An aircraft low-drag high-efficiency aerodynamic optimization method based on elite knowledge transfer

Abstract

The application relates to the technical field of aircraft aerodynamic design and optimization, and discloses an aircraft low-resistance high-efficiency aerodynamic optimization method based on elite knowledge migration, which comprises the following steps: determining design variables of an aircraft aerodynamic shape and setting geometric constraint conditions; constructing a search space of the design variables; generating an independent population for each subtask by using the search space; performing aerodynamic simulation on each individual in each population, calculating the fitness value of the individual by using a comprehensive objective function constructed by the application, so as to determine an elite individual; executing a cross-task elite knowledge migration mechanism in the evolution process of the individual of the population; independently searching each population by adopting a multi-task parallel collaborative optimization mode, and determining the updating mode of the population according to whether the migration operation is triggered; and finally determining the global optimal individual of each population as an aerodynamic design scheme of the corresponding subtask of the population.

Description

Technical Field

[0001] This invention relates to the field of aircraft aerodynamic design and optimization technology, specifically to a low-drag and high-efficiency aerodynamic optimization method for aircraft based on elite knowledge transfer. Background Technology

[0002] With the rapid development of aircraft technology, modern aircraft, such as solid rockets, often need to maintain excellent aerodynamic performance across a wide speed range (covering subsonic, transonic, and supersonic speeds) and multiple operating conditions. Aerodynamic shape optimization, as one of the core aspects of aircraft design, directly determines its range, accuracy, and overall combat effectiveness. However, existing aerodynamic optimization methods for aircraft still face the following severe challenges when dealing with complex engineering design tasks:

[0003] The "information silo" dilemma in multi-condition collaborative design. An aircraft experiences different Mach numbers and various angles of attack throughout the entire mission profile. Traditional methods typically treat these conditions as completely independent optimization tasks and solve them separately. In the update mechanism of standard algorithms (such as population optimization), each individual is only influenced by its own historical best and the global best of the population. This approach has significant drawbacks:

[0004] Local convergence risk: The population under a certain operating condition is very likely to fall into the local extreme value trap. Due to the lack of information exchange, it is unable to learn from the high-quality geometric configuration regions explored by the population under another related operating condition, resulting in the overall convergence stagnation. Waste of computational resources: The aerodynamic shape of the aircraft has the characteristic of "heterogeneous homology". The high-quality shapes under different operating conditions have a strong proximity correlation in the geometric parameter space. Independent evolution ignores this spatial correlation, resulting in a lot of computing power being wasted on repeated searches of inefficient regions.

[0005] The evaluation metrics are too simplistic and lack economic considerations. Existing optimization studies often focus on maximizing pure aerodynamic performance indicators such as lift-to-drag ratio, neglecting the crucial dimension of "low cost" in aircraft development. In aircraft design, energy consumption per unit range (primarily caused by drag) is the core indicator for evaluating the cost-effectiveness of an aircraft. Currently, there is a lack of an effective method to explicitly incorporate unit range cost into the objective function of multi-mission optimization, resulting in optimized solutions that, while meeting performance targets, suffer from poor engineering economics.

[0006] Limitations of existing algorithm frameworks for practical application: To balance computational accuracy and time consumption, existing technologies often introduce surrogate models such as radial basis functions; however, the construction and updating logic of surrogate models is extremely complex, and they are prone to prediction deviations when dealing with high-dimensional, multi-constraint problems. For practical engineering tasks at the grassroots level, there is an urgent need for a fast optimization framework that is simple in structure, robust, requires no cumbersome pre-training process, and can balance performance and cost. Summary of the Invention

[0007] The purpose of this invention is to provide a low-drag and high-efficiency aerodynamic optimization method for aircraft based on elite knowledge transfer, in order to solve the problems existing in the aerodynamic optimization of aircraft, such as low efficiency of independent design under various operating conditions, high cost of high-fidelity simulation calculation, complex construction of proxy models and difficulty in handling experience sharing of multiple tasks in a wide speed range, and lack of consideration for evaluation indicators of unit range cost.

