An active and passive control method for suppressing the shock wave buffeting of swept wings

Abstract

The present invention discloses an active and passive control method for suppressing the shock wave flutter of a swept wing, which belongs to the field of active and passive coordinated control of aircraft flow, and includes: establishing a numerical simulation geometric model of a swept wing, and a design model of an active and passive coordinated control law of a swept wing, and combining a swept wing transonic flutter flow field solver to analyze the flow field state, training the design model of the active and passive coordinated control law of the swept wing, obtaining the optimal design model of the active and passive coordinated control law of the swept wing, and performing active and passive control on the swept wing to suppress the shock wave flutter of the swept wing. The present invention adopts the idea of ​​active and passive coordinated control, and effectively reduces the amplitude of lift pulsation through the mutual cooperation of an active control device and a passive control device. At the same time, the present invention adopts deep reinforcement learning control, which can eliminate the shock wave flutter phenomenon of the swept wing within a certain range of flow states.

Description

Technical Field

[0001] The present invention belongs to the field of active and passive coordinated control of aircraft flow, and specifically is an active and passive coordinated control method of swept wing transonic shock wave flutter pulsation load based on deep reinforcement learning. Background Art

[0002] The aircraft has high flight efficiency in the transonic stage, but when the flow state enters the flutter boundary composed of Mach number and angle of attack, transonic flutter will occur on the wing surface. This is a flow instability phenomenon caused by the interference of the shock wave boundary layer. It is manifested as a large shaking of the shock wave on the wing surface, causing a large pulsation of the lift response. The oscillation of the shock wave has nothing to do with whether the aircraft structure is moving, and is generally considered to be a global flow instability phenomenon. The unstable starting boundary of the flutter is closely related to the incoming flow Mach number (0.7-0.9) and the angle of attack (0°-9°). As the Mach number increases, the starting angle of the flutter gradually decreases.

[0003] Flow control is an important means to eliminate the adverse effects of transonic buffeting. Researchers have designed a variety of transonic buffeting control schemes to eliminate the pulsating load of transonic buffeting and improve flow stability. At present, the general concern is the shock wave boundary layer interference and the sudden change of the trailing edge flow. Therefore, regulating the flow at the shock root and trailing edge has become the main design goal of the control scheme. Most of this control scheme uses a single control device, which can be divided into passive control devices and active control devices according to whether energy injection is required. Passive shock wave suppression bulges and active trailing edge rudders are the most widely used control devices.

[0004] The current passive control method can improve the flutter boundary and eliminate the flutter load under the design condition, but its shape parameters cannot be changed, which will cause it to fail or even have the opposite effect under non-design conditions. In active control, the commonly used closed-loop trailing edge control law is difficult to design and has a high dimension of parameter optimization, which limits its application in engineering practice. At the same time, wing flutter not only manifests as the chord-wise oscillation of the shock wave, but also manifests as the irregular shaking of the shock wave in the span direction, making the control design of the wing transonic flutter more difficult. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides an active and passive collaborative control method for the transonic shock wave buffeting pulsation loads of a swept wing based on deep reinforcement learning. The shape parameters and actuation laws of the optimized active and passive control devices are obtained by training the deep reinforcement learning model to achieve a better control effect than a single control device, thereby solving the problem that a single control device is currently difficult to achieve a better transonic buffeting control effect for a swept wing.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: an active and passive control method for suppressing the shock wave flutter of a swept wing, comprising the following steps:

[0007] S1: Establish the numerical simulation geometric model of the swept wing;

[0008] S2: Establish a design model for active and passive coordinated control laws for swept wings;

[0009] S3: According to the numerical simulation geometric model of the swept wing, the swept wing transonic buffeting flow field solver is used to analyze and numerically simulate the flow field state of the swept wing at the current moment;

[0010] S4: Input the current flow field state of the swept wing into the swept wing active and passive coordinated control law design model for training, and combine the swept wing transonic buffeting flow field solver and the numerical simulation geometric model of the swept wing to obtain the optimal swept wing active and passive coordinated control law design model;

[0011] S5: Use the optimal swept wing active and passive coordinated control law to design a model to perform active and passive control on the swept wing and suppress the shock wave flutter of the swept wing.

