1 ，δ l2 ，δ l3 ，δ l4 ，δ r1 ，δ r2 ，δ r3 ，δ r4 ] T To control the amount of surface deflection, delta l1 ，δ l2 ，δ l3 ，δ r1 ，δ r2 ，δ r3 Respectively, left and right elevon, delta l4 ，δ r4 Are respectively a left resistance rudder and a right resistance rudder,
the roll control moment coefficient and the pitch control moment coefficient of the ith control surface respectively,
a non-linear fitting expression, p, for the yaw channel control moment coefficient of the ith control surface i3 ，p i2 ，p i1 ，p i0 Is a cubic coefficient, a quadratic coefficient, a primary coefficient and a constant term of a nonlinear fitting expression,
three channels are respectively used for controlling moment coefficients by the cross coupling of the left third lifting aileron and the right third lifting aileron on the left side and the right third lifting aileron on the same side with the resistance rudder on the same side;
control surface failures of flying wing drones are usually damage, seizing, saturation, floatationWhen delta is the control surface deflection input command, f a (delta, t) is the control surface deflection output after the fault, and the fault relation of the control surface deflection output and the fault relation is parameterized as
Wherein Λ = diag { α) 1 ，α 2 ，α 3 ，α 4 ，α 5 ，α 6 ，α 7 ，α 8 }，Γ＝diag{γ 1 ，γ 2 ，…，γ 8 }，
α i Characterizing the type of failure as i =1, the i-th control surface is completely deactivated (seized or floating), when α is i If =0, the i-th control surface is normal or partially failed. Gamma ray i For residual efficiency under damage fault, gamma i =0 denotes a floating failure of the i-th control surface, 0 < γ i < 1 indicates the remaining operational capacity of the i-th control surface in the event of a partial failure, γ i =1 denotes that the control surface is working properly. Delta i (t F ) For control surfaces at t F A stuck position that is completely failed at that moment.
3. The method of claim 1, wherein the PSO optimization objective function is respectively established as
Wherein v is d V is a three-channel expected virtual control command calculated according to a flight control law, v is a three-channel control moment coefficient calculated by a control surface deflection position distributed by a particle swarm according to a control surface nonlinear dynamic efficiency model, delta (t) and delta (t-1) are deflection amounts of control surfaces at the time t and the time t-1 respectively, | | 2 Expressing the two norms of the matrix, and the value constraint of each control surface manipulated variable is as follows:
wherein the content of the first and second substances,
to control the varying speed of deflection of the surface, delta min ，δ max The minimum value and the maximum value of the deflection of each control surface,
the minimum value and the maximum value of the deflection speed of each control surface.
4. The fault-tolerant method for the control surface of the flying wing unmanned aerial vehicle based on the improved dynamic MOPSO is characterized in that a composite track multidirectional prediction model is adopted to perform prediction analysis on an initial population at the next moment of a particle swarm, the prediction track is formed by compounding a time sequence prediction model and a linearized control effect inverse model, and the positions of particles represent the deflection positions of the control surface;
and (4) through clustering analysis, taking the extreme value solution of each objective function in the optimal solution set PS and the center of the PS as an initial representative individual set. PS center at time t
Is defined as
Wherein, PS t Is PS, | PS at time t t I denotes PS t The number of the intermediate solutions is equal to or greater than the total number of the intermediate solutions,
is PS t The ith solution in (1). And dividing the rest individuals in the PS into different clusters formed by corresponding representative individuals, wherein each representative individual is the center of the corresponding cluster. If the number of the initial representative individuals (namely the number of the current clusters) does not meet the requirement, selecting the individual farthest from the center of the cluster from the cluster with the largest radius as a new representative individual, and reselectingThe individuals closest to the new representative individual are divided to form new clusters. Repeating the steps until the number of representative individuals and clusters meets the requirement. Wherein the Euclidean distance between the ith individual and the jth individual is
After the individuals in all the PS are divided into clusters taking K representative individuals as centers, the individuals in each cluster follow the evolution track of the corresponding cluster center together, and the same prediction model is adopted to generate the initial population at the next moment.