[0008] To achieve the above objectives, the present invention employs the following technical solution:

[0009] A method for low-drag and high-efficiency aerodynamic optimization of aircraft based on elite knowledge transfer includes:

[0010] Determine the design variables for the aerodynamic shape of the aircraft and set the geometric constraints; construct the search space for the design variables;

[0011] Based on the different flight speed ranges of the aircraft, corresponding sub-tasks are set, and an independent population is generated for each sub-task using the search space. Each individual in the population is a set of design variables; each population is initialized.

[0012] Aerodynamic simulations were performed on each individual in each population to calculate the average lift-to-drag ratio of the individual at different angles of attack. A comprehensive objective function was constructed using the average lift-to-drag ratio and the unit range cost index. The fitness value of the individual was calculated using the comprehensive objective function to determine the elite individuals.

[0013] During the individual evolution of the population, a cross-task elite knowledge transfer mechanism is implemented; in each iteration, the position vector of an elite individual in one population is copied with the migration probability and replaced with the position vector of a random individual in another population.

[0014] A multi-task parallel collaborative optimization approach is adopted to independently search each population. Before updating the velocity vector set and position vector set in each iteration of the population, the execution result of the cross-task elite knowledge transfer mechanism is checked first, and the update method of the population is determined according to whether the transfer operation is triggered. Finally, the globally optimal individual of each population is determined as the aerodynamic design scheme of the corresponding subtask of the population.

[0015] Furthermore, the design variables for the aerodynamic shape of the aircraft include the nose type and the nose length. Mid-section length of the aircraft Length of the aircraft tail and the diameter of the tail end of the aircraft ;

[0016] The geometric constraints include the total length of the aircraft. Fixed, mid-section diameter of the aircraft And the fixed diameter of the aircraft's nose and the overall length of the aircraft Length of aircraft nose and the length of the middle part of the aircraft and the length of the aircraft's tail The range of values ​​is diameter at the tail end of the aircraft The range of values ​​is .

[0017] Furthermore, by using the Latin hypercube sampling method, the value space of all design variables in the search space is sampled, and each set of design variables obtained from the sampling is regarded as an individual, thereby constructing a population;

[0018] Both the set of position vectors and the set of velocity vectors of the population are two-dimensional vectors. The first dimension corresponds to different individuals in the population, and the second dimension corresponds to different design variables.

[0019] Furthermore, the overall objective function is as follows:

[0020] ;

[0021] in, Represents an individual fitness value; For selection One angle of attack; The selected angle of attack number; The lift coefficient; This is the drag coefficient; Indicates angle of attack corresponding ; Indicates the average rise-to-drag ratio; The preset cost adjustment factor, ; This represents the cost per unit range. For the angle of attack The corresponding weighting factor, ; Indicates angle of attack The corresponding drag coefficient;

[0022] The individual with the highest fitness value in the population is considered the elite individual.

[0023] Furthermore, in the cross-task elite knowledge transfer mechanism, the initial transfer probability... Set as:

[0024] ;

[0025] in, The dimension of the design variables; The number of individuals in the population; This is an adjustment coefficient, and its value range is... ; To correct the bias, the value range is: ; It is the inverse hyperbolic tangent function;

[0026] The formula for updating the migration probability is:

[0027] ;

[0028] in This represents the updated migration probability.

[0029] Furthermore, the subtasks include subtask A and subtask B, with corresponding populations A and B, respectively; the cross-task elite knowledge transfer mechanism includes:

[0030] During each evolutionary iteration, the position vectors of elite individuals in population A and population B are locked respectively. and ;

[0031] Generate a A uniformly random number is generated within the interval. If this number is less than the current migration probability, a migration operation is triggered.

[0032] The migration process from subtask B to subtask A is as follows: the position vectors of elite individuals in population B are... Introduce an individual index into population A; select an individual index from population A. The set of current position vectors of population A The OK Directly overwrite as ;

[0033] For each individual in the population whose position vector is replaced by an elite individual, its historical best fitness value is reset to the fitness value of the elite individual.