[0012] The beneficial effects of the present invention are as follows: the present invention adopts the active-passive collaborative control concept, utilizes the passive shock wave suppression bulge to suppress the spanwise effect of the swept wing transonic flutter, and then ensures the control effect of the active trailing edge control to suppress the shock wave chord-wise sway. By actively and passively controlling the swept wing, the lift pulsation amplitude is effectively reduced, achieving a control effect that is better than when acting alone, and having little effect on the aerodynamic characteristics of the swept wing. At the same time, the present invention adopts deep reinforcement learning control to solve the multi-objective control problem, and has a good control effect on complex flow systems, strong robustness, and can eliminate the swept wing shock wave flutter phenomenon within a certain range of flow states.

[0013] Furthermore, the establishment of a numerical simulation geometric model of the swept wing specifically includes:

[0014] Set the geometry and dimensions of the swept wing;

[0015] Meshing the swept wing geometry model;

[0016] Setting the position and size parameters of the control device in the swept wing and applying the control device by modal interpolation method;

[0017] Configure the numerical solver for the swept wing and set the boundary conditions, physical model, and numerical algorithm for the numerical simulation of the swept wing.

[0018] The beneficial effects of the above further scheme are: by establishing a numerical simulation geometric model of the swept wing, it is possible to simulate and analyze the aerodynamic characteristics of the swept wing under transonic conditions, accurately predict and evaluate the flight performance of the swept wing under different controls, and facilitate the intelligent agent to learn the optimal control strategy.

[0019] Further: the specific method of establishing the active and passive coordinated control law design model of the swept wing is as follows:

[0020] Set up the agent, determine the state of the agent as the lift coefficient and moment coefficient obtained by solving the flow field, determine the action of the agent as the deflection angle of the rudder surface and the shape parameters of the passive bulge of the swept wing, and determine the reward function of the agent;

[0021] Initialize the intelligent agent and obtain the initialization swept wing active and passive cooperative control law design model;

[0022] The updating strategy for initializing the active and passive coordinated control law design model of the swept wing is determined, and the active and passive coordinated control law design model of the swept wing is obtained.

[0023] The beneficial effects of the above further scheme are: by establishing a swept-wing active and passive collaborative control law design model, it is able to handle complex flow environments and automatically adjust the control device to suppress buffeting. At the same time, the swept-wing active and passive collaborative control law design model can learn in a constantly changing flight environment and eliminate the swept-wing shock wave buffeting phenomenon within a certain range of flow states, and has strong robustness.

[0024] Further: The expression of the reward function of the agent is as follows:

[0025]

[0026]

[0027]

[0028]

[0029] in, is the agent’s reward function, is the pulsation amplitude of the wing lift coefficient response, is the change in the wing lift coefficient response at adjacent moments, is the action amplitude, , and are weights, for The lift coefficient of the swept wing is, is the change of the wing lift coefficient at adjacent moments, for to The average value of the wing lift coefficient at time is the change of the bulge parameter relative to the initial parameter and The deflection amount of the active control surface at any moment.

[0030] The beneficial effects of the above further scheme are: by setting the reward function, the intelligent agent is guided to learn in the direction of reducing the impact of vibration and optimizing aerodynamic performance, and the intelligent agent is helped to maintain stable performance when facing different flight conditions, thereby enhancing the robustness of the control strategy.

[0031] Further: the intelligent agent includes a policy network, a target network corresponding to the policy network, an evaluation network, and a target network corresponding to the evaluation network, and the policy network, the target network corresponding to the policy network, the evaluation network, and the target network corresponding to the evaluation network are all fully connected networks; the fully connected network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence;

[0032] The initialized swept wing active and passive collaborative control law design model includes a strategy network, a target network corresponding to the strategy network, a first evaluation network, a target network corresponding to the first evaluation network, a second evaluation network, and a fourth target network corresponding to the second evaluation network.

