Close-range air combat model construction method
Through the forward model construction method and PID control method, a close-range air combat model is constructed based on the energy maneuver curve, which solves the problems of sparse reward function and unfavorable model establishment in the existing technology, and realizes a close-range air combat model with strong interpretability and excellent maneuverability.
Patent Information
- Application Number
- CN202411828238.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-12
- Publication Date
- 2025-04-11
AI Technical Summary
In the close air combat, existing intelligent air combat technology has problems such as sparse reward functions, poor model establishment, inequality of strategy distribution, easy dispersion of agent behavior, and insufficient interpretability and security guarantees.
The forward model construction method is adopted, and typical state points are selected based on the typical energy maneuver curve of the aircraft, and a close-range air combat maneuvering process is constructed. The maneuver strategy and control model are constructed through air combat situation judgment and PID control methods, and iterative optimization is used for simulation.
It has realized the construction of a close-range air combat model with strong interpretability, avoided the problem of complex maneuver strategy selection, fully utilized the maneuverability of the fighter, and was suitable for air combat teaching and training and drone air combat.
Smart Images
Figure CN119808536A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of autonomous air combat decision-making, and in particular to a method for constructing a close-range air combat model. Background Art
[0002] With the advent of the era of intelligent air combat, neural networks and reinforcement learning have become a hot area of research in intelligent air combat in recent years. Based on the reward and punishment function, the intelligent agent obtains the optimization strategy and evolves into an air combat model through a large number of simulation trials and errors. Although the intelligent agent obtained by reinforcement learning can innovate air combat tactics and compete with human pilots, there are still some shortcomings: reinforcement learning has limited use of the experience of expert rules, and it is difficult to converge quickly when the reward function is sparse; the model obtained by reinforcement learning training is not conducive to the efficient establishment of a model close to air combat training; when the strategy distribution is not unique, the level of the obtained intelligent agent is low and the behavior is easy to diverge; there are problems such as interpretability, security, and inconsistency between the simulation environment and reality that hinder practical application.
[0003] Judging from the publicly available information, the current research on close-range intelligent air combat focuses on theoretical and algorithmic research, situation assessment and decision-making, and lacks consideration of energy maneuverability. How to make full use of the experience of expert rules, build a close-range air combat model based on the energy maneuverability characteristics of the aircraft, and maximize the maneuverability of the aircraft, whether for air combat teaching and training, or for future application in drone air combat, has research value. Summary of the invention
[0004] The purpose of the present invention is to provide a method for constructing a close-range air combat model. The method of the present invention is a forward model construction method with strong interpretability. By constructing the model forward, the complex maneuver strategy optimization problem is avoided.
[0005] Technical solution. A method for constructing a close-range air combat model comprises the following steps:
[0006] Step 1: Based on the typical energy maneuvering curve of the aircraft, select typical state points and construct the close-range air combat maneuvering process;
[0007] Step 2: Based on the position and speed relationship between our aircraft and the enemy aircraft, make an air combat situation judgment; based on the air combat situation judgment in step 2 and the close air combat maneuvering process in step 1, construct a maneuvering strategy based on energy maneuvering under various situations;
[0008] Step 3: Based on the maneuvering strategy in step 2, use the PID control method to build the aircraft's lateral and longitudinal control models.
[0009] The aforementioned close-range air combat model construction method also includes step 4: using simulation to iteratively optimize the PID control parameters in step 3 and each typical state point in step 1.
[0010] In the aforementioned close-range air combat model construction method, in step 1, the construction process of the close-range air combat maneuver history is as follows:
[0011] Step 11. Select typical state points: on the typical energy maneuvering curve, select the point corresponding to the entry speed of the aircraft entering the air combat, the point corresponding to the maximum normal overload at the entry speed, the point corresponding to the maximum circling angular rate at the corner point speed, the point corresponding to the minimum air combat speed, the point corresponding to the circling angular rate of 0 at the minimum air combat speed, and the point corresponding to the maximum stable circling angular rate; the unit residual power of the point corresponding to the minimum air combat speed is 0;
[0012] Step 12. Match the typical maneuvering process of the aircraft with the process between typical state points to obtain the close-range air combat maneuvering process.