[0034] Furthermore, the multi-task parallel collaborative optimization method includes:

[0035] In each generation iteration, the evolutionary calculations of population A and population B are triggered simultaneously;

[0036] Before updating the population's velocity vector set, the execution result of the cross-task elite knowledge transfer mechanism is checked in real time; if the transfer operation is triggered, the following adaptive update formula is executed on the population of each subtask:

[0037] ;

[0038] ;

[0039] in: and These represent the updated set of position vectors and the set of velocity vectors, respectively. This is the set of historical best position vectors for each individual. and These represent the current set of position vectors and the set of velocity vectors, respectively.

[0040] ;

[0041] in, This represents the globally optimal position vector; when a migration operation is triggered... Indicates population The migration operation updated the position vector of the current elite individual after updating the position vector set; This indicates the population without migration operations. The position vector of the current elite individual; , As random factors, respectively in Independent, uniform, random sampling within the interval; For inertial weights, , For learning factors;

[0042] ;

[0043] in These are the preset lower and upper limits for the weight values;

[0044] ;

[0045] ;

[0046] in, These are the preset lower and upper limits for the learning factor.

[0047] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, it implements the aircraft low-drag high-efficiency aerodynamic optimization method based on elite knowledge transfer.

[0048] A computer-readable storage medium storing a computer program; when executed by a processor, the computer program implements the aircraft low-drag high-efficiency aerodynamic optimization method based on elite knowledge transfer.

[0049] Compared with the prior art, the present invention has the following technical features:

[0050] 1. Significantly improves design iteration efficiency: Through a cross-task elite knowledge transfer mechanism, the similarity of the design space under high and low Mach number conditions is fully utilized, enabling the algorithm to quickly identify high-quality search regions through "experience learning." Compared to independent condition optimization, the convergence speed is significantly improved without using a surrogate model, greatly shortening the development cycle.

[0051] 2. Achieving true low-cost design evaluation: Unlike traditional methods that only focus on aerodynamic performance indicators, this invention explicitly incorporates unit range cost into the objective function, so that the optimization results not only meet the aerodynamic performance standards, but also have significant advantages in economic efficiency (unit range cost), accurately responding to the technical indicators of low-drag and high-efficiency key technologies.

[0052] 3. The algorithm is robust and easy to implement in engineering: It adopts a population algorithm framework, which avoids the complex mutation operator design of the differential evolution algorithm in the paper by Chen Xuan et al., and does not require the construction and training of a cumbersome radial basis function surrogate model, reducing the programming difficulty and dependence on computing resources, and has strong engineering applicability.

[0053] 4. Adaptable to wide speed range and full-condition performance requirements: By weighted optimization of performance at various angles of attack, the aircraft is ensured to have good steady-state flight characteristics across the entire mission profile, avoiding performance failures in other conditions caused by single-point optimization. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0055] Figure 2 This is a schematic diagram showing some design variables and geometric constraints of the aircraft in an embodiment of the present invention;

[0056] Figure 3 This is the convergence curve of the comprehensive objective function value of the method of the present invention;

[0057] Figure 4 This is the convergence curve of the performance function value of the method of the present invention;

[0058] Figure 5 This is the convergence curve of the cost function value of the method of the present invention;

[0059] Figure 6 This invention provides a comparison of the convergence curves of the cross-task elite knowledge transfer mechanism and the comprehensive objective function values ​​of independent evolution of various populations.

[0060] Figure 7 This paper compares the convergence curves of the multi-task parallel collaborative optimization method of the present invention with those of the comprehensive objective function value of serial optimization. Detailed Implementation

[0061] How to combine multi-task transfer learning to break down the "information silos" between populations, utilize cross-condition design experience sharing to improve optimization efficiency, and achieve a balance between performance and cost has become a key technical problem that urgently needs to be solved in the field of aircraft aerodynamic design.