[0033] The beneficial effects of the above further scheme are as follows: the present invention designs a specific intelligent agent network structure and provides a clear intelligent agent learning framework, which can effectively capture the nonlinear characteristics and dynamic changes in the flutter control of the swept wing, improve the learning efficiency of the intelligent agent, and output more accurate control decisions, effectively suppressing the shock wave flutter of the swept wing.

[0034] Further: the specific steps of S4 are as follows:

[0035] S41: according to the flow field state of the swept wing at the current moment, the state value is calculated by using the design model of the active and passive cooperative control law of the swept wing;

[0036] S42: outputting the actuation law of the control device at the next moment through the strategy network of the swept wing active and passive coordinated control law design model according to the state value, and inputting it into the numerical simulation geometric model of the swept wing to obtain the flow field state of the swept wing at the next moment;

[0037] S43: according to the actuation law of the control device at the next moment, the evaluation network of the swept wing active and passive coordinated control law design model is used for evaluation, and the evaluation result is transmitted back to the swept wing transonic buffeting flow field solver, and the hyper parameters of the evaluation network are updated by calculating the TD-error time difference error;

[0038] S44: Calculate the reward function of the swept wing active and passive collaborative control law design model, and determine whether the reward function converges. If so, obtain the optimal swept wing active and passive collaborative control law design model. Otherwise, return to S42.

[0039] The beneficial effect of the above further scheme is that through the swept-wing transonic buffeting flow field solver and the numerical simulation geometric model of the swept-wing, accurate environmental feedback and physical reality mapping can be provided for the training of the swept-wing active and passive collaborative control law design model, thereby improving the training efficiency and the accuracy of the control performance.

[0040] Further: the hyperparameters of the evaluation network are updated by calculating the TD-error temporal difference error, and the expression is as follows:

[0041]

[0042]

[0043] in, TD-error is the timing differential error. is the target value obtained according to the target network corresponding to the evaluation network, is the first sampled from the experience replay pool Group data, To evaluate the output value of the network, is the output value of the target network corresponding to the evaluation network, are the labels of the two evaluation networks, is the output value of the target policy network, is the discount factor, To satisfy the normal distribution of noise, For a normal distribution with truncation, is a normal distribution, is the variance of the normal distribution, is the amplitude of the truncated normal distribution.

[0044] The beneficial effects of the above further scheme are: by calculating the TD-error temporal difference error, the learning efficiency of the intelligent agent can be improved, the stability can be improved, and the hyperparameters of the evaluation network can be optimized, so that the intelligent agent can learn more effectively in a complex environment.

[0045] Further: the design model of the optimal swept wing active and passive coordinated control law is used to perform active and passive control on the swept wing, specifically:

[0046] The closed-loop trailing edge control surface is used to control the chord-wise oscillation of the shock wave; and under transonic buffeting conditions, the pressure pulsation at the wing tip is severe, and the active trailing edge control surface is set at the first percentage threshold span of the swept wing, and the chord-wise length is the second percentage threshold of the cross-sectional chord length, and the incoming flow angle of attack and the control surface deflection angle are actively set to change;

[0047] The passive shock wave suppression bulge is used to suppress the spanwise sway of the shock wave. The shape of the passive bulge is expressed by a Hicks-Henne type function, and its expression is as follows:

[0048]

[0049]

[0050]

[0051] in, is the chordal position of the point on the drum relative to the drum, is the spanwise position of a point on the bulge relative to the bulge, is the bulge height, is the chord-wise bulge shape function, is the spanwise bulge shape function, is the mth power, It is the ratio of the chordal position of the highest point of the bulge to the chordal length of the bulge. It is the ratio of the spanwise position of the highest point of the bulge to the spanwise length of the bulge. is the relative chord-wise coordinate of the point on the bulge, is the spanwise coordinate of the point on the bulge relative to the wing, is the starting chordal position of the drum, is the starting spanwise position of the bulge, is the chordal length of the drum, is the span length of the bulge.

