A method for real-time calculation of tactical control distance in air combat simulation environment

Abstract

The application discloses a kind of real-time solving method of tactical control distance in air combat simulation environment, and is obtained by generating and training real-time solving model, and the model training process includes: constructing the three-dimensional motion model of beyond-visual-range air combat and air-to-air missile and its constraint model and setting simulation limit condition to obtain offline solving model;With each group of sample motion data, the maximum initial search range of the boundary of tactical control distance is used to obtain the corresponding tactical control distance value by enemy missile and aircraft motion simulation, and then a sample database is constructed;Based on sparse auto-encoding network, an initial real-time solving model is constructed, and a target real-time solving model is obtained by network training using the sample database.The corresponding tactical control distance value is obtained by inputting each group of motion data to be tested into the target real-time solving model, and then the boundary of tactical control distance can be obtained.The application first proposes the concept of tactical control distance and provides theoretical guidance for air combat decision through quantitative model, with high solving accuracy and strong real-time performance.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of computer simulation and artificial intelligence, and particularly relates to a real-time calculation method for a tactical control distance in an air combat simulation environment. BACKGROUND

[0002] Both air combat game and air combat simulation system are detailed and realistic simulations of the entire combat process of a fighter by means of computer simulation. In order to effectively improve the authenticity of user experience and the ease of operation of the game and simulation system, it is necessary to simulate and design the game and simulation system from the actual air combat perspective, and more importantly, to simulate the tactics and design the convenient interaction, so as to restore the authenticity of air combat while improving the operation level of users in the game and simulation system.

[0003] In the air combat game and air combat simulation system, the air combat environment in which the fighter is located is generally divided into two types: over-the-horizon air combat and close-range air combat. Compared with close-range air combat, over-the-horizon air combat starts from a head-on situation and the distance is far, the angle condition of attack occupation is relatively easy to meet, and more attention is paid to the game and confrontation at the distance level.

[0004] Therefore, how to analyze the distance position data of the two sides in the air combat simulation to provide reference quantitative information for the tactical decision of the air combat simulation system has been one of the key research problems in the field of air combat simulation. The current related theoretical models mainly include air-to-air missile attack area and dynamic escape area.

[0005] The air-to-air missile attack area is analyzed from the attack level, and the research results mainly include cooperative attack area, three-dimensional attack area, and omnidirectional attack area, etc. This kind of method is based on the mathematical model of the dynamics equation and the guidance law of the air-to-air missile, combined with the three-degree-of-freedom or six-degree-of-freedom model of the carrier aircraft, to quickly search the boundary of the attack area. The attack area calculation methods mainly include real-time simulation method, table lookup interpolation method, polynomial fitting method and neural network fitting method, etc.

[0006] The dynamic escape area analyzes the quantitative index of the distance between the enemy and oneself when the typical event of air combat is triggered from the defense perspective, which is important support information for the escape maneuver decision of the pilot. This kind of method is based on the differential game theory and the simplified geometric model, and designs a fast solving method for the avoidance maneuver opportunity, which is used as the basis for dividing the aggressive adventure tactics and the rational conservative tactics, thereby providing a theoretical basis for the tactical control and intelligent decision of over-the-horizon air combat.

[0007] Beyond visual range air combat is a process of high coordination and coupling between attack and defense. In the existing research work, the attack area and the escape area only analyze the tactical distance of air combat from the single perspective of attack or defense, which is difficult to fully reflect the situation change of both sides of the combat, and cannot accurately give the conversion time between attack and defense in beyond visual range air combat, which seriously affects the quality of air combat simulation system. At the same time, there are many problems in the existing attack area calculation method. For example, although the real-time simulation method has high solution accuracy, its low timeliness cannot meet the needs of high dynamic air combat environment; although the table lookup interpolation method improves the simulation timeliness, the solution result error is large, which cannot meet the accuracy requirements of air combat; the polynomial fitting method needs artificial design of relationship, when the input dimension increases, the dependence relationship between input and output is difficult to design through intuitive observation, which greatly limits the application scene and fitting performance of this method; although the neural network fitting method improves the calculation accuracy, it needs a large number of labeled data samples for training, and is prone to "overfitting" and "gradient dispersion" phenomenon, which greatly limits the application in high fidelity air combat simulation environment. SUMMARY