[0062] This invention provides a low-drag, high-efficiency aerodynamic optimization method for aircraft based on elite knowledge transfer. It employs a multi-task evolutionary algorithm framework, abstracting the aerodynamic optimization requirements under different flight conditions into multiple parallel sub-tasks. A population optimization algorithm is used as the underlying search engine, and a cross-task elite knowledge transfer mechanism is introduced between the populations to achieve design experience sharing across flight conditions. Simultaneously, a unit range cost function is explicitly introduced into the objective function, enabling efficient collaborative optimization of comprehensive aerodynamic performance and economic cost across the entire angle of attack range without relying on complex proxy models. The overall flowchart of this invention is as follows. Figure 1 As shown, the specific steps include:

[0063] Step 1: Determine the design variables of the aircraft's aerodynamic shape and set geometric constraints; construct the search space for the design variables.

[0064] Step 1-1: Perform geometric modeling of the aircraft using the design variables of its aerodynamic shape; the design variables include:

[0065] aircraft head type This is an enumeration type; it includes conical, oval, and von Kármán shapes, each represented by an enumeration value. , ,express.

[0066] aircraft nose length , is a scalar (floating-point) type, and the unit is meters.

[0067] Mid-section length of the aircraft , is a scalar (floating-point) type, and the unit is meters.

[0068] aircraft tail length , is a scalar (floating-point) type, and the unit is meters.

[0069] aircraft tail end diameter , is a scalar (floating-point) type, and the unit is meters.

[0070] Steps 1-2: Based on the overall design requirements of the aircraft, specify the geometric constraints, including:

[0071] Total length of the aircraft Fixed, in meters, limiting the longitudinal range of the pneumatic control surface arrangement; in the embodiment, it is set as follows: .

[0072] Mid-section diameter of the aircraft and the diameter of the aircraft's nose The units are fixed, all in meters, and are used as the reference parameters for calculating the pneumatic base. In this embodiment, they are set as follows: .

[0073] Overall length of the aircraft The length of the aircraft's nose Mid-section length of the aircraft Length of the aircraft tail The sum of .

[0074] aircraft nose length Mid-section length of the aircraft Length of the aircraft tail The range of values ​​is .

[0075] aircraft tail end diameter The range of values ​​is .

[0076] The selection of design variables and geometric constraints in this scheme is as follows: Figure 2 As shown.

[0077] Steps 1-3 assign a corresponding value space to each design variable, and the value spaces of all design variables constitute the search space. In addition, different value spaces can be assigned to design variables for different sub-tasks, thus constructing a search space for that sub-task.

[0078] Step 2: Based on the different flight speed ranges of the aircraft, set corresponding sub-tasks, and use the search space to generate an independent population for each sub-task. Each individual in the population is a set of design variables; initialize each population.

[0079] The spacecraft described in this invention, such as a solid rocket, exhibits wide-range flight characteristics, including subsonic, transonic, and supersonic regions. Multiple sub-tasks are defined for different flight speed ranges, and an independent population is initialized for each sub-task. Each population is generated by sampling the search space. Each individual represents a set of design variables; construct the initial set of position vectors and velocity vectors for the population.

[0080] That is, set the number of subtasks according to optimization requirements. For example, in one embodiment of the present invention Includes constructing low Mach numbers for the subsonic region. Subtask A, and constructing high Mach numbers for the transonic region. Subtask B. For each subtask, initialize an independent population; in the population, each individual represents a set of design variables, obtained by sampling from the search space; for example, the Latin hypercube sampling method can be used to sample the value space of all design variables in the search space, and each set of design variables obtained by sampling is regarded as an individual; that is, the form of the individual is [ , , , , ]; In the embodiments, it is set .

[0081] Both the set of position vectors and the set of velocity vectors in the population are two-dimensional vectors. The first dimension (rows) corresponds to different individuals in the population, and the second dimension (columns) corresponds to different design variables. Initial set of position vectors. The initial velocity vector set consists of the initial values ​​of all design variables for all individuals in the population. The update direction of all design variables representing all individuals in the population is set to... .