[0052] The beneficial effect of the above further scheme is: the present invention utilizes passive shock wave suppression bulge to suppress the spanwise effect of the swept wing transonic flutter, and utilizes closed-loop control surface control to suppress the control effect of the shock wave chordwise shaking. The two methods cooperate with each other to achieve a control effect that is better than when each acts alone. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A flow chart of an active and passive control method for suppressing shock wave flutter of a swept wing;

[0054] Figure 2 Design a system block diagram for active and passive coordinated control of shock wave buffeting of swept wings;

[0055] Figure 3This is the effect diagram of using only the active control method based on the trailing edge control surface;

[0056] Figure 4 This is the effect diagram of using only the passive control method based on shock wave suppression bulge;

[0057] Figure 5 This is a diagram showing the effect of using both the active control method based on the trailing edge control surface and the passive control method based on the shock wave suppression bulge. DETAILED DESCRIPTION

[0058] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0059] like Figure 1 As shown, a flow chart of an active and passive control method for suppressing the shock wave flutter of a swept wing includes the following steps:

[0060] S1: Establish the numerical simulation geometric model of the swept wing;

[0061] S2: Establish a design model for active and passive coordinated control laws for swept wings;

[0062] S3: According to the numerical simulation geometric model of the swept wing, the swept wing transonic buffeting flow field solver is used to analyze and numerically simulate the flow field state of the swept wing at the current moment;

[0063] S4: Input the current flow field state of the swept wing into the swept wing active and passive coordinated control law design model for training, and combine the swept wing transonic buffeting flow field solver and the numerical simulation geometric model of the swept wing to obtain the optimal swept wing active and passive coordinated control law design model;

[0064] S5: Use the optimal swept wing active and passive coordinated control law to design a model to perform active and passive control on the swept wing and suppress the shock wave flutter of the swept wing.

[0065] In one embodiment of the present invention, the numerical simulation geometric model of the swept wing is established in S1, specifically including:

[0066] Set the geometry and dimensions of the swept wing;

[0067] The swept wing geometric model is meshed. The structured and unstructured hybrid mesh generation method can be used to generate the uncontrolled swept wing surface and computational domain space mesh. The RBF dynamic mesh method is used to achieve adaptive changes in meshing when the control mode changes.

[0068] Setting the position and size parameters of the control devices in the swept wing, including determining the number of passive bulge control devices, the relevant initial parameters in the position and shape control function, the position, chord-wise and span-wise lengths of the active control surface control devices, and the position of the shaft, and constructing the passive bulge and active control surface modes based on the modal interpolation method, so as to facilitate the subsequent application of the control law;

[0069] A numerical solver for the swept wing is configured, and boundary conditions, physical models and numerical algorithms for numerical simulation of the swept wing are set. In the present invention, in order to reduce the time loss of solving the three-dimensional swept wing flow field and improve the training efficiency of the active and passive coordinated control law design model of the swept wing, the control equation can be selected as the unsteady Reynolds averaged Navier-Stokes equation, and its expression is as follows:

[0070]

[0071] in, is a conserved variable, is the inviscid flux, For a viscous flux, is the grid speed, For the control body, is the control volume boundary, is the unit vector of the normal direction outside the control volume boundary, is the surface integral, is the volume fraction, It is the acronym of the word inviscid in the inviscid flux. It is the abbreviation of the word viscid in viscous flux. The turbulence model used to solve the unsteady Reynolds-averaged Navier-Stokes equations is the SST model. The finite volume method is used for discretization. The spatial discretization format is the "AUSM+UP" format, and the time advancement method is the dual time advancement method.