[0008] The purpose of the embodiment of the present application is to provide a real-time solution method of tactical control distance in air combat simulation environment, so as to realize the simulation of beyond visual range air combat in high fidelity air combat simulation environment, and improve the accuracy and real-time performance of the solution of tactical control distance. The specific technical scheme is as follows:

[0009] In the first aspect, the embodiment of the present application provides a method for training a tactical control distance solution model in an air combat simulation environment, which comprises:

[0010] By constructing a three-dimensional motion model of the aircraft and the air-to-air missile in the beyond visual range air combat and its constraint model, and setting the simulation limit condition, an offline solution model of the tactical control distance is obtained;

[0011] For each group of obtained sample motion data, the sample motion data, the preset maximum initial search range of the boundary of the tactical control distance, and the offline solution model are used to obtain the tactical control distance value corresponding to the sample motion data by simulating the motion of the enemy missile and the aircraft. Each group of sample motion data represents the motion state of the aircraft and the enemy aircraft, including the speed of the aircraft, the height of the aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, the target entry angle, the target azimuth angle and the aircraft maneuvering overload;

[0012] Each group of sample motion data and the corresponding tactical control distance value are combined as a sample data group, and a sample database is obtained based on all sample data groups;

[0013] An initial real-time calculation model for tactical control distance is constructed based on a sparse autoencoder network, and the initial real-time calculation model is trained using the sample database to obtain a trained target real-time calculation model, which is used to output the corresponding tactical control distance value for any set of input motion data.

[0014] In one embodiment of the present invention, the three-dimensional motion model and its constraint model include:

[0015] The three-degree-of-freedom motion model of the carrier aircraft, the three-degree-of-freedom motion model of the missile, the relative motion model of the missile and the carrier aircraft, and the guidance law model.

[0016] In one embodiment of the present invention, the simulation constraints include:

[0017] When the distance between the missile and the target is less than the missile's maximum damage radius and the time and speed limits are not triggered, the missile is deemed to have successfully hit the target; otherwise, the missile attack is deemed to have failed.

[0018] The time constraint is: when the missile's flight time exceeds the missile's controllable flight time, the missile's energy is exhausted and it cannot hit the target.

[0019] The speed limit condition is: when the missile speed is less than the missile's minimum flight speed, the missile's maneuverability decreases and it cannot hit the target.

[0020] In one embodiment of the present invention, the step of using the set of sample motion data, the maximum initial search range of the preset tactical control distance boundary, and the offline solution model to obtain the tactical control distance value corresponding to the set of sample motion data by performing motion simulation of the enemy missile and the carrier aircraft includes:

[0021] For the search range of the tactical control distance boundary used in the current iteration, calculate its bisection division point; wherein, the search range of the tactical control distance boundary used in the first iteration is the maximum initial search range of the tactical control distance boundary;

[0022] Input the bisection point corresponding to the current iteration and the motion data of the sample group into the offline solution model, and output a miss distance value;

[0023] Determine whether the miss distance value indicates that the aircraft was not hit by the missile, and based on different determination results of whether the miss distance value indicates that the aircraft was not hit by the missile, reduce the search range of the tactical control distance boundary used in the current iteration accordingly;

[0024] Determine whether the search range of the narrowed tactical control distance boundary meets the preset error requirements;

[0025] If no, the search range of the reduced tactical control distance boundary is used as the search range of the tactical control distance boundary used in the next iteration, and the step of calculating the bisection point of the search range of the tactical control distance boundary used in the current iteration is returned.

[0026] If yes, the bisection point of the search range of the reduced tactical control distance boundary is calculated, and is used as the tactical control distance value corresponding to the group of sample motion data.