[0082] Step 3: Perform aerodynamic simulations for each individual in each population to calculate the average lift-to-drag ratio at different angles of attack. Construct a comprehensive objective function using the average lift-to-drag ratio and the unit range cost index. Calculate the fitness value of each individual using the comprehensive objective function to determine elite individuals. Update the historical best position vector for each individual. .

[0083] To meet the low-drag and high-efficiency requirements of this invention, cost is taken as one of the optimization objectives. Considering multi-angle-of-attack performance, the following comprehensive objective function is constructed:

[0084] ;

[0085] in, For performance functions, It is the cost function; Represents an individual The fitness value (i.e., the comprehensive objective function value); For the selected angle of attack, The selected angle of attack number; The lift coefficient, This is the drag coefficient; Indicates angle of attack corresponding ; This represents the average rise-to-drag ratio; This is a preset cost adjustment factor used to balance aerodynamic performance and economy. It can be set according to requirements, for example, its value is 1. This represents the cost per unit range. For the angle of attack The corresponding weighting factor, ; Indicates angle of attack The corresponding drag coefficient.

[0086] In the embodiments of the present invention, the setting of the comprehensive objective function mainly considers the following two aspects:

[0087] Overall performance across multiple angles of attack: Based on engineering experience, considering the main range of angles of attack values ​​in aerodynamic optimization scenarios, Three typical angles of attack, 0°, 5°, and 10°, were selected as representatives to calculate the lift coefficients at these three angles of attack. and drag coefficient Thus, each angle of attack is obtained. The lift-to-drag ratio ,but This represents the average lift-to-drag ratio.

[0088] Unit range cost calculation: Define unit range cost index This indicator is related to flight drag. A positive correlation is shown, set at the angle of attack. drag coefficient below The sum of , The weighting factor is used to reduce drag and extend flight range, thereby reducing the operational cost per unit distance; in this embodiment, the weighting factor is set to... .

[0089] Step 4: During the individual evolution of the population, execute the cross-task elite knowledge transfer mechanism; in each iteration, copy the position vector of an elite individual in one population and replace the position vector of a random individual in another population with the migration probability.

[0090] To avoid blind migration disrupting the stability of the population, this scheme abandons the practice of blindly setting fixed probabilities based on experience, and instead uses a cross-task elite knowledge transfer mechanism for migration; the initial migration probability is defined as:

[0091]

[0092] in, The dimension of the design variables (in this embodiment, ,correspond (different design variables); The number of individuals in the population (in this embodiment, ); It is the inverse hyperbolic tangent function, used to map the ratio to a nonlinear interval; Adjustment coefficient (range of values) In this embodiment, ); To correct the bias (range of values) In this embodiment, ).

[0093] This formula reflects the dynamic balance between dimensionality and population size: the higher the dimensionality, the more difficult the search becomes, and the greater the need for external knowledge assistance; the larger the population size, the greater the internal diversity, and the less dependent it is on external factors. In the embodiment, the initial migration probability is calculated. .

[0094] With evolution and iteration, the formula is adopted. The migration probability is adaptively updated while gradually decreasing; where, This represents the updated migration probability.

[0095] In each generation of evolutionary iterations, taking a scenario with two sub-tasks as an example, namely sub-task A (low Mach number task) and sub-task B (high Mach number task), the corresponding populations are population A and population B, respectively; the cross-task elite knowledge transfer mechanism is as follows:

[0096] Step 4-1, Elite Individual Extraction: During each evolutionary iteration, the position vectors of elite individuals in population A and population B are locked respectively. and Elite individuals refer to those with the highest fitness values ​​in the population.

[0097] Step 4-2, Probabilistic Decision; Generate a A uniformly random number within an interval; if this number is less than If so, the migration operation will be triggered, and step 4-3 will be executed.