[0072] After establishing the numerical simulation geometric model of the swept wing, the simulation calculation results are compared with the results of the same research object in the existing literature to ensure the accuracy of the digital simulation geometric model and the numerical calculation method.

[0073] In one embodiment of the present invention, a design model of active and passive coordinated control law of swept wing is established in S2, and the specific method is as follows:

[0074] S21: Setting an agent, the agent includes a policy network, a target network corresponding to the policy network, an evaluation network, and a target network corresponding to the evaluation network, the policy network, the target network corresponding to the policy network, the evaluation network, and the target network corresponding to the evaluation network are all fully connected networks; the fully connected network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence;

[0075] Among them, the number of neurons in each layer of the policy network and the target network corresponding to the policy network is "state dimension-128-256-64-action dimension", which means that the number of neurons in the input layer of the policy network and the target network corresponding to the policy network corresponds to the state space dimension, the number of neurons in the first hidden layer is 128, the number of neurons in the second hidden layer is 256, the number of neurons in the third hidden layer is 64, and the number of neurons in the output layer corresponds to the action space dimension; the number of neurons in each layer of the evaluation network and the target network corresponding to the evaluation network is "state dimension+action dimension-128-256-128-1", which means that the number of neurons in the input layer of the evaluation network and the target network corresponding to the evaluation network corresponds to the state space dimension plus the action space dimension, the number of neurons in the first hidden layer is 128, the number of neurons in the second hidden layer is 256, the number of neurons in the third hidden layer is 128, and the number of neurons in the output layer is 1;

[0076] The state of the intelligent agent is determined as the lift coefficient and moment coefficient obtained by solving the flow field, the action of the intelligent agent is determined as the deflection angle of the rudder surface and the shape parameters of the passive bulge of the swept wing, and the reward function of the intelligent agent is determined. The expression of the reward function of the intelligent agent is as follows:

[0077]

[0078]

[0079]

[0080]

[0081] in, is the agent’s reward function, is the pulsation amplitude of the wing lift coefficient response, is the change in the wing lift coefficient response at adjacent moments, is the action amplitude, , and are weights, for The lift coefficient of the swept wing is, is the change of the wing lift coefficient at adjacent moments, for to The average value of the wing lift coefficient at time is the change of the bulge parameter relative to the initial parameter and The deflection amount of the active rudder surface at any moment; the weight can be set as: , , .

[0082] S22: Initialize the intelligent agent to obtain an initialized swept wing active and passive cooperative control law design model; the dual-delay deep deterministic policy gradient TD3 algorithm can be used for initialization;

[0083] The initialization swept wing active and passive collaborative control law design model includes a strategy network, a target network corresponding to the strategy network, a first evaluation network, a target network corresponding to the first evaluation network, a second evaluation network, and a fourth target network corresponding to the second evaluation network.

[0084] S23: determining an update strategy for initializing the swept wing active and passive coordinated control law design model, and obtaining the swept wing active and passive coordinated control law design model;

[0085] Among them, the update strategy of the active and passive coordinated control law design model of the initial swept wing is as follows:

[0086] The evaluation network adopts the TD-error update method based on the reward function; the policy network adopts the gradient update method; both the evaluation network and the policy network adopt asynchronous update strategies to ensure the stability of the policy network and the convergence of the reinforcement learning model.

[0087] In one embodiment of the present invention, the writing language of the intelligent agent is Python, and the language environment of the swept-wing transonic buffeting flow field solver is Fortran. A data transfer interface between the two languages ​​can be written to realize the interaction between the intelligent agent and the flow field solver; then, according to the numerical simulation geometric model of the swept-wing, the swept-wing transonic buffeting flow field solver is used to perform analysis, and the flow field state of the swept-wing at the current moment is numerically simulated.