[0013] In the aforementioned close-range air combat model construction method, in step 1, on the typical energy maneuvering curve, point 1 is selected as the entry speed of the air combat; point 2 is the point corresponding to the maximum normal overload at the entry speed, and from point 1 to point 2, the aircraft establishes the maximum normal overload; point 3 is the point of the maximum circling angular rate corresponding to the corner point speed; point 4 is the minimum air combat speed, and its unit residual power is 0; point 5 is the point where the circling angular rate is 0 at the minimum air combat speed, and point 6 is the point of the maximum stable circling angular rate; then the typical maneuvering process of close-range air combat is as follows:
[0014]
[0015]
[0016] In the aforementioned close-range air combat model construction method, in step 2, the air combat situation judgment includes static situation judgment, and the judgment process is as follows:
[0017] Calculate the angle between the vector of our aircraft looking at the enemy aircraft and the direction of the nose, that is, the sight angle θ LOS ; The angle between our aircraft's vector relative to the enemy aircraft and the direction of the enemy aircraft's nose, that is, the target offset angle θ TDA :
[0018]
[0019] Among them, V A 、V T Respectively represent the velocity vectors of our aircraft and enemy aircraft, V P The vector indicating the position of our aircraft pointing to the position of the enemy aircraft;
[0020] Define our situation function f1(θ LOS , D):
[0021]
[0022] In the formula, D is the distance between two machines, θ M1 D is the maximum off-axis angle allowed for our close-range combat missile to be launched when it meets the guidance. Mmax1 is the maximum range of our close combat missile in the current state, D Mmin1 is the minimum range; 0 means the enemy is outside our close-range combat missile off-axis launch angle; 1 means the enemy is within our close-range combat missile off-axis launch angle and within the range, and can launch missiles to attack; 2 means the enemy is within our close-range combat missile off-axis launch angle but outside the range; 3 means the enemy is within our close-range combat missile off-axis launch angle but outside the range;
[0023] Similarly, the enemy situation function f2(θ TDA , D) are as follows:
[0024]
[0025] In the formula, θ M2 D is the maximum off-axis launch angle allowed when the enemy close-range combat projectile meets the guidance requirements. Mmax2 is the maximum range of the enemy's close combat projectile in the current state, D Mmin2 is the minimum range;
[0026] According to f1(θ LOS , D) and f2(θ TDA , D), the close air combat situation at the preset moment is divided into seven types: attack, defense, mutual disadvantage, active, passive, balance and confrontation, and then the situation matrix is constructed.
[0027] In the aforementioned close-range air combat model construction method, in step 2, the air combat situation judgment also includes dynamic situation judgment, and the judgment process is as follows:
[0028] Based on the static situation, the line of sight angular rate θ is introduced LOS ′, the distance change rate D′ is used for decision-making maneuvers; the line of sight angular rate θ LOS ′ is positive, indicating the sight angle θ LOS If the distance change rate D′ is positive, the distance between the two aircraft is increasing, and if it is negative, the distance between the two aircraft is decreasing. ΔH is defined as the height difference between the enemy and us:
[0029]
[0030] Where H1 is the altitude of our aircraft, H2 is the altitude of the enemy aircraft, and h is a constant;
[0031] Based on the situation matrix, the line of sight angular velocity θ LOS ′, distance change rate D′ and the height difference ΔH between the enemy and us are used to construct the maneuver strategy based on energy maneuver under various situations.
[0032] In the aforementioned close-range air combat model construction method, in step 2, the maneuver strategy is as follows:
[0033]
[0034]
[0035] In the aforementioned close air combat model construction method, in step 3, the construction process of the longitudinal control model is as follows:
[0036] The expected speed is calculated by situation judgment and reinforcement learning model as the control input, and the actual flight speed V A As PID feedback, a speed controller is built to output the throttle control value δ T , thus achieving speed control; after situation judgment and reinforcement learning model calculation of expected overload as pitch control command input, with actual overload as feedback, the rudder control amount is calculated, and then the actual pitch angle, pitch angle velocity θ, θ′ are used as series PID feedback to build a pitch controller, and the output elevator deflection amount δ e , thereby achieving pitch angle control.
[0037] In the above-mentioned close-range air combat model construction method, when the throttle lever input is at the maximum, the speed still cannot reach the expected speed. T_real The correction value K is calculated and used to correct the pitch control command.
[0038] In the aforementioned close air combat model construction method, in step 3, the construction process of the lateral heading control model is as follows:
[0039] The expected sideslip angle β0 of the lateral sideslip control is kept at 0, and the expected roll angle is calculated by the situation judgment and reinforcement learning model. Then, the actual sideslip angle β is used as the input of the sideslip PID controller, and the actual roll angle and roll angular velocity φ, φ′ are used as the input of the roll PID controller to calculate the rudder deflection angle δ respectively. r and aileron deflection angle δ a , thereby achieving lateral heading control of the aircraft.