[0098] Step 4-3, Position Replacement; The migration process from subtask B to subtask A is: The position vectors of elite individuals in population B are... Introduced into population A; in order to maintain population size Without changing the index, a random elimination strategy is used to select an individual index from population A. The set of current position vectors of population A The OK Directly overwrite as ,Right now Similarly, the migration process from subtask A to subtask B follows the same logic, migrating elite individuals from population A to population B, which will not be elaborated here.

[0099] Through the above operations, the set of position vectors after the migration operation based on population A can be made possible. When determining the current elite individuals, the position vector information of elite individuals in population B will be comprehensively considered.

[0100] Step 4-4, historical memory reset: For each individual in the population whose position vector has been replaced by an elite individual, its historical best fitness value (i.e., the maximum fitness value in each iteration) is reset to the fitness value of the elite individual, ensuring that it can immediately start searching around this new position vector in subsequent iterations.

[0101] Step 5: Use a multi-task parallel collaborative optimization method to independently search each population; before updating the velocity vector set and position vector set in each iteration of the population, first check the execution result of the cross-task elite knowledge transfer mechanism, and determine the update method of the population based on whether the transfer operation is triggered; finally, determine the globally optimal individual of each population as the aerodynamic design scheme of the corresponding subtask of the population.

[0102] To overcome the inefficiency of traditional sequential optimization (i.e., optimizing subtask A first and then subtask B), this invention employs a multi-task parallel collaborative optimization method; the specific execution actions are as follows:

[0103] Step 5-1, parallel state update; in each generation iteration, the computing unit simultaneously triggers the evolutionary calculation of population A and population B; this parallelism is reflected in the fact that the two populations each perform individual aerodynamic simulation and fitness evaluation in an independent search space, without occupying each other's computing resources.

[0104] Step 5-2, real-time embedding of transfer information; this is the core link in forming a whole between parallel logic and the transfer mechanism; before updating the population's velocity vector set, the execution result of the cross-task elite knowledge transfer mechanism is checked in real time; if a transfer operation is triggered, the following adaptive update formula is executed on the population of each subtask:

[0105] ;

[0106] ;

[0107] in: and These represent the updated set of position vectors and the set of velocity vectors, respectively. This is the set of historical best position vectors for each individual. and These represent the current set of position vectors and the set of velocity vectors, respectively.

[0108] ;

[0109] in, This represents the globally optimal position vector; when a migration operation is triggered... Indicates population The migration operation updated the position vector of the current elite individual after updating the position vector set; This indicates the population without migration operations. The position vector of the current elite individual; , As random factors, respectively in Independent, uniform, and random sampling is performed on the interval. For inertial weights, , For learning factor ( As a self-awareness learning factor, (For social cognitive learning factors), adaptive strategies that are dynamically adjusted with the number of iterations are adopted to ensure that the parallel search has strong exploratory power in the early stage and strong convergence in the later stage.

[0110] Specifically: Inertia weight The adaptive dynamic adjustment is performed within a preset range. The linear decreasing strategy on, In the early stages of iteration, Larger (close to) This endows individuals with greater kinetic energy, enabling them to search over a wide area and maintain population diversity; in the later stages of iteration, Smaller (closer) This facilitates precise searches by individuals near elite individuals; in this embodiment, it is set... .

[0111] Learning factor , Adaptive dynamic adjustment employs a time-varying asymmetric strategy and self-awareness learning factors. Social cognitive learning factors . Early stage big, When individuals are young, they tend to engage in "self-exploration" to avoid getting trapped in local optima too early; later on... Small, Larger values ​​mean that individuals tend to converge towards the "global optimum," accelerating convergence. In this embodiment, we set... .

[0112] This invention views "parallelism" and "migration" as an organic whole: parallelism provides the breadth of the search (simultaneously covering different Mach number conditions), while migration provides the depth of the search (utilizing the physical similarity between conditions to escape local optima). Under this mechanism, subtask A not only searches within its own population space but can also observe the progress of subtask B in real time through parallel collaboration. Once subtask B discovers individuals that better conform to aerodynamic laws (i.e., elite individuals), this characteristic will immediately permeate the evolutionary process of subtask A with a certain probability, thereby guiding the population of subtask A to gather in more promising areas, achieving an acceleration effect of "1+1>2".