5. The method as claimed in claim 4, wherein the number of representative individuals is selected according to the severity of environmental changes, the environmental changes caused by the failure of the control surface of the UAV are considered, a measure λ (t) reflecting the severity of the environmental changes is provided according to the influence of the failure, and N is a measure before and after the environmental changes P The variation degree of the objective function value of each individual is expressed as:
wherein Δ f i (Ar i )＝[(f j (Ar i ，g)-f j (Ar i ，g-1))/(u j (g)-l j (g))]，
Δf j (Ar i ) Is the jth objective function value u of the ith individual in the external file at the g iteration j (g) And l j (g) Respectively representing the maximum and minimum values, η, of the jth objective function in the g-th iteration 1 ，η 2 ，…，η M Weights respectively corresponding to the variation degrees of the M objective functions; the number of representative near-optimal solutions can be adaptively determined according to the degree of environmental change as defined above
K 1 And K 2 Respectively the upper and lower boundaries of K, and selecting K through test verification 1 = M +1 and K 2 ＝3M。
6. The fault-tolerant control surface method for flying-wing unmanned aerial vehicles based on the improved dynamic MOPSO as claimed in claim 4, characterized in that the time sequence prediction model constructs a plurality of time sequence models based on historical information provided by representative individuals in the first two environments to generate several different possible evolution directions, and fully predicts the motion of PS, namely the change trajectory of the optimal distribution solution set of the control surface deflection possible under the redistribution condition; two representative sets of individuals at times t and t-1 are labeled separately
And
a set of centers for each cluster; for C t Of (2)
The proposed strategy first comes in the set C t-1 Find the closest representative individual, record
Consider it as
The parent of (2) adopts Euclidean distance as the measure of the closeness degree between individuals, and when the control surface does not have fault and dynamic environment changes, the individuals
The direction of evolution is
When partial control surface faults occur, if the environment is not changed, the optimal solution set which is a part of the new initial population is
If the dynamic environment changes, the evolution track is
When a control surface with faults is newly added, an optimal solution set containing fault information of the first two moments is used as a part of a new initial population, namely
7. The fault-tolerant method for the control surface of the flying-wing unmanned aerial vehicle based on the improved dynamic MOPSO is characterized in that according to the requirements of control distribution and fault tolerance, when the deflection angle of a resistance rudder is small, a linear rudder effect inverse model is adopted to provide the change direction of a solution set so as to reduce the failure probability of the control surface, and an auxiliary evolution track of a prediction solution is obtained; the objective function may be at point δ t Near linearization as Δ v t ＝BΔδ t Wherein δ t ＝[δ l1 t ，δ l2 t ，δ l3 t ，δ l4 t ，δ r1 t ，δ r2 t ，δ r3 t ，δ r4 t ] T For the optimal allocation result at time t, B is at δ t Control distribution efficiency matrix after nearby linearization:
Δv t ＝[ΔC tδ ，ΔC mδ ，ΔC nδ ] T the torque coefficient increment is controlled for the three channels to be allocated,
to control the moment coefficient C i About control surface delta j Partial differential of (a); according to the deflection relation of the control surface in the fault state, the control moment coefficient increment is converted into:
when the fault state of the control surface at the time t is consistent with the time t-1, the least square solution delta of the formula is adopted t The evolution track as the prediction optimal solution guides the population to evolve near the optimal solution after the environment changes, namely delta t ＝Γ -1 (I-Λ) -1 (B T B) -1 B T (v t -v t-1 ) Wherein Γ = Γ t ＝Γ t-1 ，Λ＝Λ t ＝Λ t-1
8. The fault-tolerant method for control surfaces of flying-wing drones based on the improved dynamic MOPSO as claimed in claim 4, characterized in that, according to the time sequence prediction model and the linearized inverse model, when the environment change is detected, the possible individuals in the jth cluster of the new prediction population
Can be selected from corresponding individuals
Is specifically represented by
Wherein
In order to predict the set of evolution directions,
to predict the set of linear inverse models, i =1,2, …, N P ，j＝1，2，…，K，ξ 1 And xi 2 Is a weight of the prediction model and satisfies ξ 1 +ξ 1 =1, gaussian disturbance epsilon t -(0，σ t ) Is based on the mean value of 0 and the variance σ t For increasing the diversity of the population, sigma t Satisfy the requirement of
Since the cross-coupling nonlinearity caused by the resistance rudder deflection affects the prediction accuracy, the weight ξ 2 Related to the deflection of the resistance rudder at time t, in particular
N may be generated from a predictive model according to the above process P New possible individuals, the rest of the initial population N-N P Individuals are randomly generated in a decision space;