[0040] Beneficial effects: The method of the present invention is a forward model construction method with strong interpretability. By constructing the model forward, the complex maneuver strategy optimization problem is avoided. Based on the close-range air combat model construction method, energy maneuverability is fully considered, and the maneuverability of the fighter can be better exerted. It has application value for air combat teaching and training, drone air combat, etc. In addition, the model reservation ability combined with reinforcement learning facilitates the subsequent use of reinforcement learning methods to expand many-to-many complex air combat scenarios and tactical innovations. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a schematic diagram of a typical energy maneuvering curve;
[0042] Figure 2 It is the geometric relationship of close air combat;
[0043] Figure 3 is the static situation matrix;
[0044] Figure 4 To control the model longitudinally;
[0045] Figure 5 For model lateral heading control. DETAILED DESCRIPTION
[0046] Example 1. A method for constructing a close-range air combat model, see Figure 1-Figure 5 ,include:
[0047] Step 1: Based on the typical energy maneuvering curve of the aircraft, select typical state points and construct the close-range air combat maneuvering process;
[0048] Step 2: Based on the position and speed relationship between our aircraft and the enemy aircraft, see Figure 2 , make an air combat situation judgment; based on the air combat situation judgment in step 2 and the close air combat maneuvering process in step 1, construct a maneuvering strategy based on energy maneuvering under each situation;
[0049] Step 3: Based on the maneuvering strategy in step 2, use the PID control method to build the aircraft's lateral and longitudinal control models.
[0050] Step 4: Use simulation to iteratively optimize the PID control parameters in step 3 and the typical state points in step 1.
[0051] In step 1, the construction process of the close air combat maneuver history is as follows:
[0052] Step 11. Select typical state points: on the typical energy maneuvering curve, select the point corresponding to the entry speed of the aircraft entering the air combat, the point corresponding to the maximum normal overload at the entry speed, the point corresponding to the maximum circling angular rate at the corner point speed, the point corresponding to the minimum air combat speed, the point corresponding to the circling angular rate of 0 at the minimum air combat speed, and the point corresponding to the maximum stable circling angular rate; the unit residual power of the point corresponding to the minimum air combat speed is 0;
[0053] Step 12. Match the typical maneuvering process of the aircraft with the process between typical state points to obtain the close-range air combat maneuvering process.
[0054] See also Figure 1On the typical energy maneuvering curve, point 1 is selected as the entry speed of air combat; point 2 is the point corresponding to the maximum normal overload at the entry speed, and from point 1 to point 2, the aircraft establishes the maximum normal overload; point 3 is the point of the maximum circling angular rate corresponding to the corner speed; point 4 is the minimum speed of air combat, and its unit residual power is 0; point 5 is the point where the circling angular rate is 0 at the minimum speed of air combat, and point 6 is the point of the maximum stable circling angular rate; the typical maneuvering process of close-range air combat is as follows:
[0055]
[0056] The speeds at points 1 and 4 are not fixed values, and the initial values of the model can be roughly obtained based on the energy maneuverability characteristics of the fighter. Select several typical height values and perform interpolation processing at different heights.
[0057] In step 2, the air combat situation judgment includes static situation judgment, and the judgment process is as follows:
[0058] Calculate the angle between the vector of our aircraft looking at the enemy aircraft and the direction of the nose, that is, the sight angle θ LOS ; The angle between our aircraft's vector relative to the enemy aircraft and the direction of the enemy aircraft's nose, that is, the target offset angle θ TDA :
[0059]
[0060] Among them, V A 、V T Respectively represent the velocity vectors of our aircraft and enemy aircraft, V P The vector indicating the position of our aircraft pointing to the position of the enemy aircraft;
[0061] Define our situation function f1(θ LOS , D):
[0062]
[0063] In the formula, D is the distance between two machines, θ M1 D is the maximum off-axis angle allowed for our close-range combat missile to be launched when it meets the guidance. Mmax1 is the maximum range of our close combat missile in the current state, D Mmin1 is the minimum range; 0 means the enemy is outside our close-range combat missile off-axis launch angle; 1 means the enemy is within our close-range combat missile off-axis launch angle and within the range, and can launch missiles to attack; 2 means the enemy is within our close-range combat missile off-axis launch angle but outside the range; 3 means the enemy is within our close-range combat missile off-axis launch angle but outside the range;
[0064] Similarly, the enemy situation function f2(θ TDA , D) are as follows:
[0065]
[0066] In the formula, θ M2 D is the maximum off-axis launch angle allowed when the enemy close-range combat projectile meets the guidance requirements. Mmax2 is the maximum range of the enemy's close combat projectile in the current state, D Mmin2 is the minimum range;
[0067] According to f1(θ LOS , D) and f2(θ TDA , D), the close air combat situation at the preset time is divided into seven types: attack, defense, mutual disadvantage, active, passive, balance and confrontation, and then the situation matrix is constructed. Figure 3 .