[0113] Step 5-3 involves repeatedly executing the parallel update process until the preset maximum number of iterations is reached (in this embodiment, it is set to...). After the iteration stops, the globally optimal position vector is extracted from population A and population B respectively. and The corresponding global optimal individuals; these two global optimal individuals are the optimal solutions for subtask A and subtask B; the set of design variables contained in each global optimal individual is the design scheme for the corresponding subtask that has the best overall aerodynamic performance and the lowest cost per unit range within the multi-angle-of-attack range.

[0114] Based on actual experimental results, the populations in all embodiments of this invention are in The iterations converged within the specified number of iterations. The relevant calculation results and conclusions of the example are as follows:

[0115] The convergence of the overall objective function value is as follows: Figure 3 As shown; from this, we can conclude that the optimization of subtask A is effective when the number of iterations is greater than [a certain threshold]. The optimization of subtask B tends to converge when the number of iterations is greater than a certain threshold. As the process converges, the overall objective function value (fitness value) after optimization of subtask A is slightly higher than that after optimization of task B.

[0116] The changes in the performance function value and the cost function value are as follows: Figure 4 and Figure 5 As shown, it can be concluded that the performance function values ​​of subtask A and subtask B after optimization convergence are almost the same, while the cost function value of subtask B is slightly higher. This is because, under the same conditions, the higher the Mach number, the greater the drag coefficient, resulting in a higher cost per unit range.

[0117] A comparison is made between the cross-task elite knowledge transfer mechanism and the optimization of each subtask, and the convergence of the overall objective function value is as follows: Figure 6 As shown, it can be concluded that the cross-task elite knowledge transfer mechanism accelerated the convergence of subtask A and subtask B, and improved the comprehensive objective function value after the convergence of subtask A and subtask B.

[0118] The multi-task parallel collaborative optimization method in this invention is compared with the serial (non-parallel) optimization method, and the convergence of the overall objective function value is as follows: Figure 7 As shown, it can be concluded that the convergence time of the parallel collaborative optimization of subtasks A and B is reduced by nearly half compared to that of the serialization optimization.

[0119] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for optimizing low-drag and high-efficiency aerodynamics of aircraft based on elite knowledge transfer, characterized in that, include: Determine the design variables for the aerodynamic shape of the aircraft and set the geometric constraints; Construct the search space for the design variables; Based on the different flight speed ranges of the aircraft, corresponding sub-tasks are set, and an independent population is generated for each sub-task using the search space. Each individual in the population is a set of design variables; each population is initialized. Aerodynamic simulations were performed on each individual in each population to calculate the average lift-to-drag ratio at different angles of attack. A comprehensive objective function was constructed using the average lift-to-drag ratio combined with the unit range cost index. This comprehensive objective function was then used to calculate the fitness value of each individual, thereby identifying elite individuals. The comprehensive objective function is as follows: ； in, Represents an individual fitness value; For selection One angle of attack; The selected angle of attack number; The lift coefficient; This is the drag coefficient; Indicates angle of attack corresponding ; Indicates the average rise-to-drag ratio; The preset cost adjustment factor, ; This represents the cost per unit range. For the angle of attack The corresponding weighting factor, ; Indicates angle of attack The corresponding drag coefficient; The individual with the highest fitness value in the population is considered the elite individual. During the individual evolution of the population, a cross-task elite knowledge transfer mechanism is implemented; in each iteration, the position vector of an elite individual in one population is copied with the migration probability and replaced with the position vector of a random individual in another population. A multi-task parallel collaborative optimization approach is adopted to independently search each population. Before updating the velocity vector set and position vector set in each iteration of the population, the execution result of the cross-task elite knowledge transfer mechanism is checked first, and the update method of the population is determined according to whether the transfer operation is triggered. Finally, the globally optimal individual of each population is determined as the aerodynamic design scheme of the corresponding subtask of the population.

2. The method for low-drag and high-efficiency aerodynamic optimization of aircraft based on elite knowledge transfer according to claim 1, characterized in that, The design variables for the aerodynamic shape of the aircraft include the nose type and the nose length. Mid-section length of the aircraft Length of the aircraft tail and the diameter of the tail end of the aircraft ; The geometric constraints include the total length of the aircraft. Fixed, mid-section diameter of the aircraft And the fixed diameter of the aircraft's nose and the overall length of the aircraft Length of aircraft nose and the length of the middle part of the aircraft and the length of the aircraft's tail The range of values ​​is diameter at the tail end of the aircraft The range of values ​​is .

3. The method for low-drag and high-efficiency aerodynamic optimization of aircraft based on elite knowledge transfer according to claim 1, characterized in that, The Latin hypercube sampling method is used to sample the value space of all design variables in the search space. Each set of design variables obtained from the sampling is regarded as an individual, thereby constructing a population. Both the set of position vectors and the set of velocity vectors of the population are two-dimensional vectors. The first dimension corresponds to different individuals in the population, and the second dimension corresponds to different design variables.

4. The method for low-drag and high-efficiency aerodynamic optimization of aircraft based on elite knowledge transfer according to claim 1, characterized in that, In the cross-task elite knowledge transfer mechanism, the initial transfer probability Set as: ； in, The dimension of the design variables; The number of individuals in the population; This is an adjustment coefficient, and its value range is... ; To correct the bias, the value range is: ; It is the inverse hyperbolic tangent function; The formula for updating the migration probability is: ； in This represents the updated migration probability.

5. The method for low-drag and high-efficiency aerodynamic optimization of aircraft based on elite knowledge transfer according to claim 1, characterized in that, The subtasks include subtask A and subtask B, with corresponding populations A and B, respectively; the cross-task elite knowledge transfer mechanism includes: During each evolutionary iteration, the position vectors of elite individuals in population A and population B are locked respectively. and ; Generate a A uniformly random number is generated within the interval. If this number is less than the current migration probability, a migration operation is triggered. The migration process from subtask B to subtask A is as follows: the position vectors of elite individuals in population B are... Introduce an individual index into population A; select an individual index from population A. The set of current position vectors of population A The OK Directly overwrite as ; For each individual in the population whose position vector is replaced by an elite individual, its historical best fitness value is reset to the fitness value of the elite individual.

6. The method for low-drag and high-efficiency aerodynamic optimization of aircraft based on elite knowledge transfer according to claim 1, characterized in that, The multi-task parallel collaborative optimization method includes: In each generation iteration, the evolutionary calculations of population A and population B are triggered simultaneously; Before updating the population's velocity vector set, the execution result of the cross-task elite knowledge transfer mechanism is checked in real time; if the transfer operation is triggered, the following adaptive update formula is executed on the population of each subtask: ； ； in: and These represent the updated set of position vectors and the set of velocity vectors, respectively. This is the set of historical best position vectors for each individual. and These represent the current set of position vectors and the set of velocity vectors, respectively. ； in, This represents the globally optimal position vector; when a migration operation is triggered... Indicates population The migration operation updated the position vector of the current elite individual after updating the position vector set; This indicates the population without migration operations. The position vector of the current elite individual; , As random factors, respectively in Independent, uniform, random sampling within the interval; For inertial weights, , For learning factors; ； in These are the preset lower and upper limits for the weight values; ； ； in, These are the preset lower and upper limits for the learning factor.

7. A terminal device, comprising a processor, a memory, and a computer program stored in the memory; characterized in that, When the processor executes the computer program, it implements the low-drag, high-efficiency aerodynamic optimization method for aircraft based on elite knowledge transfer as described in any one of claims 1-6.

8. A computer-readable storage medium storing a computer program; characterized in that, When the computer program is executed by the processor, it implements the low-drag, high-efficiency aerodynamic optimization method for aircraft based on elite knowledge transfer as described in any one of claims 1-6.

Images