[0027] In an embodiment of the present application, the search range of the tactical control distance boundary used in the current iteration is reduced according to different results of the determination based on whether the miss distance value indicates that the carrier is not hit by the missile, comprising:

[0028] If the miss distance value indicates that the carrier is hit by the missile, the lower limit value in the search range of the tactical control distance boundary used in the current iteration is replaced by the bisection point corresponding to the current iteration;

[0029] If the miss distance value indicates that the carrier is not hit by the missile, it is determined whether at least one of the time limit condition and the speed limit condition is met;

[0030] If not, the step of inputting the bisection point corresponding to the current iteration and the group of sample motion data into the offline solving model is returned to re-simulate with the original input data;

[0031] If yes, the upper limit value in the search range of the tactical control distance boundary used in the current iteration is replaced by the bisection point corresponding to the current iteration.

[0032] In an embodiment of the present application, the determination of whether the search range of the reduced tactical control distance boundary meets the preset error requirement comprises:

[0033] Determining whether the absolute value of the difference between the upper limit and the lower limit of the search range of the reduced tactical control distance boundary is less than a preset boundary precision constant.

[0034] In an embodiment of the present application, the structure of the initial real-time solving model comprises:

[0035] An input layer, four sparse auto-encoding networks and an output layer; wherein each sparse auto-encoding network only contains one hidden layer;

[0036] Wherein, the number of nodes of the input layer is 7; the number of nodes of the hidden layers of the four sparse auto-encoding networks are 14, 16, 10 and 8 respectively; the number of nodes of the output layer is 1; the number of nodes of the hidden layers of the four sparse auto-encoding networks is determined by network performance influence evaluation optimization.

[0037] In an embodiment of the present application, the network training of the initial real-time solution model by using the sample database comprises:

[0038] Each sparse auto-encoding network in the initial real-time solution model is pre-trained separately to obtain corresponding pre-trained network weights; wherein the first sparse auto-encoding network is pre-trained by using the sample database; and the input in pre-training of the second sparse auto-encoding network is the excitation of each neuron in the hidden layer after pre-training of the previous sparse auto-encoding network.

[0039] After superimposing all the pre-trained sparse auto-encoding networks and adding the input layer and the output layer, the global optimization of the network weights of the overall network structure is performed by using the sample database to obtain the target real-time solution model after training.

[0040] In a second aspect, an embodiment of the present application provides a method for real-time solution of a tactical control distance in a simulated air combat environment, which comprises:

[0041] For any group of acquired motion data in the simulated air combat environment, the group of motion data is input into the pre-trained target real-time solution model to obtain a tactical control distance value corresponding to the group of motion data; wherein any group of motion data represents the motion state of an aircraft and an enemy aircraft, including the speed of the aircraft, the height of the aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, a target entry angle, a target azimuth angle and a maneuvering overload of the aircraft; and the target real-time solution model is trained according to the method for solution of a tactical control distance in a simulated air combat environment of the first aspect.

[0042] In an embodiment of the present application, after obtaining the corresponding tactical control distance value for any group of motion data, the method further comprises:

[0043] The tactical control distance values obtained from each group of motion data in the same simulated air combat environment and with continuous situation are sequentially connected to obtain a tactical control distance boundary; wherein only the target entry angle is different in each group of motion data in the same simulated air combat environment and with continuous situation, and each target entry angle changes at a preset step length.

[0044] The present application has the following beneficial effects:

[0045] 1) The embodiment of the present application proposes the concept of tactical control range (TCR) for the problem of decision-making information support in over-the-horizon air combat simulation confrontation environment, and proposes a quantitative representation method of tactical control range. In the provided tactical control range calculation model training method in air combat simulation environment, firstly, an offline calculation model of tactical control range is established. The offline calculation model can provide a high-accuracy motion simulation environment of enemy missiles and aircrafts because it fully combines the three-dimensional motion model and constraint model of aircrafts and air-to-air missiles in over-the-horizon air combat and adds simulation constraints. Therefore, it can simulate more accurate TCR values for each group of sample motion data. Next, a sample database is constructed based on the offline calculation model to provide training samples with high accuracy and high real matching degree of air combat simulation environment for the real-time calculation model. Then, an initial real-time calculation model based on sparse auto-encoding network is designed according to the characteristics of the data in the sample database. The initial real-time calculation model adds a sparsity constraint on the basis of the traditional auto-encoding network, which can effectively improve the network feature learning ability and improve the training effect. The "pre-training + global fine-tuning" strategy is used to train the initial real-time calculation model, so that the simulation results of the target real-time calculation model after training can meet the requirements of calculation accuracy and timeliness, and the accuracy and real-time performance are high when used to fit and calculate TCR.

[0046] 2) The tactical control distance real-time solving method provided by the embodiment of the present application realizes real-time solving of TCR based on a target real-time solving model constructed based on a sparse auto-encoding network, and the target real-time solving model is obtained by using the tactical control distance solving model training method in the air combat simulation environment provided by the embodiment of the present application, so that the simulation result can meet the requirements of calculation accuracy and timeliness. Therefore, when performing real-time solving of the tactical control distance, any group of to-be-measured motion data in the air combat simulation environment is input into the pre-trained target real-time solving model, and since the target real-time solving model introduces a sparse auto-encoding network to fit and solve TCR, extracts primary features of samples through the sparse auto-encoding network, obtains high-dimensional feature quantities representing the nonlinear law of the sample library, and establishes a deep network model to fit the high-dimensional feature quantities, the accurate value of TCR under the corresponding situation can be quickly obtained. The embodiment of the present application fully considers the decision-making requirements of fighter aircrafts in the over-the-horizon air combat simulation environment in terms of safety and task completion, establishes a tactical control distance (TCR) theoretical model based on air combat process analysis. A TCR real-time solving method based on a sparse auto-encoding network is designed. The method fully excavates the potential law of data samples, overcomes the gradient dispersion problem of a deep neural network, and effectively improves the fitting solving accuracy and solving timeliness. Existing research work only analyzes air combat tactics from the perspective of attack or defense, and it is difficult to fully reflect the situation changes of both combatants. The embodiment of the present application establishes a theoretical model of the tactical control distance based on a commonly used maneuver control model in air combat simulation, and designs a corresponding quantification method. The embodiment of the present application provides a tactical control distance real-time solving method in an air combat simulation environment based on a sparse auto-encoding network, which fully considers the characteristics of high dynamics, strong real-time and uncertain air combat simulation systems, and the decision-making conflict between approaching and turning out, and can effectively make up for the gap in situation information perception in over-the-horizon air combat simulation confrontation, better guide users (pilots) to make real-time decision selection, thereby improving the combat effectiveness of fighter aircrafts in the air combat simulation system, and has strong engineering practicability and effectiveness. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 A schematic diagram for understanding the tactical control distance in the embodiment of the present application;

[0048] Figure 2 A flowchart of a tactical control distance solving model training method in an air combat simulation environment provided by the embodiment of the present application;

[0049] Figure 3 A search flowchart of the tactical control distance for a group of sample motion data in the embodiment of the present application;

[0050] Figure 4 An abstract representation diagram of the sparse auto-encoding network in the embodiment of the present application;

[0051] Figure 5 A preliminary structural diagram of the sparse auto-encoding network in the embodiment of the present application;

[0052] Figure 6 A schematic diagram of the training process of the sparse auto-encoding network in the embodiment of the present application;

[0053] Figure 7 An algorithm structure schematic diagram of the method for training the tactical control distance calculation model in the air combat simulation environment provided by the embodiment of the present application;

[0054] Figure 8 A flow understanding schematic diagram of the method for real-time calculation of the tactical control distance in the air combat simulation environment provided by the embodiment of the present application;

[0055] Figure 9 A flow understanding schematic diagram of another method for real-time calculation of the tactical control distance in the air combat simulation environment provided by the embodiment of the present application;