[0068] In step 2, the air combat situation judgment also includes dynamic situation judgment, and the judgment process is as follows:
[0069] Based on the static situation, the line of sight angular rate θ is introduced LOS ′, the distance change rate D′ is used for decision-making maneuvers; the line of sight angular rate θ LOS ′ is positive, indicating the sight angle θ LOS If the distance change rate D′ is positive, the distance between the two aircraft is increasing, and if it is negative, the distance between the two aircraft is decreasing. ΔH is defined as the height difference between the enemy and us:
[0070]
[0071] Where H1 is the altitude of our aircraft, H2 is the altitude of the enemy aircraft, and h is a constant;
[0072] Based on the situation matrix, the line of sight angular velocity θ LOS ′, distance change rate D′ and the height difference ΔH between the enemy and us are used to construct the maneuver strategy based on energy maneuver under various situations.
[0073] In step 2, the maneuver strategy is as follows:
[0074]
[0075]
[0076] The construction process of step 3 is as follows: The longitudinal automatic flight control system is a system that automatically controls the longitudinal movement of the aircraft. The longitudinal controller includes a speed controller and an altitude controller. The speed controller controls the throttle lever of the engine by T To control the flight speed, the altitude controller controls the elevator rudder deflection angle δ eTo control the flight altitude. In this model, after situation judgment and reinforcement learning model calculation, the desired speed is obtained as the control input, and the actual flight speed V A As PID feedback, a speed controller is built, outputting δ T , thus achieving speed control; after situation judgment and reinforcement learning model calculation, the expected overload is obtained as the pitch control command input, and the actual overload is used as feedback to calculate the rudder control amount. Then, the actual pitch angle, pitch angle velocity θ, θ′ are used as series PID feedback to build a pitch controller, and the output δ e , thereby achieving pitch angle control. When the throttle stick input is at its maximum, the speed still cannot reach the desired speed. T_real The correction value K is calculated and used to correct the pitch control command, thereby further increasing the speed.
[0077] The lateral automatic flight control system is used to implement automatic control of the lateral movement of the aircraft, with the aileron as the main control object and the rudder used to eliminate the sideslip angle to ensure the coordinated movement during aileron control, that is, the sideslip angle is zero. This control with the aileron as the main control and the rudder as the auxiliary control is also called coordinated turning control.
[0078] The lateral heading controller includes a roll controller and a sideslip controller, and the control principle is basically the same as that of the longitudinal control. The desired sideslip angle β0 of the lateral heading sideslip control is kept at 0, and the expected roll angle value is calculated by the situation judgment and reinforcement learning model. Then, the actual sideslip angle β is used as the input of the sideslip PID controller, and the actual roll angle and roll angular velocity φ, φ′ are used as the input of the roll PID controller to calculate the rudder deflection angle δ respectively. r and aileron deflection angle δ a , thereby achieving lateral heading control of the aircraft.
Claims
1. A method for constructing a close air combat model, characterized in that: The steps include: Step 1: Based on the typical energy maneuvering curve of the aircraft, select typical state points and construct the close-range air combat maneuvering process; Step 2: Based on the position and speed relationship between our aircraft and the enemy aircraft, make an air combat situation judgment; based on the air combat situation judgment in step 2 and the close air combat maneuvering process in step 1, construct a maneuvering strategy based on energy maneuvering under various situations; Step 3: Based on the maneuvering strategy in step 2, use the PID control method to build the aircraft's lateral and longitudinal control models.
2. The close air combat model construction method according to claim 1, characterized in that: The method further includes step 4: iteratively optimizing the PID control parameters in step 3 and each typical state point in step 1 by using simulation.
3. The close air combat model construction method according to claim 1, characterized in that: In step 1, the construction process of the close air combat maneuver history is as follows: Step 11. Select typical state points: on the typical energy maneuvering curve, select the point corresponding to the entry speed of the aircraft entering the air combat, the point corresponding to the maximum normal overload at the entry speed, the point corresponding to the maximum circling angular rate at the corner point speed, the point corresponding to the minimum air combat speed, the point corresponding to the circling angular rate of 0 at the minimum air combat speed, and the point corresponding to the maximum stable circling angular rate; the unit residual power of the point corresponding to the minimum air combat speed is 0; Step 12. Match the typical maneuvering process of the aircraft with the process between typical state points to obtain the close-range air combat maneuvering process.
4. The close air combat model construction method according to claim 3 is characterized in that: In step 1, on the typical energy maneuvering curve, point 1 is selected as the entry speed of air combat; point 2 is the point corresponding to the maximum normal overload at the entry speed, and from point 1 to point 2, the aircraft establishes the maximum normal overload; point 3 is the point of the maximum circling angular rate corresponding to the corner point speed; point 4 is the minimum speed of air combat, and its unit residual power is 0; point 5 is the point where the circling angular rate is 0 at the minimum speed of air combat, and point 6 is the point of the maximum stable circling angular rate; then the typical maneuvering process of close-range air combat is as follows:
5. The close air combat model construction method according to claim 1, characterized in that: In step 2, the air combat situation judgment includes static situation judgment, and the judgment process is as follows: Calculate the angle between the vector of our aircraft looking at the enemy aircraft and the direction of the nose, that is, the sight angle θ LOS ; The angle between our aircraft's vector relative to the enemy aircraft and the direction of the enemy aircraft's nose, that is, the target offset angle θ TDA : Among them, V A 、V T Respectively represent the velocity vectors of our aircraft and enemy aircraft, V P The vector indicating the position of our aircraft pointing to the position of the enemy aircraft; Define our situation function f1(θ LOS , D): In the formula, D is the distance between two machines, θ M1 D is the maximum off-axis angle allowed for our close-range combat missile to be launched when it meets the guidance. Mmax1 is the maximum range of our close combat missile in the current state, D Mmin1 is the minimum range; 0 means the enemy is outside our close-range combat missile off-axis launch angle; 1 means the enemy is within our close-range combat missile off-axis launch angle and within the range, and can launch missiles to attack; 2 means the enemy is within our close-range combat missile off-axis launch angle but outside the range; 3 means the enemy is within our close-range combat missile off-axis launch angle but outside the range; Similarly, the enemy situation function f2(θ TDA , D) are as follows: In the formula, θ M2 D is the maximum off-axis launch angle allowed for the enemy close-range combat missile to meet the guidance requirements. Mmax2 The maximum range of the enemy's close combat projectile in the current state, D Mmin2 is the minimum range; According to f1(θ LOS , D) and f2(θ TDA , D), the close air combat situation at the preset moment is divided into seven types: attack, defense, mutual disadvantage, active, passive, balance and confrontation, and then the situation matrix is constructed.
6. The close air combat model construction method according to claim 5, characterized in that: In step 2, the air combat situation judgment also includes dynamic situation judgment, and the judgment process is as follows: Based on the static situation, the line of sight angular rate θ is introduced LOS ′, the distance change rate D′ is used for decision-making maneuvers; the line of sight angular rate θ LOS ′ is positive, indicating the sight angle θ LOS Increase, the angle trend is disadvantageous; A negative value indicates that the angle trend is advantageous; a positive value for the distance change rate D′ indicates that the distance between the two aircraft is increasing, and a negative value indicates that the distance between the two aircraft is decreasing; ΔH is defined as the height difference between the enemy and us: Where H1 is the altitude of our aircraft, H2 is the altitude of the enemy aircraft, and h is a constant; Based on the situation matrix, the line of sight angular velocity θ LOS ′, distance change rate D′ and the height difference ΔH between the enemy and us are used to construct the maneuver strategy based on energy maneuver under various situations.
7. The close air combat model construction method according to claim 6, characterized in that: In step 2, the maneuver strategy is as follows:
8. The close air combat model construction method according to claim 7, characterized in that: In step 3, the construction process of the longitudinal control model is as follows: The expected speed is calculated by situation judgment and reinforcement learning model as the control input, and the actual flight speed V A As PID feedback, a speed controller is built to output the throttle control value δ T , thus achieving speed control; after situation judgment and reinforcement learning model calculation of expected overload as pitch control command input, with actual overload as feedback, the rudder control amount is calculated, and then the actual pitch angle, pitch angle velocity θ, θ′ are used as series PID feedback to build a pitch controller, and the output elevator deflection amount δ e , thereby achieving pitch angle control.
9. The close air combat model construction method according to claim 8, characterized in that: When the throttle lever input is at its maximum, the speed still cannot reach the expected speed. T_real The correction value K is calculated and used to correct the pitch control command.
10. The close air combat model construction method according to claim 7, characterized in that: In step 3, the construction process of the lateral and directional control model is as follows: The expected sideslip angle β0 of the lateral sideslip control is kept at 0, and the expected roll angle is calculated by the situation judgment and reinforcement learning model. Then, the actual sideslip angle β is used as the input of the sideslip PID controller, and the actual roll angle and roll angular velocity φ, φ′ are used as the input of the roll PID controller to calculate the rudder deflection angle δ respectively. r and aileron deflection angle δ a , thereby achieving lateral heading control of the aircraft.