[0088] In one embodiment of the present invention, the specific steps of S4 are as follows:

[0089] S41: according to the flow field state of the swept wing at the current moment, the state value is calculated by using the design model of the active and passive cooperative control law of the swept wing;

[0090] S42: outputting the action law of the control device at the next moment through the strategy network of the swept wing active and passive coordinated control law design model according to the state value, and inputting it into the numerical simulation geometric model of the swept wing to obtain the flow field state of the swept wing at the next moment;

[0091] S43: According to the actuation law of the control device at the next moment, the evaluation network of the swept wing active and passive coordinated control law design model is used for evaluation, and the evaluation result is transmitted back to the swept wing transonic buffeting flow field solver. The hyperparameters of the evaluation network are updated by calculating the TD-error time difference error. The expression is as follows:

[0092]

[0093]

[0094] in, TD-error is the timing differential error. is the target value obtained according to the target network corresponding to the evaluation network, is the first sampled from the experience replay pool Group data, To evaluate the output value of the network, is the output value of the target network corresponding to the evaluation network, are the labels of the two evaluation networks, is the output value of the target policy network, is the discount factor, To satisfy the normal distribution of noise, For a normal distribution with truncation, is a normal distribution, is the variance of the normal distribution, is the amplitude of the truncated normal distribution.

[0095] S44: Calculate the reward function of the swept wing active and passive collaborative control law design model, and determine whether the reward function converges. If so, obtain the optimal swept wing active and passive collaborative control law design model. Otherwise, return to S42.

[0096] In one embodiment of the present invention, Figure 2 As shown, the swept wing is actively and passively controlled by using the optimal swept wing active and passive coordinated control law design model to suppress the shock wave buffeting of the swept wing, including: an active control method based on the trailing edge control surface and a passive control method based on the shock wave suppression bulge, respectively using the closed-loop trailing edge control surface to control the chord-wise oscillation of the shock wave, and using the passive shock wave suppression bulge to suppress the span-wise swaying of the shock wave. The parameter settings of the two control devices are as follows:

[0097] Active control method based on trailing edge control surface: Use closed-loop trailing edge control surface to control the chord-wise oscillation of shock wave. Under transonic buffeting conditions, the pressure pulsation at the wing tip is severe. The active trailing edge control surface is set at the first percentage threshold span of the swept wing, and the chord-wise length is the second percentage threshold of the cross-sectional chord length. The first percentage threshold is set to 80%-100%, and the second percentage threshold is set to 15%. The angle of attack of the incoming flow is actively changed. and rudder angle , and its control law design model is a fully trained deep reinforcement learning model.

[0098] Passive control method based on shock wave suppression bulge: The passive shock wave suppression bulge is used to suppress the spanwise sway of the shock wave. The shape of the passive bulge is described by the commonly used Hicks-Henne type function:

[0099]

[0100]

[0101]

[0102] in, is the chordal position of the point on the drum relative to the drum, is the spanwise position of a point on the bulge relative to the bulge, is the bulge height, is the chord-wise bulge shape function, is the spanwise bulge shape function, is the mth power, It is the ratio of the chordal position of the highest point of the bulge to the chordal length of the bulge. It is the ratio of the spanwise position of the highest point of the bulge to the spanwise length of the bulge. is the relative chord-wise coordinate of the point on the bulge, is the spanwise coordinate of the point on the bulge relative to the wing, is the starting chordal position of the drum, is the starting spanwise position of the bulge, is the chordal length of the drum, is the span length of the bulge;

[0103] In this embodiment, a bulge array consisting of four bulges is used as a passive control method, and the shape parameters of the bulges are given based on experience; the starting chord positions of the four bulges are , chord length and span length are the same, respectively , and ,in is the chord length of each section of the wing, is the wing span, the starting position of the four bulges in the span They are , , and The bulge is symmetrical, so The bulge height is , is the average aerodynamic chord length of the wing.

[0104] like Figure 3 As shown in the figure, it is the effect diagram of using only the active control method based on the trailing edge control surface. The figure shows the effect of single trailing edge control surface control of the swept wing. The single trailing edge control surface control can only suppress the pulsation amplitude of the lift coefficient response of the swept wing by about 50%.