the particle velocity and position are updated according to the following expression:
in the formula, v i Denotes the particle velocity, x i As the position of the current particle, p ibest Represents the optimal position of the ith particle, g ibest Representing the global optimum position of the ith particle, c 1 ，c 2 Is a learning factor, r 1 ，r 2 Has a range of [0,1]A random number in between. And updating the individual optimal solution sets according to a Pareto domination relation when the particles in the two generations of individual optimal solution sets before and after updating meet the convergence index, performing variation operation on the particles when the convergence index is not met, sorting the updated particles according to the concentration, screening the individual optimal solution sets and the global optimal solution sets, and selecting one global optimal solution as a control plane deflection instruction for controlling redistribution of the fault control surface.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211306318.3A CN115657479A (en) | 2022-10-24 | 2022-10-24 | Improved dynamic MOPSO-based control surface fault tolerance method for flying-wing unmanned aerial vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211306318.3A CN115657479A (en) | 2022-10-24 | 2022-10-24 | Improved dynamic MOPSO-based control surface fault tolerance method for flying-wing unmanned aerial vehicle |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115657479A true CN115657479A (en) | 2023-01-31 |
Family
ID=84990899
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211306318.3A Pending CN115657479A (en) | 2022-10-24 | 2022-10-24 | Improved dynamic MOPSO-based control surface fault tolerance method for flying-wing unmanned aerial vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115657479A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116185057A (en) * | 2023-03-24 | 2023-05-30 | 西北工业大学 | Attitude fault-tolerant control method for wing body fusion flying unmanned aerial vehicle |
-
2022
- 2022-10-24 CN CN202211306318.3A patent/CN115657479A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116185057A (en) * | 2023-03-24 | 2023-05-30 | 西北工业大学 | Attitude fault-tolerant control method for wing body fusion flying unmanned aerial vehicle |
| CN116185057B (en) * | 2023-03-24 | 2023-09-01 | 西北工业大学 | Attitude fault-tolerant control method for wing body fusion flying unmanned aerial vehicle |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107065539B (en) | Control surface fault self-adaptive fault-tolerant control method of flying wing layout aircraft | |
| CN108628167B (en) | Fault-tolerant control method of flight control system based on adaptive distribution technology | |
| CN110703787A (en) | Aircraft redundancy control method based on mixed multi-target PSO algorithm of preference matrix | |
| CN115657479A (en) | Improved dynamic MOPSO-based control surface fault tolerance method for flying-wing unmanned aerial vehicle | |
| CN110262448A (en) | Hypersonic aircraft elevator failure tolerant control method with state limit | |
| Ma et al. | Deep auto-encoder observer multiple-model fast aircraft actuator fault diagnosis algorithm | |
| CN112327926A (en) | Self-adaptive sliding mode control method for unmanned aerial vehicle formation | |
| CN112068594B (en) | JAYA algorithm optimization-based course control method for small unmanned helicopter | |
| Huang et al. | Coordinated disturbance observer-based flight control of fixed-wing UAV | |
| CN116185057B (en) | Attitude fault-tolerant control method for wing body fusion flying unmanned aerial vehicle | |
| CN115659502A (en) | Control surface fault diagnosis method of flying wing unmanned aerial vehicle based on neural network adaptive observer | |
| Shuang et al. | An anti-windup fault tolerant control scheme with guaranteed transient performance for tailless flying wing aircrafts | |
| CN116088556A (en) | Intelligent fault-tolerant control method for aircraft based on deep reinforcement learning | |
| CN116859728A (en) | Ship dynamic positioning system thrust distribution method based on improved multi-target particle swarm algorithm | |
| Zhang et al. | Multi-target encirclement with collision avoidance via deep reinforcement learning using relational graphs | |
| CN114035543B (en) | Self-repairing control method under damaged state of airplane | |
| Kamalasadan | A new generation of adaptive control: An intelligent supervisory loop approach | |
| CN112698666B (en) | Aircraft route optimization method based on meteorological grid | |
| CN112464557A (en) | Improved hybrid multi-target PSO-based redundant control surface control method for flying wing unmanned aerial vehicle | |
| Li et al. | Multi-effectors distribution of flying wing with stealthy optimization | |
| CN112487552A (en) | Envelope dividing and gain scheduling method of flying wing unmanned aerial vehicle based on fuzzy clustering | |
| Yu et al. | Control of Fixed-wing UAV Using Optimized PID Controller with the Adaptive Genetic Algorithm | |
| Zhang et al. | Research of fault diagnosis in flight control system based on fuzzy neural network | |
| Li et al. | Incremental dual heuristic dynamic programming based hybrid approach for multi-channel control of unstable tailless aircraft | |
| Xu et al. | An innovative control allocation framework for a novel tiltrotor aircraft |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |