Flexible spacecraft layout optimization method based on multi-target hybrid particle swarm
By optimizing the sensor installation position using a multi-objective hybrid particle swarm optimization algorithm, the problems of inaccurate modal coordinate observation of large flexible spacecraft and local optima of traditional algorithms were solved, thus achieving accurate observation and control of modal coordinates.
Patent Information
- Application Number
- CN202410990166.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-08
AI Technical Summary
Existing methods cannot effectively observe and control the modal coordinates of highly flexible spacecraft, resulting in slow divergence or convergence of modal coordinates. Furthermore, traditional optimization algorithms are prone to getting trapped in local optima, making it impossible to accurately control the spacecraft.
A flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm optimization is adopted. A model is established through finite element analysis, and the optimization criteria are set as the observability optimal criterion under modal effective mass weighting and the condition number of modal filter matrix under sensor number truncation. The hybrid particle swarm optimization algorithm is used to optimize the sensor installation position.
It improves the accuracy of modal coordinate observation, avoids modal overflow, ensures the global optimization of sensor layout, and is suitable for the control and application of flexible spacecraft of various sizes.
Smart Images

Figure CN118917192A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a flexible spacecraft layout optimization method and belongs to the field of structural vibration observation. Background Art
[0002] With the advancement and development of science and technology, the missions of spacecraft have become increasingly complex and large-scale. In the field of aerospace, the role of structures such as solar panels and satellite antennas is becoming increasingly important. When such large flexible structures are in orbit, they will be affected by various factors. For example, during the process of spacecraft orbit change, maneuvering, and on-orbit service, the low-frequency characteristics brought by the large scale of large flexible accessories will cause continuous vibration of the flexible structure, especially when the vibration frequency is close to the natural frequency of the structure. The effect will be more obvious; secondly, some actuators, such as control torque gyroscopes and flywheels, will cause vibration of the flexible structure due to the coupling characteristics brought by their high-frequency vibrations during excitation and control. Therefore, in order to avoid this situation, passive damping or active control is usually used for vibration suppression. Among them, active vibration control is mostly used, and sensors are often used to measure the vibration of the flexible structure and feedback it to the controller.
[0003] The distribution of flexible structure sensors plays a vital role in the observation of modal coordinates. Since the actual modal coordinates of the physical displacement measured by the sensor are infinite order, and the number of sensors is limited, the coordinates converted to the modal space will inevitably overflow the modal observation. Due to the complexity and large size of the structure, the sensor installation position matrix is a sparse matrix. When the installation position is inappropriate, the modal coordinates obtained by the modal filter often cannot converge or converge slowly, which in turn affects the vibration calculation and control of the entire surface; and too many sensors will also bring about unfavorable factors such as complex wiring and increased maintenance costs. In view of the above reasons, it is necessary to optimally layout the sensor positions according to the structural characteristics of large flexible bodies.
[0004] Currently, there are many optimization criteria for device layout, such as observability criterion, energy criterion, modal independence criterion, etc. However, due to the sparse characteristics of the installation matrix of large spacecraft, it is also necessary to consider the accuracy of the calculated modal coordinates. Modal overflow and ill-conditioned installation matrices may lead to inaccurate modal coordinates. Therefore, it is necessary to consider the layout under multiple factors at the same time; secondly, due to the complex structure, high dimension, and multiple constraints of large-scale flexible spacecraft, traditional improved heuristic algorithms are easy to fall into local optimality, relying on the experience of algorithm developers to debug parameters and conduct multiple experiments to obtain the optimal solution. Therefore, it is necessary to consider the global and local convergence of the algorithm at the same time. Summary of the invention
[0005] The purpose of the present invention is to solve the problem that the modal coordinates of the spacecraft flexible structure will diverge or converge slowly, resulting in the inability of existing methods to correctly observe the modal coordinates of the large flexible structure of the spacecraft and to accurately control the spacecraft, and to propose a flexible spacecraft layout optimization method based on a multi-objective hybrid particle swarm.
[0006] A flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm is as follows:
[0007] Step 1: Use finite element analysis software to establish a flexible attachment model under constrained mode, perform modal analysis on the flexible attachment model under constrained mode, and obtain the modal natural frequency, modal damping, vibration mode matrix and effective mass of each order mode;
[0008] Step 2: Based on step 1, a flexible structure dynamic model described by modal coordinates when using a displacement sensor is established;
[0009] Step 3: Based on step 2, set optimization criterion 1 as the optimal criterion for observability under the weighting of modal effective mass;
[0010] Set optimization criterion 2 to the number of sensors n sensor Modal filter matrix under truncation The two-norm condition number of the truncated modal filter matrix C dis Φ n Take the first n sensor List;
[0011] Step 4: Take the optimization criterion as the multi-objective fitness function, optimize the multi-objective fitness function, and obtain the sensor installation freedom when the fitness function takes the minimum case.
[0012] The beneficial effects of the present invention are:
[0013] In view of the fact that traditional optimization indicators cannot correctly observe modal displacement, the present invention improves the indicators and introduces the modal filter matrix condition number as one of the optimization indicators to ensure that the actual physical displacement can be effectively converted to the modal space, especially when the observation matrix of a large-scale flexible spacecraft is sparse; while considering the optimal sensor observation performance, the present invention introduces the modal effective mass to reduce the impact of modal overflow; it solves the problem that the modal coordinates of the flexible structure of the spacecraft diverge or converge slowly, causing the existing method to be unable to correctly observe the modal coordinates of the flexible structure of the spacecraft and to accurately control the spacecraft;
[0014] There are a large number of local optimal solutions for large-scale flexible spacecraft. Traditional intelligent optimization algorithms will fall into local optimal solutions and have low efficiency. The present invention uses a hybrid particle swarm algorithm to effectively avoid the emergence of local optimal solutions for the complex situation of multi-peak distribution of large-scale flexible spacecraft. It solves the problem that the modal coordinates of the flexible structure of the spacecraft diverge or converge slowly, causing the existing methods to be unable to correctly observe the modal coordinates of the flexible structure of the spacecraft and to accurately control the spacecraft.
[0015] The present invention has good applicability and can be applied to the layout of flexible spacecraft of various sizes, and has certain application value in terms of control and application in the modal space of actual structures. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a flow chart of the present invention;
[0017] Figure 2 A physical structure diagram of a large flexible spacecraft targeted by the present invention;
[0018] Figure 3 This is the fitness curve of indicator 1 when the number of sensors is 5;
[0019] Figure 4 This is the fitness curve of indicator 2 when the number of sensors is 5;
[0020] Figure 5 This is the result diagram of the optimized sensor installation position;
[0021] Figure 6 It is the location diagram of the excitation points where the excitation is applied during simulation;
[0022] Figure 7 When the number of sensors is 5, the first 5 theoretical modal coordinates and the modal coordinate diagram measured by the sensors;
[0023] Figure 8 The first five theoretical modal coordinates and the modal coordinate error diagram measured by the sensor;
[0024] Fig. 9 This is a diagram of the modal overflow of each sensor detection point. DETAILED DESCRIPTION
[0025] Specific implementation method 1: This implementation method is based on a multi-objective hybrid particle swarm to optimize the layout of a flexible spacecraft. The specific process is as follows:
[0026] Step 1: Use finite element analysis software to establish a flexible attachment model under constrained mode, perform modal analysis on the flexible attachment model under constrained mode, and obtain the modal natural frequency, modal damping, vibration mode matrix and effective mass of each order mode;
[0027] Step 2: Based on step 1, a flexible structure dynamic model described by modal coordinates when using a displacement sensor is established;
[0028] Step 3: Based on step 2, set optimization criterion 1 as the optimal criterion for observability under the weighting of modal effective mass;
[0029] Set optimization criterion 2 to the number of sensors n sensor Cut off The two-norm condition number of ;
[0030] Step 4: Take the optimization criterion as the multi-objective fitness function, optimize the multi-objective fitness function, and obtain the sensor installation freedom when the fitness function takes the minimum case.
[0031] Specific implementation method 2: This implementation method is different from the specific implementation method 1 in that: in step 1, a finite element analysis software is used to establish a flexible attachment model under a constrained mode, and a modal analysis is performed on the flexible attachment model under a constrained mode to obtain the modal natural frequency, modal damping, vibration mode matrix and the effective mass of each order mode; the specific process is:
[0032] Use Ansys finite element analysis software to build a physical model of the flexible beam structure. According to the required working conditions, establish constraints on the physical model of the flexible beam structure, and perform modal analysis (modal analysis is performed using Ansys finite element analysis software) to obtain the vibration characteristics of the constructed physical model of the flexible beam structure. The vibration characteristics of the physical model of the flexible beam structure include modal natural frequency, modal damping, vibration mode matrix and effective mass of each order mode.
[0033] The other steps and parameters are the same as those in the first embodiment.
[0034] Specific implementation method three: This implementation method is different from specific implementation methods one or two in that: in step 2, based on step 1, a flexible structure dynamics model described by modal coordinates when using a displacement sensor is established; the expression is:
[0035]
[0036] in
[0037] C and K are the damping matrix and stiffness matrix respectively, All are diagonal matrices; Φ n is the vibration mode matrix; the superscript T means transposition; B d , B c are the interference force and control force action matrices respectively; f d is the interference force, f c For control; C disis the displacement sensor installation matrix, the number of rows of the displacement sensor installation matrix is the number of sensors, and the number of columns is the total number of degrees of freedom n;
[0038] Assume that a sensor is installed in the z-axis direction of node 1, C dis The representation is:
[0039]
[0040] W o are the modal coordinates, is the modal velocity, is the modal acceleration; Q is the output value of the displacement sensor.
[0041] The other steps and parameters are the same as those in the first or second embodiment.
[0042] Specific implementation method 4: This implementation method is different from any one of specific implementation methods 1 to 3 in that: in step 3, the optimization criterion 1 is set based on step 2 as the optimal criterion of observability under the weighting of modal effective mass (the purpose is to ensure that the energy obtained by the sensor is the highest, while considering the influence of each order mode on the system dynamics);
[0043] Set optimization criterion 2 to the number of sensors n sensor Cut off The two-norm condition number of ;
[0044] The specific process is:
[0045] Optimization criterion 1 is:
[0046]
[0047] Optimization criterion 2 is:
[0048]
[0049] Among them, W o represents optimization index 1, Cirt_2 represents optimization index 2; n sensor Represents the order observable by the displacement sensor, which is the same as the number of displacement sensors; j represents the order index observable by the displacement sensor, 1≤j≤n sensor ;M eff,j represents the modal effective mass, expressed as represents the degree of participation of each mode in the system dynamics, M is the mass matrix, r is the unit excitation vector, Φ j is the jth column of the vibration mode matrix; j represents the observable Gramian matrix W o The displacement sensor can observe the eigenvalues of the modal sub-block; nn sensorrepresents the unobservable order of the displacement sensor, where n is the total number of degrees of freedom of the node, satisfying The total number of nodes established for finite element analysis (the number of nodes in the finite element modeling setting in Ansys); j′ represents the unobservable order index of the displacement sensor, n sensor +1≤j′≤nn sensor ; j′ represents the observable Gramian matrix W o Eigenvalue of the unobservable mode sub-block of the displacement sensor; M eff,j′ represents the modal effective mass, expressed as represents the degree of participation of each mode in the system dynamics, M is the mass matrix, r is the unit excitation vector, Φ j′ is the j′th column of the vibration mode matrix; σ() represents the standard deviation; · represents the multiplication sign; ||||2 represents the calculation of the matrix bi-norm; Indicates the first n sensor Modal coordinate η nsensor and the sensor measures the physical coordinate δ nsensor The transformation matrix of
[0050] Indicator 2 aims to In sparse cases, it is ensured that the actual physical coordinates measured by the sensor can be optimally converted to the modal coordinates in the modal space.
[0051] The other steps and parameters are the same as those in Specific Embodiments 1 to 3.
[0052] Specific implementation method 5: This implementation method is different from any one of the specific implementation methods 1 to 4 in that: in step 4, the optimization criterion is used as a multi-objective fitness function, and the multi-objective fitness function is optimized to obtain the group optimal solution G when the fitness function takes the minimum case k (Step 4.5 determines the final group optimal solution G k );
[0053] The group optimal solution G k Refers to the installation freedom of the sensor;
[0054] Each iteration will generate a group optimal solution, which will be updated in the iteration to take a smaller value;
[0055] The specific process is:
[0056] Step 4.1: Initialize the number of particles particlesize, position x, velocity v, and the ratio of elite particles to ordinary particles λ r , crossover selection probability λ c , mutation selection probability λ m, inertia weight ω, acceleration coefficients c1, c2, criterion weight ratio weight, particle position constraints, particle velocity constraints; set the maximum number of iterations Iter;
[0057] Step 4.2: Set the number of iterations k = 1;
[0058] Step 4.3: Set up the multi-objective fitness function, as well as the elite subgroup and the ordinary subgroup;
[0059] The specific process is:
[0060] Step 4.3.1: Take optimization criteria 1 and optimization criteria 2 as multi-objective fitness functions, where optimization criteria 1 is negative, and calculate the values of optimization index 1 and optimization index 2;
[0061]
[0062] Solve the actuator installation position layout when the fitness function Cirt_1 is maximum and the fitness function Cirt_2 is minimum. Cirt_1 is negative, so it is converted into a multi-objective optimization problem in which both criteria are minimized.
[0063] Step 4.3.2: Normalize the values of optimization index 1 and optimization index 2 to obtain the normalized values of optimization index 1 and optimization index 2;
[0064] Step 4.3.3: Set the weight ratio of optimization index 1 and optimization index 2, weight the normalized values of optimization index 1 and optimization index 2, and obtain the particle segmentation standard; the expression is:
[0065] weighted fitness =weighted×[norm cirt_1 ,norm cirt_2 ] T
[0066] Among them, weighted represents the weight ratio; norm cirt_1 is the normalized optimization index 1; norm cirt_2 is the normalized optimization index 2; the superscript T indicates transposition;
[0067] The vector obtained by weighting the particle fitness index is only used for particle sorting here and is not used for subsequent individual optimal updates. r To split the population, is a real number;
[0068] Step 4.3.4: Weighted fitness Vector descending order (after descending order, use λr Particle segmentation);
[0069] Step 4.3.5: According to the population proportion λ r (For example, 50%) The particles in descending order are divided into the elite subgroup, and the number of particles in the elite subgroup is N e (For example, there are 10 particles, particles 1 to 5 are elite subgroups), and the latter part is the ordinary subgroup, and the number of particles in the ordinary subgroup is N n (For example, there are 10 particles, particles 6 to 10 are common subgroups), satisfying N e +N n =particlesize;
[0070] The elite subgroup is responsible for guiding the learning process of subsequent ordinary particles and according to the crossover probability λ c Learn the information of other particles to increase the global search capability; the ordinary subgroup follows the mutation probability λ m Randomize its particle learning process to improve local search capabilities;
[0071] Step 4.4: Calculate the speed and individual position of all particles in the elite subgroup; the specific process is:
[0072] Step 4.4.1: Let i = 1, calculate the d-dimensional individual position of the i-th particle at the k-th iteration, i = 1, 2, ..., N e , N e is the total number of particles in the elite subgroup;
[0073] The expression is:
[0074]
[0075] The subscript k is the kth iteration, the subscript k-1 is the k-1th iteration, i represents the order of particles in the elite subgroup, d represents the particle dimension, k is the time, r1 and r2 are two random variables uniformly distributed in the interval [0,1]; Represents the optimal historical solution P of the i-th particle individual updated in the k-1th iteration i,k-1 The dth dimension of , that is, the x corresponding to the small Cirt_1 and Cirt_2 in the history iteration of the i-th particle i,k-1 ; The optimal solution P of the individual history of the i-th particle i,k-1 is the x corresponding to which Cirt_1 and Cirt_2 are both small in the history iteration of the i-th particle i,k-1 ; Represents the optimal historical solution P of the mth particle individual updated in the k-1th iteration m,k-1The dth dimension of (the initial value is a random installation position within the installable range), that is, the x corresponding to Cirt_1 and Cirt_2 are small in each particle history iteration process i,k-1 , m is a random particle in the elite subgroup, m≠i, m=1,2,…,N e ; is the individual historical optimal solution P of the i-th particle updated in the k-1th iteration i,k-1 The d1th dimension of (the initial value is a random installation position within the installable range), that is, the x corresponding to Cirt_1 and Cirt_2 are small in each particle history iteration process i,k-1 , and d1≠d; rand represents any value between (0,1), thereby ensuring that the particle update has a certain degree of randomness; represents the particle velocity v of the ith particle after update in the k-1th iteration i,k-1 The dth dimension of represents the position x of the i-th particle after the update in the k-1th iteration i,k-1 The dth dimension of Represents the particle velocity v of the i-th particle after update in the k-th iteration i,k The dth dimension of represents the individual position x of the i-th particle after update in the k-th iteration i,k The dth dimension of c represents the crossover probability;
[0076] is the updated optimal solution G of the group in the k-1th iteration k-1 The dth dimension of (the initial value is a random installation position within the installable range);
[0077] Step 4.4.2: Let i = i + 1, and repeat step 4.4.1 until all particles in the elite subgroup are calculated, and then execute step 4.5;
[0078] Step 4.5: Calculate the speed and individual position of all particles in the common subgroup; the specific process is:
[0079] Step 4.5.1: Let i′ = 1, calculate the d-dimensional individual position of the i′th particle at the kth iteration, i′ = 1, 2, …, N n , N n is the total number of particles in the common subgroup; the expression is:
[0080]
[0081] in is the particle velocity v of the next i′th particle after the kth iteration update i′,k The dth dimension of is the particle velocity v of the i′th particle after update in the k-1th iteration i′,k-1 The dth dimension of ; r1 and r2 are random variables uniformly distributed in the interval [0,1]; Represents the optimal solution P of the individual history of the lth particle after the update in the k-1th iteration l,k-1 The dth dimension of (the initial value is a random installation position within the installable range), that is, the x corresponding to Cirt_1 and Cirt_2 are small in each particle history iteration process i,k-1 , l is the particle number randomly selected from [1,i′-1], l≠i′, l=1,2,…,N n ; is the updated optimal solution G of the group in the k-1th iteration k-1 The dth dimension of (the initial value is a random installation position within the installable range); represents the optimal solution P of the individual history of the i′th particle in the k-1th iteration i′,k-1 The dth dimension of (the initial value is a random installation position within the installable range); represents the individual position x of the i′th particle at the k-1th iteration i′,k-1 The dth dimension of represents the individual position x of the i′th particle at the kth iteration i′,k The dth dimension of represents the particle velocity v of the i′th particle after the update in the kth iteration i′,k The dth dimension of represents the particle velocity v of the i′th particle after the update in the k-1th iteration i′,k-1 The dth dimension of m represents the mutation probability;
[0082] When the Ansys model is established, the vertices and connecting rods of the established structure are set as the locations where the sensors can be installed. Without loss of generality, in the iteration of the particle swarm algorithm, the variables to be optimized x of all particles i,k 、Individual historical optimal solution P i,k , group optimal solution G k The initial value x i,0 , P i,0 , G0 is taken as a random installation position within the installable range. For particles outside the range that appear during the iteration process, they are excluded and randomly moved to an installable position.
[0083] Step 4.5.2: Let i′=i′+1, and repeat step 4.5.1 until all particles in the common subgroup are calculated, and then execute step 4.6;
[0084] Step 4.6: Calculate all elite particles (number is N e ) and ordinary particles (number is Nn )'s fitness value (in elite particles (number is N e ) and ordinary particles (number is N n ) is fully updated, calculate all elite particles (number is N e ) and ordinary particles (number is N n ) fitness index), and for all particles (elite particles (number is N e ) and ordinary particles (number is N n )) The individual historical optimal solution P at the kth moment i,k and the group optimal solution G k Make updates;
[0085] Step 4.7: Set the number of iterations k = k + 1, and re-execute steps 4.3 to 4.7 until k = Iter, and obtain the final group optimal solution G k , which is the optimal installation position for iteration.
[0086] Sensors can be installed at the vertices and connecting rod positions of the truss structure established by Ansys, thereby imposing certain constraints on the installation position of the structure.
[0087] In the iterative optimization process using the heuristic algorithm, when weighting and sorting the particle fitness, the particle fitness parts are normalized to ensure the consistency of the dimension; at the same time, the sorting index is applied globally to ensure that the speed and position are for the same particle.
[0088] The other steps and parameters are the same as those in Specific Implementation 1 to 4-1.
[0089] Specific implementation method 6: This implementation method is different from any one of specific implementation methods 1 to 5 in that: in step 4.3.1, optimization criteria 1 and optimization criteria 2 are used as multi-objective fitness functions, index 1 is negative, and the values of optimization index 1 and optimization index 2 are calculated;
[0090]
[0091] The specific process is:
[0092] 1) Displacement sensor installation matrix C dis The number of rows is the number of sensors, and the number of columns is the total number of degrees of freedom n;
[0093] 2) The value of represents the displacement sensor installation matrix C dis Middle The column value is 1;
[0094] Indicates the kth iteration The individual positions of the particles; The particle is the sum of elite particles and ordinary particles. Particles;
[0095] 3) Replace C in 2) dis Substitute the formula y=C dis Φ n T dis z=Cz, calculate C matrix;
[0096] Where y is the sensor output, z is the state vector, η1, represents the first-order modal displacement and modal velocity, η2, represents the second-order modal displacement and modal velocity, η n , represents the nth-order modal displacement and modal velocity, and the superscript T is the transposition; Φ n is the vibration mode matrix; T dis For even columns all zero, odd columns are unit column vectors, we have:
[0097]
[0098] 4) Input the A matrix and C matrix into the Gram matrix function, and the Gram matrix function outputs the block matrix W o , the block matrix W o First n sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j ; after nn sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j′ ;
[0099] Among them A i Because of A α The diagonal array is composed of
[0100] ξ α is the modal damping of the αth order mode, 1≤α≤n; ω α is the modal natural frequency of the αth order mode, 1≤α≤n;
[0101] 5) j , j′ Substitute optimization index 1 and optimization index 2 to obtain the values of optimization index 1 and optimization index 2.
[0102] The other steps and parameters are the same as those in Specific Implementation Methods 1 to 5-1.
[0103] Specific implementation method 7: This implementation method is different from the specific implementation methods 1 to 6 in that: in step 4.6, all elite particles (number N) are calculated. e) and ordinary particles (number is N n )'s fitness value (in elite particles (number is N e ) and ordinary particles (number is N n ) is fully updated, calculate all elite particles (number is N e ) and ordinary particles (number is N n ) fitness index), and for all particles (elite particles (number is N e ) and ordinary particles (number is N n )) The individual historical optimal solution P at the kth moment i,k and the group optimal solution G k Make updates;
[0104] The specific process is:
[0105] Step 4.6.1: Calculate all elite particles (number is N e ) and ordinary particles (number is N n )’s fitness value; the specific process is:
[0106] 1) Displacement sensor installation matrix C dis The number of rows is the number of sensors, and the number of columns is the total number of degrees of freedom n;
[0107] 2) The value of represents the displacement sensor installation matrix C dis Middle The column value is 1;
[0108] Indicates the kth iteration The individual positions of the particles; The particle is the sum of the elite particles and ordinary particles. Particles;
[0109] 3) Replace C in 2) dis Substitute the formula y=C dis Φ n T dis z=Cz, calculate C matrix;
[0110] Where y is the sensor output; z is the state vector, η1, represents the first-order modal displacement and modal velocity, η2, represents the second-order modal displacement and modal velocity, η n , represents the nth-order modal displacement and modal velocity, and the superscript T is the transposition; Φ n is the vibration mode matrix; T dis For even columns all zero, odd columns are unit column vectors, we have:
[0111]
[0112] 4) Input the A matrix and C matrix into the Gram matrix function, and the Gram matrix function outputs the block matrix W o , the block matrix W o First n sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j ; after nn sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j′ ;
[0113] Among them A i Because of A α The diagonal array is composed of
[0114] ξ α is the modal damping of the αth order mode, 1≤α≤n; ω α is the modal natural frequency of the αth order mode, 1≤α≤n;
[0115] 5) j , j′ Substitute optimization index 1 and optimization index 2 to obtain the values of optimization index 1 and optimization index 2;
[0116] Calculate the matrix C, and then calculate the observable Gram matrix W of the system through the coefficients (A, B, C, D) in the state space equation o :
[0117] When the system is stable, there is A T W o +W o A=-C T C. When there is no control output, the flexible structure is stable, and the Gram matrix W of the system can be calculated from this o ;
[0118] Since the matrix A is composed of A α The block matrix composed of o It is also a block matrix, and the eigenvalue of each sub-block is the λ in the index i , W o Refers to a certain time T f Gram matrices as they go to infinity;
[0119] y=Q is the sensor output, A is the i The diagonal array is composed of ξ i ,ω i is the modal damping and modal frequency of the i-th mode;
[0120] Step 4.6.2: When the particle k When the values of optimization index 1 and optimization index 2 are both less than the values of optimization index 1 and optimization index 2 at time k-1, particle Individual historical optimal solution Update to time k Otherwise the particle Individual historical optimal solution Keep it as k-1 moment The value of
[0121] Indicates the kth iteration The individual positions of the particles
[0122] No. Particles are elite particles (the number is N e ) and ordinary particles (number is N m ) Particles;
[0123] Step 4.6.3: When the particle k Individual historical optimal solution The values of optimization index 1 and optimization index 2 are both less than the optimal solution G of the group at k-1 time k-1 When the values of optimization index 1 and optimization index 2 of the dth dimension (the initial value is a random installation position within the installable range), the optimal solution G at time k k Update to k-time particle Individual historical optimal solution Otherwise, the optimal group solution G at time k k Keep the optimal solution G of the group at k-1 times k-1 .
[0124] The other steps and parameters are the same as those in Specific Implementations 1 to 6-1.
[0125] The following examples are used to verify the beneficial effects of the present invention:
[0126] Embodiment 1: Below, in combination with the accompanying drawings and specific implementation methods, the large-scale flexible spacecraft layout optimization method based on hybrid particle swarm and optimal modal observation provided by the present invention is described in detail.
[0127] Step 1. Use Ansys finite element analysis software to build a physical model of the flexible beam structure. According to the required working conditions, establish constraints on the physical model of the flexible beam structure, and perform modal analysis (using Ansys finite element analysis software to perform modal analysis) to obtain the vibration characteristics of the constructed physical model of the flexible beam structure. The vibration characteristics of the physical model of the flexible beam structure include relevant parameters such as natural frequency, modal damping, vibration mode matrix, and effective mass of each order mode.
[0128] Step 2: For the dynamics of flexible structures, the physical displacement of the nodes is is the number of node degrees of freedom, and the flexible structure dynamic equation (1) is established
[0129]
[0130] Where M, C, and K are the mass matrix, damping matrix, and stiffness matrix respectively; B d , B c are the interference force and control force action matrices respectively; Q is the sensor output value, C dis Installing matrix for displacement sensors;
[0131] According to the conversion relationship between the node physical displacement δ and the modal coordinates δ = Φ n η, obtain the flexible structure dynamic equation under modal coordinate expression (2)
[0132]
[0133] in ξ n ,ω n is the modal damping and modal frequency of the nth order mode; C and K are the damping matrix and stiffness matrix respectively, All are diagonal matrices; Φ n is the vibration mode matrix; the superscript T means transpose; I is a diagonal matrix with 1 on the diagonal; B d , B c are the interference force and control force action matrices respectively; f d is the interference force, f c For control; C dis is the displacement sensor installation matrix, the number of rows of the displacement sensor installation matrix is the number of sensors, and the number of columns is the total number of degrees of freedom of the node n;
[0134] Assume that a sensor is installed in the z-axis direction of node 1, C dis The representation is:
[0135]
[0136] η is the modal coordinate, is the modal velocity, is the modal acceleration; Q is the output value of the displacement sensor.
[0137] Step 3: In order to obtain the optimal sensor installation position, it is necessary to select a suitable optimization performance index and use the index as the fitness function of the heuristic optimization algorithm;
[0138] Ideally, the number of nodes is infinite, which leads to infinite modal displacement η. In finite element analysis, the number of node degrees of freedom n is often large, and the high-order modes have less influence and decay faster, so the first n are usually considered. o The first n order modes are observed, and the number of node degrees of freedom n is truncated, thereby reducing the amount of calculation in the optimization algorithm. That is, in actual use, the first n of the total number of node degrees of freedom n are truncated. o As the total number of degrees of freedom of the nodes in actual use; then, formula (2) is modified as follows:
[0139]
[0140] make η1, represents the first-order modal displacement and modal velocity, η2, represents the second-order modal displacement and modal velocity, Represents the nth o The first-order modal displacement and modal velocity, the superscript T is the transpose, and z is the state variable; is the vibration mode matrix after mode truncation;
[0141] Transform formula (3) into formula (4):
[0142]
[0143] in, is the state variable derivative, y=Q is the sensor output, and A is the value obtained by A i The diagonal array is composed of ξ α ,ω α is the modal damping and modal frequency of the αth order mode, 1≤α≤n o ;
[0144] B is a matrix with all odd-numbered rows being 0 and even-numbered rows being unit rows.
[0145] Here n o This corresponds to the total number of degrees of freedom n of the nodes above. n is used in the derivation process; however, in actual calculations, n is infinite or very large. To facilitate calculations, n will be truncated and the number of truncations is set to n. o (n o=10); that is, in the actual simulation experiment, the infinite value or a large value n is regarded as a truncation value n o ;
[0146] C is an intermediate variable, representing the conversion matrix from state variable z to output y, which is:
[0147]
[0148] Where T dis The even columns are all 0, and the odd columns are unit column vectors, so:
[0149]
[0150] D is 0,
[0151] In order to maximize the energy obtained by the sensor, from the initial state to a certain time T f , define the signal energy J received by the sensor:
[0152]
[0153] Among them, z0 is the initial state, W o (T f ) is T f The Gram matrix of the output equation of system (4) at time τ, y(τ) is the sensor output at time τ;
[0154] According to formula (7), we have:
[0155]
[0156] Thus:
[0157]
[0158] Among them, W o Refers to a certain time T f Gram matrices as they go to infinity;
[0159] From equations (7) and (9), it can be concluded that the energy received by the sensor is only proportional to the Gram matrix of the system. Equation (10) is used to measure the signal energy J received by the sensor. When the system is unobservable, the value is 0; the better the observation performance, the higher the value.
[0160]
[0161] Among them, the matrix trace (W o ) as energy output, represents the geometric mean of the eigenvalues, σ(λ i) is used to penalize higher or lower eigenvalues to prevent poorer eigenvalues from being hidden by better eigenvalues;
[0162] Since each mode contributes differently to energy, we use equation (10) to convert the modal effective mass M eff As the weight, the eigenvalue corresponding to each mode is weighted, and formula (10) is modified to formula (11):
[0163]
[0164] In order to make the sensor observed modal effect better and reduce the impact of the unobserved mode, the high-order unobserved mode is used as the subtrahend, and the sensor observed mode minus the high-order unobserved mode is used as the updated index. sensor is the number of sensors, n o -n sensor is the number of unobserved modes, and the index update is shown in formula (12):
[0165]
[0166] For the second indicator, consider n sensor sensors, the sensor modal cutoff order is n sensor , for the calculation of modal coordinates, since When the corresponding number of sensor data is obtained, The modal coordinates are obtained with the same number of sensors. However, when the flexible spacecraft structure is large, the number of nodes is large, and C dis The matrix is sparse, resulting in It is often an ill-conditioned matrix, that is, the actual coordinates Small errors can lead to large changes in the modal coordinate values.
[0167] Therefore, in order to ensure accurate calculation of the modal coordinates, it is necessary to ensure that the sensor obtains as much energy as possible. The matrix has good characteristics. Let the condition number of the matrix be the optimization index 2, and we have:
[0168]
[0169] Where Cirt_1 represents optimization index 1, and Cirt_2 represents optimization index 2; n sensor Represents the order observable by the displacement sensor, which is the same as the number of displacement sensors; j represents the order index observable by the displacement sensor, 1≤j≤n sensor ;M eff,j represents the modal effective mass, expressed as represents the degree of participation of each mode in the system dynamics, M is the mass matrix, r is the unit excitation vector, Φ j is the jth column of the vibration mode matrix; j Denotes the observable Gramian matrix W o The displacement sensor can observe the eigenvalues of the modal sub-block; n o -n sensor represents the unobservable order of the displacement sensor; j′ represents the unobservable order index of the displacement sensor, n sensor +1≤j≤n o -n sensor ; j′ represents the observable Gramian matrix W o Eigenvalue of the unobservable mode sub-block of the displacement sensor; M eff,j′ represents the modal effective mass, expressed as represents the degree of participation of each mode in the system dynamics, M is the mass matrix, r is the unit excitation vector, Φ j′ is the j′th column of the vibration mode matrix; σ() represents the standard deviation; · represents the multiplication sign; || ||2 represents the calculation of the matrix bi-norm; Indicates the first n sensor Modal coordinates and sensors to measure physical coordinates The transformation matrix of
[0170] Indicator 2 aims to In sparse cases, it is ensured that the actual physical coordinates measured by the sensor can be optimally converted to the modal coordinates in the modal space.
[0171] Step 4: Use the optimization index as the fitness function of the hybrid particle swarm algorithm, where index 1 takes the sign, so that the optimization direction of the two indexes is the same, that is, the optimization goal is to minimize both objectives as much as possible. Due to the high dimension of the object, the index (13) is often extremely large in power, and can even reach the order of 10 to the twenty-first power when the optimization result is poor, while the ideal value is close to 1. If the index (13) is integrated into (12) as a penalty term, it is often difficult to balance the magnitudes of the two.
[0172] Therefore, the problem is transformed into a multi-objective optimization problem to deal with the situation where the magnitude difference between the two indicators is large, as shown in formula (14).
[0173]
[0174] Step 41: Initialize the number of particles particlesize, position x, velocity v, and the ratio of elite particles to ordinary particles λ r , crossover selection probability λ c , mutation selection probability λm , inertia weight ω, acceleration coefficients c1, c2, criterion weight ratio weight, particle position constraints, particle velocity constraints; set the maximum number of iterations Iter;
[0175] Step 4.2: Set the number of iterations k = 1;
[0176] Step 4.3: Set up the multi-objective fitness function, as well as the elite subgroup and the ordinary subgroup; the specific process is:
[0177] Step 4.3.1: Take optimization criteria 1 and optimization criteria 2 as multi-objective fitness functions, where optimization criteria 1 is negative, and calculate the values of optimization index 1 and optimization index 2;
[0178]
[0179] Solve the actuator installation position layout when the fitness function Cirt_1 is maximum and the fitness function Cirt_2 is minimum. Cirt_1 is negative, so it is converted into a multi-objective optimization problem in which both criteria are minimized.
[0180] Step 4.3.2: Normalize the values of optimization index 1 and optimization index 2 to obtain the normalized values of optimization index 1 and optimization index 2;
[0181] Step 4.3.3: Set the weight ratio of optimization index 1 and optimization index 2, weight the normalized values of optimization index 1 and optimization index 2, and obtain the particle segmentation standard; the expression is:
[0182] weighted fitness =weighted×[norm cirt_1 ,norm cirt_2 ] T
[0183] Among them, weighted represents the weight ratio; norm cirt_1 is the normalized optimization index 1; norm cirt_2 is the normalized optimization index 2; the superscript T indicates transposition;
[0184] The vector obtained by weighting the particle fitness index is only used for particle sorting here and is not used for subsequent individual optimal updates. r Perform population segmentation;
[0185] Step 4.3.4: Weighted fitness Vector descending order (after descending order, use λ r Particle segmentation);
[0186] Step 4.3.5: According to the population proportion λ r (For example, 50%) The particles in descending order are divided into the elite subgroup, and the number of particles in the elite subgroup is N e (For example, there are 10 particles, particles 1 to 5 are elite subgroups), and the latter part is the ordinary subgroup, and the number of particles in the ordinary subgroup is N n (For example, there are 10 particles, particles 6 to 10 are common subgroups), satisfying N e +N n =particlesize;
[0187] The elite subgroup is responsible for guiding the learning process of subsequent ordinary particles and according to the crossover probability λ c Learn the information of other particles to increase the global search capability; the ordinary subgroup follows the mutation probability λ m Randomize its particle learning process to improve local search capabilities;
[0188] Step 4.4: Calculate the speed and individual position of all particles in the elite subgroup; the specific process is:
[0189] The learning strategy of the standard particle swarm algorithm will lead to the loss of particle diversity during the iteration process, increasing the possibility of falling into the local optimum. In order to make the optimization take into account both global search ability and local search ability to cope with the complex and multi-peak characteristics of the objective function, the population ratio λ is used r To divide particles into elite particles and ordinary particles, elite particles are responsible for guiding the learning process of ordinary particles and obtaining information from particles in other dimensions to increase population diversity;
[0190] Step 4.4.1: Let i = 1, calculate the d-dimensional individual position of the i-th particle at the k-th iteration, i = 1, 2, ..., N e , N e is the total number of particles in the elite subgroup; the expression is:
[0191]
[0192] The subscript k is the kth iteration, the subscript k-1 is the k-1th iteration, i represents the order of particles in the elite subgroup, d represents the particle dimension, k is the time, r1 and r2 are two random variables uniformly distributed in the interval [0,1]; Represents the optimal historical solution P of the i-th particle individual updated in the k-1th iteration i,k-1 The dth dimension of , that is, the x corresponding to the small Cirt_1 and Cirt_2 in the history iteration of the i-th particle i,k-1 ; The optimal solution P of the individual history of the i-th particle i,k-1is the x corresponding to which Cirt_1 and Cirt_2 are both small in the history iteration of the i-th particle i,k-1 ; Represents the optimal historical solution P of the mth particle individual updated in the k-1th iteration m,k-1 The dth dimension of (the initial value is a random installation position within the installable range), that is, the x corresponding to Cirt_1 and Cirt_2 are small in each particle history iteration process i,k-1 , m is a random particle in the elite subgroup, m≠i, m=1,2,…,N e ; is the individual historical optimal solution P of the i-th particle updated in the k-1th iteration i,k-1 The d1th dimension of (the initial value is a random installation position within the installable range), that is, the x corresponding to Cirt_1 and Cirt_2 are small in each particle history iteration process i,k-1 , and d1≠d; rand represents any value between (0,1), thereby ensuring that the particle update has a certain degree of randomness; represents the particle velocity v of the ith particle after update in the k-1th iteration i,k-1 The dth dimension of represents the position x of the i-th particle after the update in the k-1th iteration i,k-1 The dth dimension of Represents the particle velocity v of the i-th particle after update in the k-th iteration i,k The dth dimension of represents the individual position x of the i-th particle after update in the k-th iteration i,k The dth dimension of c represents the crossover probability, λ c ∈[0,1]; crossover selection probability λ c It is used to ensure the coordinated execution of the two learning methods, so that the population can learn more search information and ensure the randomness of the learning method and the diversity of the population.
[0193] is the updated optimal solution G of the group in the k-1th iteration k-1 The dth dimension of (the initial value is a random installation position within the installable range);
[0194] For the above formula, both social learning and cognitive learning operators can be reflected in the calculation. A particle can use the social learning operator to perform cross operations between different particles in the same dimension. Since each particle has its own advantages, the local optimization ability is enhanced through local resource exchange. At the same time, the cognitive learning operator is used to perform cross calculations on data of different dimensions of the same particle in order to escape the local minimum.
[0195] Step 4.4.2: Let i = i + 1, repeat step 4.4.1 until all particles in the elite subgroup are calculated, and then execute step 4.6;
[0196] Step 4.5: Calculate the speed and individual position of all particles in the common subgroup; the specific process is:
[0197] The adaptability of ordinary subgroup particles is poor. For each particle, it is asked to randomly select a common particle in front of it as a learning benchmark to update its velocity vector and position characteristics and learn from the experience of better particles.
[0198] Step 4.5.1: Let i′ = 1, calculate the d-dimensional individual position of the i′th particle at the kth iteration, i′ = 1, 2, …, N n , N n is the total number of particles in the common subgroup;
[0199] The expression is:
[0200]
[0201]
[0202] in represents the particle velocity v of the next i′th particle after the kth iteration update i′,k The dth dimension of represents the particle velocity v of the i′th particle after the update in the k-1th iteration i′,k-1 The dth dimension of ; r1 and r2 represent random variables uniformly distributed in the interval [0,1]; It represents the optimal solution P of the individual history of the lth particle after the update in the k-1th iteration. l,k-1 The dth dimension of (the initial value is a random installation position within the installable range), that is, the x corresponding to Cirt_1 and Cirt_2 are small in each particle history iteration process i,k-1 , l is the particle number randomly selected from [1,i′-1], l≠i′; represents the particle with better fitness than particle i′ at the k-1th iteration, that is, any particle that is ranked before particle i during sorting. When particle i is updated, the optimal solution of a particle is randomly selected from the pool of better particles, and crossover operation is performed with it to guide the poorer particles. At the same time, too much information may cause particles to not converge, so the mutation selection probability λ is introduced m , use the global optimal solution to guide the update and speed up the convergence. By selecting the mutation probability, the particles can converge faster and avoid the local optimum at the same time. It represents the optimal solution G of the group after update in the k-1th iteration k-1The dth dimension of (the initial value is a random installation position within the installable range); represents the optimal solution P of the individual history of the i′th particle in the k-1th iteration i′,k-1 The dth dimension of (the initial value is a random installation position within the installable range); represents the individual position x of the i′th particle at the k-1th iteration i′,k-1 The dth dimension of represents the individual position x of the i′th particle at the kth iteration i′,k The dth dimension of represents the particle velocity v of the i′th particle after the update in the kth iteration i′,k The dth dimension of represents the particle velocity v of the i′th particle after the update in the k-1th iteration i′,k-1 The dth dimension of m represents the mutation probability;
[0203] When the Ansys finite element model is established, the model vertices and connecting rods are set as the locations where the sensors can be installed. Without loss of generality, during the iteration of the particle swarm algorithm, the variables to be optimized x of all particles are i,k 、Individual historical optimal solution P i,k , group optimal solution G k The initial value x i,0 , P i,0 , G0 is taken as a random installation position within the installable range. For particles outside the range that appear during the iteration process, they are excluded and randomly moved to an installable position.
[0204] Step 4.5.2: Let i′=i′+1, and repeat step 4.5.1 until all particles in the common subgroup are calculated, and then execute step 4.6;
[0205] Step 4.6: Calculate all elite particles (number is N e ) and ordinary particles (number is N n )'s fitness value (in elite particles (number is N e ) and ordinary particles (number is N n ) is fully updated, calculate all elite particles (number is N e ) and ordinary particles (number is N n ) fitness index), and for all particles (elite particles (number is N e ) and ordinary particles (number is N n )) The individual historical optimal solution P at the kth moment i,k and the group optimal solution G k Make updates;
[0206] After completing the update of the two particle populations, continue to perform the historical optimal solution P of each particle. k and the global optimal solution G k When the two fitness values of the updated particles are both smaller than the original particles, the historical optimal solution of the particles is updated; when the two fitness values of all updated particles are both smaller than G k When , the global optimal solution is updated.
[0207] Step 4.7: Set the number of iterations k = k + 1, and re-execute steps 4.3 to 4.7 until k = Iter, and obtain the final group optimal solution G k , which is the optimal installation position for iteration.
[0208] This implementation method is to build a physical model in Ansys and perform layout optimization simulation based on the derived parameters. The physical structural parameters of the truss are shown in Table 1;
[0209] Table 1 Parameters of large space truss structure
[0210]
[0211] The sensor parameter settings are shown in Table 2;
[0212] Table 2 Sensor parameter settings
[0213]
[0214] Using the multi-objective hybrid particle swarm algorithm proposed in the present invention, the algorithm parameters are selected as follows, and the algorithm flow is shown in Table 3:
[0215] Table 3 Some parameters of multi-objective hybrid particle swarm algorithm
[0216]
[0217] In order to measure the observation effect of the designed algorithm, a short-time step excitation is applied at the excitation point. According to the modal space dynamics, the theoretical modal coordinates under excitation are obtained, and the modal coordinates measured by the sensor after the observed overflow truncation based on the acquired physical displacement are compared. The excitation point is selected as follows: Figure 6 As shown, the excitation is selected in the x-axis direction. Without loss of generality, the excitation point number, i.e., the excitation size, is shown in Table 4:
[0218] Table 4 Excitation point excitation and number
[0219]
[0220] At the same time, the displacement difference between the actual displacement measured by the sensor and the sensor observed modal order is used as the observation overflow, as shown in the following formula:
[0221]
[0222] Where Q r is the observed overflow calculation error, Q real is the theoretical physical displacement without sensor number truncation, and Q is the actual physical displacement after sensor number truncation. In the experiment, the actual physical displacement calculated with an order greater than the number of sensors is taken as the sensor true value, and the physical displacement calculated with the modal order after sensor number truncation is taken as the observed overflow physical displacement. The difference between the two is used to measure the observed overflow of the structure.
[0223] The simulation results are as follows:
[0224] According to the data in Table 1 and Table 3, after the above layout optimization simulation, the fitness curves of the two indicators are as follows: Figure 3 , Figure 4 As shown, the optimized value of index 1 is -0.0003, and the optimized value of index 2 is 16.6521; the layout position is as follows Figure 5 As shown, the target point installation position is shown in Table 5:
[0225] Table 5 Target installation position
[0226]
[0227] Among them, 24x represents the x-axis installation direction of node No. 24, the user-defined numbering system is the numbering system of the Ansys end and the external interface, and the Ansys internal numbering system is the numbering system corresponding to the internal parameters such as the vibration matrix and mass matrix, and matrix mapping can be performed between the two; in the structure, a beam in the Ansys model is regarded as a beam unit node.
[0228] According to Table 4 and Figure 6 The excitation point excitation information is used to apply excitation, and the modal coordinate observation and modal overflow at this time are obtained. The effect is as follows Figure 7-Figure 9 As shown. Figure 7 , Figure 8 , it can be seen that the sensor observed modal coordinates are similar to the theoretical modal coordinates; at the same time, the overflow observed at each sensor target point is as follows Fig. 9 This implementation scheme takes into account the observation performance and the model obtained. Fig. 9 Standard performance, introduced (C dis Φ no ) nsensor The matrix condition number is used as one of the optimization indicators to convert the problem into a multi-objective optimization problem. At the same time, in order to prevent falling into the local optimal solution, a multi-objective hybrid particle swarm algorithm is introduced. According to the particle population segmentation and randomized weight update problem, the optimization result takes into account both locality and globality, and solves the abnormal modal coordinate calculation problem caused by matrix sparsity when optimizing the layout of large truss structures.
[0229] The present invention may also have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.
Claims
1. A flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm, characterized by: The specific process of the method is: Step 1: Use finite element analysis software to establish a flexible attachment model under constrained mode, perform modal analysis on the flexible attachment model under constrained mode, and obtain the modal natural frequency, modal damping, vibration mode matrix and effective mass of each order mode; Step 2: Based on step 1, a flexible structure dynamic model described by modal coordinates when using a displacement sensor is established; Step 3: Based on step 2, set optimization criterion 1 as the optimal criterion for observability under the weighting of modal effective mass; Set optimization criterion 2 to the number of sensors n sensor Modal filter matrix under truncation The two-norm condition number of the truncated modal filter matrix C dis Φ n Take the first n sensor List; Step 4: Take the optimization criterion as the multi-objective fitness function, optimize the multi-objective fitness function, and obtain the sensor installation freedom when the fitness function takes the minimum case.
2. The flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm according to claim 1 is characterized by: In the step 1, a finite element analysis software is used to establish a flexible attachment model under a constrained mode, and a modal analysis is performed on the flexible attachment model under a constrained mode to obtain the modal natural frequency, modal damping, vibration mode matrix and effective mass of each order mode; the specific process is: Use Ansys finite element analysis software to build a physical model of the flexible beam structure, establish constraints on the physical model of the flexible beam structure, and perform modal analysis to obtain the vibration characteristics of the constructed physical model of the flexible beam structure. The vibration characteristics of the physical model of the flexible beam structure include modal natural frequency, modal damping, vibration mode matrix and effective mass of each order mode.
3. The flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm according to claim 2 is characterized by: In step 2, based on step 1, a flexible structure dynamic model described by modal coordinates when using a displacement sensor is established; the expression is: in C and K are the damping matrix and stiffness matrix respectively, All are diagonal matrices; Φ n is the vibration mode matrix; the superscript T means transposition; B d , B c are the interference force and control force action matrices respectively; f d is the interference force, f c For control; C dis is the displacement sensor installation matrix, the number of rows of the displacement sensor installation matrix is the number of sensors, and the number of columns is the total number of degrees of freedom of the node n; η is the modal coordinate, is the modal velocity, is the modal acceleration; Q is the output value of the displacement sensor.
4. The flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm according to claim 3 is characterized by: In step 3, based on step 2, the optimization criterion 1 is set as the optimal criterion of observability under the weighting of modal effective mass; Set optimization criterion 2 to the number of sensors n sensor Cut off The two-norm condition number of ; The specific process is: Optimization criterion 1 is: Optimization criterion 2 is: Among them, Cirt_1 represents optimization index 1, and Cirt_2 represents optimization index 2; n sensor It represents the order observable by the displacement sensor, which is the same as the number of displacement sensors; j represents the order index of the displacement sensor observable, 1≤j≤n sensor ; M eff,j represents the modal effective mass, expressed as M is the mass matrix, r is the unit excitation vector, Φ j is the jth column of the vibration mode matrix; λ j Denotes the observable Gramian matrix W o The displacement sensor can observe the eigenvalues of the modal sub-block; nn sensor represents the unobservable order of the displacement sensor, where n is the total number of degrees of freedom of the node, satisfying The total number of nodes established for the finite element analysis; j′ represents the unobservable order index of the displacement sensor, n sensor +1≤j′≤nn sensor ; λ j′ Denotes the observable Gramian matrix W o Eigenvalues of unobservable modal sub-blocks of displacement sensors; M eff,j′ represents the modal effective mass, expressed as M is the mass matrix, r is the unit excitation vector, Φ j′ is the j′th column of the vibration mode matrix; σ() represents the standard deviation; · represents the multiplication sign; 2 represents the calculation of the matrix's second norm; Indicates the first n sensor Modal coordinates and sensors to measure physical coordinates The transformation matrix.
5. The flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm according to claim 4 is characterized by: In step 4, the optimization criterion is used as the multi-objective fitness function, and the multi-objective fitness function is optimized to obtain the group optimal solution G when the fitness function takes the minimum condition. k ; The group optimal solution G k Refers to the installation freedom of the sensor; The specific process is: Step 4.1: Initialize the number of particles, particle size, position x, velocity v, and the ratio of elite particles to ordinary particles λ r , crossover selection probability λ c , mutation selection probability λ m , inertia weight ω, acceleration coefficients c1, c2, criterion weight ratio weight, particle position constraints, particle velocity constraints; set the maximum number of iterations Iter; Step 4.2: Set the number of iterations k = 1; Step 4.3: Set up the multi-objective fitness function, as well as the elite subgroup and the ordinary subgroup; the specific process is: Step 4.3.1: Take optimization criteria 1 and optimization criteria 2 as multi-objective fitness functions, where optimization criteria 1 is negative, and calculate the values of optimization index 1 and optimization index 2; Step 4.3.2: Normalize the values of optimization index 1 and optimization index 2 to obtain the normalized values of optimization index 1 and optimization index 2; Step 4.3.3: Set the weight ratio of optimization index 1 and optimization index 2, weight the normalized values of optimization index 1 and optimization index 2, and obtain the particle segmentation standard; the expression is: weighted fitness =weighted×[norm cirt_1 ,norm cirt_2 ] T Among them, weighted represents the weight ratio; norm cirt_1 is the normalized optimization index 1; norm cirt_2 is the normalized optimization index 2; the superscript T indicates transposition; weighted fitness The vector obtained by weighting the particle fitness index; Step 4.3.4: Weighted fitness Sort the vector in descending order; Step 4.3.5: According to the population proportion λ r The particles in descending order are divided into the elite subgroup, and the number of particles in the elite subgroup is N. e , the latter part is the ordinary subgroup, the number of particles in the ordinary subgroup is N n ; Step 4.4: Calculate the speed and individual