[0105] like Figure 4 As shown in the figure, it is the effect diagram of using only the passive control method based on shock wave suppression bulge. The figure shows the control effect of the initial passive shock wave suppression bulge array. It can be seen from the figure that the passive bulge array can adjust the high-frequency shock wave flutter phenomenon to a lower frequency and more regular state, which means that the passive bulge can effectively suppress the spanwise shaking phenomenon of the shock wave, but the suppression effect on the pulsation amplitude of the lift coefficient of the swept wing is not good, and the pulsation amplitude of the lift coefficient of the swept wing can hardly be suppressed.

[0106] like Figure 5 As shown in the figure, it is the effect diagram of using the active control method based on the trailing edge control surface and the passive control method based on the shock wave suppression bulge at the same time. The figure shows the control effect of the active and passive cooperative control law design model of the swept wing. It can be seen from the figure that the active and passive cooperative control can reduce the pulsation amplitude of the swept wing lift coefficient by 87%, achieving the best control effect that is better than that achieved when the two control devices are actuated separately, and improving the average lift coefficient of the wing after the control is completed.

[0107] The beneficial effects of the present invention are as follows: the present invention adopts the active-passive collaborative control concept, utilizes the passive shock wave suppression bulge to suppress the spanwise effect of the swept wing transonic flutter, and then ensures the control effect of the active trailing edge control to suppress the shock wave chord-wise shaking. Through the mutual cooperation of the active control device and the passive control device, the lift pulsation amplitude is effectively reduced, and the control effect is better than that when acting alone, and the aerodynamic characteristics of the swept wing are less affected. At the same time, the present invention adopts deep reinforcement learning control to solve the multi-objective control problem, and has a good control effect on complex flow systems, strong robustness, and can eliminate the swept wing shock wave flutter phenomenon within a certain range of flow states.

Claims

1. An active and passive control method for suppressing the shock wave flutter of a swept wing, characterized in that: The following steps are involved: S1: Establish the numerical simulation geometric model of the swept wing; S2: Establish a design model for active and passive coordinated control laws for swept wings; S3: According to the numerical simulation geometric model of the swept wing, the swept wing transonic buffeting flow field solver is used to analyze and numerically simulate the flow field state of the swept wing at the current moment; S4: Input the current flow field state of the swept wing into the swept wing active and passive coordinated control law design model for training, and combine the swept wing transonic buffeting flow field solver and the numerical simulation geometric model of the swept wing to obtain the optimal swept wing active and passive coordinated control law design model; The specific steps of S4 are as follows: S41: according to the flow field state of the swept wing at the current moment, the state value is calculated by using the design model of the active and passive cooperative control law of the swept wing; S42: outputting the actuation law of the control device at the next moment through the strategy network of the swept wing active and passive coordinated control law design model according to the state value, and inputting it into the numerical simulation geometric model of the swept wing to obtain the flow field state of the swept wing at the next moment; S43: according to the actuation law of the control device at the next moment, the evaluation network of the swept wing active and passive coordinated control law design model is used for evaluation, and the evaluation result is transmitted back to the swept wing transonic buffeting flow field solver, and the hyper parameters of the evaluation network are updated by calculating the TD-error time difference error; S44: Calculate the reward function of the swept wing active and passive cooperative control law design model, and determine whether the reward function converges. If so, obtain the optimal swept wing active and passive cooperative control law design model. Otherwise, return to S42. S5: Use the optimal swept wing active and passive coordinated control law design model to perform active and passive control on the swept wing and suppress the shock wave buffeting of the swept wing; The design model of the optimal swept wing active and passive coordinated control law is used to perform active and passive control on the swept wing, specifically: The closed-loop trailing edge control surface is used to control the chord-wise oscillation of the shock wave; and under transonic buffeting conditions, the pressure pulsation at the wing tip is severe, and the active trailing edge control surface is set at the first percentage threshold span of the swept wing, and the chord-wise length is the second percentage threshold of the cross-sectional chord length, and the incoming flow angle of attack and the control surface deflection angle are actively set to change; The passive shock wave suppression bulge is used to suppress the spanwise sway of the shock wave. The shape of the passive bulge is expressed by a Hicks-Henne type function, and its expression is as follows: in, is the chordal position of the point on the drum relative to the drum, is the spanwise position of a point on the bulge relative to the bulge, is the bulge height, is the chord-wise bulge shape function, is the spanwise bulge shape function, is the mth power, It is the ratio of the chordal position of the highest point of the bulge to the chordal length of the bulge. It is the ratio of the spanwise position of the highest point of the bulge to the spanwise length of the bulge. is the relative chord-wise coordinate of the point on the bulge, is the spanwise coordinate of the point on the bulge relative to the wing, is the starting chordal position of the drum, is the starting spanwise position of the bulge, is the chordal length of the drum, is the span length of the bulge.

2. The active and passive control method for suppressing the shock wave buffeting of a swept wing according to claim 1, characterized in that: The establishment of the numerical simulation geometric model of the swept wing specifically includes: Set the geometry and dimensions of the swept wing; Meshing the swept wing geometry model; Setting the position and size parameters of the control device in the swept wing and applying the control device by modal interpolation method; Configure the numerical solver for the swept wing and set the boundary conditions, physical model, and numerical algorithm for the numerical simulation of the swept wing.

3. The active and passive control method for suppressing the shock wave buffeting of a swept wing according to claim 1, characterized in that: The specific method of establishing the active and passive coordinated control law design model of the swept wing is as follows: Set up the agent, determine the state of the agent as the lift coefficient and moment coefficient obtained by solving the flow field, determine the action of the agent as the deflection angle of the rudder surface and the shape parameters of the passive bulge of the swept wing, and determine the reward function of the agent; Initialize the intelligent agent and obtain the initialization swept wing active and passive cooperative control law design model; The updating strategy for initializing the active and passive coordinated control law design model of the swept wing is determined, and the active and passive coordinated control law design model of the swept wing is obtained.

4. The active and passive control method for suppressing the shock wave buffeting of a swept wing according to claim 3, characterized in that: The expression of the agent's reward function is as follows: in, is the agent’s reward function, is the pulsation amplitude of the wing lift coefficient response, is the change in the wing lift coefficient response at adjacent moments, is the action amplitude, , and are weights, for The lift coefficient of the swept wing is, is the change of the wing lift coefficient at adjacent moments, for to The average value of the wing lift coefficient at time is the change of the bulge parameter relative to the initial parameter and The deflection amount of the active control surface at any moment.

5. The active and passive control method for suppressing the shock wave buffeting of a swept wing according to claim 3, characterized in that: The intelligent agent includes a policy network, a target network corresponding to the policy network, an evaluation network, and a target network corresponding to the evaluation network, wherein the policy network, the target network corresponding to the policy network, the evaluation network, and the target network corresponding to the evaluation network are all fully connected networks; the fully connected network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence; The initialized swept wing active and passive collaborative control law design model includes a strategy network, a target network corresponding to the strategy network, a first evaluation network, a target network corresponding to the first evaluation network, a second evaluation network, and a fourth target network corresponding to the second evaluation network.

6. The active and passive control method for suppressing the shock wave buffeting of a swept wing according to claim 1, characterized in that: The hyperparameters of the evaluation network are updated by calculating the TD-error temporal difference error, and the expression is as follows: in, TD-error is the timing differential error. is the target value obtained according to the target network corresponding to the evaluation network, is the first sampled from the experience replay pool Group data, To evaluate the output value of the network, is the output value of the target network corresponding to the evaluation network, are the labels of the two evaluation networks, is the output value of the target policy network, is the discount factor, To satisfy the normal distribution of noise, For a normal distribution with truncation, is a normal distribution, is the variance of the normal distribution, is the amplitude of the truncated normal distribution.

Images