position of all particles in the elite subgroup; the specific process is: Step 4.4.1: Let i = 1, calculate the d-dimensional individual position of the i-th particle at the k-th iteration, i = 1, 2, ..., N e , N e is the total number of particles in the elite subgroup; The expression is: The subscript k is the kth iteration, the subscript k-1 is the k-1th iteration, i represents the order of particles in the elite subgroup, d represents the particle dimension, k is the time; r1 and r2 are two random variables uniformly distributed in the interval [0,1]; represents the optimal solution P of the individual history of the i-th particle in the k-1th iteration i,k-1 The dth dimension of The optimal solution P of the individual history of the i-th particle i,k-1 is the x corresponding to which Cirt_1 and Cirt_2 are both small in the history iteration of the i-th particle i,k-1 ; represents the optimal solution P of the mth particle in the k-1th iteration m,k-1 The d-th dimension, m is a random particle in the elite subgroup, m≠i, m=1,2,…,N e ; is the individual historical optimal solution P of the i-th particle in the k-1th iteration i,k-1 The d1th dimension of , and d1≠d; rand represents any value between (0,1); represents the particle velocity v of the i-th particle at the k-1th iteration i,k-1 The dth dimension of represents the position x of the i-th particle in the k-1th iteration i,k-1 The dth dimension of represents the particle velocity v of the i-th particle at the k-th iteration i,k The dth dimension of represents the position x of the i-th particle at the k-th iteration i,k The dth dimension of λ c represents the crossover probability; is the optimal solution G of the group under the k-1th iteration k-1 The dth dimension of Step 4.4.2: Let i = i + 1, and repeat step 4.4.1 until all particles in the elite subgroup are calculated, and then execute step 4.5; Step 4.5: Calculate the speed and individual position of all particles in the common subgroup; the specific process is: Step 4.5.1: Let i′ = 1, calculate the d-dimensional individual position of the i′th particle at the kth iteration, i′ = 1, 2, …, N n , N n is the total number of particles in the common subgroup; The expression is: in is the particle velocity v of the i′th particle in the kth iteration i′,k The dth dimension of is the particle velocity v of the i′th particle in the k-1th iteration i′,k-1 The dth dimension of r1 and r2 are random variables uniformly distributed in the interval [0,1]; represents the optimal solution P of the individual history of the lth particle in the k-1th iteration l,k-1 The dth dimension of , l is the particle number randomly selected from [1,i′-1], l≠i′, l=1,2,…,N n ; is the optimal solution G of the group under the k-1th iteration k-1 The dth dimension of represents the optimal solution P of the individual history of the i′th particle in the k-1th iteration i′,k-1 The dth dimension of represents the individual position x of the i′th particle at the k-1th iteration i′,k-1 The dth dimension of represents the individual position x of the i′th particle at the kth iteration i′,k The dth dimension of represents the particle velocity v of the i′th particle at the kth iteration i′,k The dth dimension of represents the particle velocity v of the i′th particle in the k-1th iteration i′,k-1 The dth dimension of λ m represents the mutation probability; Step 4.5.2: Let i′=i′+1, and repeat step 4.5.1 until all particles in the common subgroup are calculated, and then execute step 4.6; Step 4.6: Calculate the fitness values of all elite particles and ordinary particles, and calculate the individual historical optimal solution P of all particles at the kth moment i,k and the group optimal solution G k Make updates; Step 4.7: Set the number of iterations k = k + 1, and re-execute steps 4.3 to 4.7 until k = Iter, where Iter is the maximum number of iterations set, to obtain the final group optimal solution G k , which is the optimal installation position for iteration.
6. The flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm according to claim 5 is characterized by: In step 4.3.1, optimization criteria 1 and optimization criteria 2 are used as multi-objective fitness functions, optimization criteria 1 is negative, and the values of optimization index 1 and optimization index 2 are calculated; The specific process is: 1) Displacement sensor installation matrix C dis The number of rows is the number of sensors, and the number of columns is the total number of degrees of freedom n; 2) The value of represents the displacement sensor installation matrix C dis Middle The column value is 1; Indicates the kth iteration The individual positions of the particles; No. The particle is the sum of elite particles and ordinary particles. Particles; 3) Replace C in 2) dis Substitute the formula y=C dis Φ n T dis z=Cz, calculate C matrix; Where y is the sensor output; z is the state vector, η1, represents the first-order modal displacement and modal velocity, η2, represents the second-order modal displacement and modal velocity, η n , represents the nth-order modal displacement and modal velocity, and the superscript T indicates the transposition; Φ n is the vibration mode matrix; T dis The even columns are all 0, and the odd columns are unit column vectors, so: 4) Input the A matrix and C matrix into the Gram matrix function, and the Gram matrix function outputs the block matrix W o , block matrix W o First n sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j ; after nn sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j′ ; Where A is α The diagonal array is composed of ξ α is the modal damping of the αth order mode, 1≤α≤n; ω α is the modal natural frequency of the αth order mode, 1≤α≤n; 5) j , j′ Substitute optimization index 1 and optimization index 2 to obtain the values of optimization index 1 and optimization index 2.
7. The flexible spacecraft layout optimization method based on multi-objective hybrid particle swarm according to claim 6 is characterized by: In step 4.6, the fitness of all elite particles and ordinary particles is calculated, and the individual historical optimal solution P of all particles at the kth moment is calculated. i,k and the group optimal solution G k Update; the specific process is: Step 4.6.1: Calculate the fitness values of all elite particles and ordinary particles; the specific process is: 1) Displacement sensor installation matrix C dis The number of rows is the number of sensors, and the number of columns is the total number of degrees of freedom n; 2) The value of represents the displacement sensor installation matrix C dis Middle The column value is 1; Indicates the kth iteration The individual positions of the particles; No. The particle is the sum of elite particles and ordinary particles. Particles; 3) Replace C in 2) dis Substitute the formula y=C dis Φ n T dis z=Cz, calculate C matrix; Where y is the sensor output; z is the state vector, η1, represents the first-order modal displacement and modal velocity, η2, represents the second-order modal displacement and modal velocity, η n , represents the nth-order modal displacement and modal velocity, and the superscript T indicates the transposition; Φ n is the vibration mode matrix; T dis For even columns all zero, odd columns are unit column vectors, we have: 4) Input the A matrix and C matrix into the Gram matrix function, and the Gram matrix function outputs the block matrix W o , block matrix W o First n sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j ; after nn sensor The eigenvalue of the sub-block is λ in the multi-objective fitness function j′ ; Where A is α The diagonal array is composed of ξ α is the modal damping of the αth order mode, 1≤α≤n; ω α is the modal natural frequency of the αth order mode, 1≤α≤n; 5) j ,λ j′ Substitute optimization index 1 and optimization index 2 to obtain the values of optimization index 1 and optimization index 2; Step 4.6.2: When the particle k When the values of optimization index 1 and optimization index 2 are both less than the values of optimization index 1 and optimization index 2 at time k-1, particle Individual historical optimal solution Update to time k Otherwise the particle Individual historical optimal solution Keep it as k-1 moment The value of Indicates the kth iteration The individual positions of the particles No. The particle is the sum of elite particles and ordinary particles. Particles; Step 4.6.3: When the particle k Individual historical optimal solution The values of optimization index 1 and optimization index 2 are both less than the optimal solution G of the group at k-1 time k-1 When the values of optimization index 1 and optimization index 2 of the dth dimension are equal, the optimal solution G of the group at time k k Update to k-time particle Individual historical optimal solution Otherwise, the optimal group solution G at time k k Keep the optimal solution G of the group at k-1 times k-1 .
Citation Information
Cited By
Method, device and equipment for determining layout position of spaceflight surveying vessel based on particle swarm, medium and product
CN119129297A
Sensor array multi-objective optimization method for magnetic field signal denoising
CN120297042A
A multi-objective optimization method for sensor arrays for magnetic field signal denoising
CN120297042B
Method for collaborative optimization of vibration control through sensor arrangement and feedback algorithm
CN120821213A
Feature extraction method based on machine tool spindle
CN121542711A