[0056] Figure 10 A simulation result diagram of the real-time calculation of the tactical control distance in the embodiment of the present application;

[0057] Figure 11 A comparison diagram of the simulation results of the real-time calculation and the offline calculation of the tactical control distance in the embodiment of the present application;

[0058] Figure 12 A comparison diagram of the time effectiveness of the real-time calculation and the offline calculation of the tactical control distance in the embodiment of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0060] In order to facilitate the understanding of the difference between the embodiments of the present application and the prior art and the corresponding contribution, the tactical control distance proposed by the embodiments of the present application is briefly introduced below, and the inventive concept is briefly described on this basis. Specifically:

[0061] In view of the design defects of the reference quantitative information of the tactical decision of the air combat confrontation simulation system in the prior art, the embodiment of the present application first proposes a tactical control distance (TCR) as a kind of reference quantitative information. The tactical control distance is developed around the air combat distance, and a plurality of tactical control distance values are connected in sequence to form a tactical control distance boundary, which represents the boundary of a space region.

[0062] In order to facilitate the understanding of the tactical control distance in the embodiment of the present application, the following will be briefly described in combination with Figure 1 the accompanying drawings. Figure 1 The schematic diagram for understanding the tactical control distance in the embodiment of the present application. As shown in Figure 1 , the tactical control distance boundary is the boundary of a space region, which is obtained by fitting the curves of the respective tactical control distance values. Figure 1 The dark aircraft represents the aircraft of our side (our aircraft), and the light aircraft represents the aircraft of the enemy side (enemy aircraft). If the our aircraft is kept outside the tactical control distance boundary, that is Figure 1 , on the right side of the curve, it can be considered that the our aircraft is relatively safe, that is, even if the enemy aircraft launches a missile at this time, the our aircraft can evade the missile of the enemy aircraft through certain maneuvering mode; if the our aircraft is inside the tactical control distance boundary, that is Figure 1 , on the left side of the curve, the missile launched by the enemy aircraft will hit the our aircraft with a large probability. Therefore, it is of great guiding significance to calculate the respective tactical control distance values and then fit the tactical control distance boundary for accurately judging the air combat situation and timely adjusting the attack and defense tactical strategy.

[0063] It can be seen that the tactical control distance proposed in the embodiment of the present application can be used as one of the important maneuvering criteria in the air combat simulation confrontation. Using the tactical control distance, the user can accurately judge the air combat situation and timely adjust the attack and defense tactical strategy according to the indication information of the tactical control distance, guide the fighter aircraft to select the appropriate way and time for attack and defense conversion, and it is the core information for guiding the tactical behavior. Therefore, the calculation accuracy of the tactical control distance plays a decisive role in assisting the user to accurately grasp the air combat process, and will directly affect the combat effectiveness of the over-the-horizon air combat in the air combat simulation environment. It can be seen that how to design a high-precision and high-timeliness tactical control distance calculation method is of great significance for improving the fidelity, reliability and user control level of the air combat confrontation simulation system. The main purpose of the embodiment of the present application is to do this.

[0064] It should be noted that in the disclosed research model, the embodiment of the application is the first to propose the concept of tactical control distance, and is the first method to provide theoretical guidance for air combat decision-making through quantitative model. The existing research work only analyzes air combat tactics from the perspective of attack or defense, and it is difficult to fully reflect the situation changes of both combatants. The embodiment of the application proposes the tactical control distance and performs quantitative model representation, can fully consider attack and defense, and comprehensively balance the conflict between risk and near and conservative turn, so as to better guide the user (pilot) to make real-time decision selection.

[0065] In summary, in order to realize the super-visual distance air combat simulation in the high-fidelity air combat simulation environment, and improve the accuracy and real-time of the tactical control distance calculation, the embodiment of the application provides a tactical control distance calculation model training method in an air combat simulation environment and a real-time tactical control distance calculation method in an air combat simulation environment. In a first aspect, the embodiment of the application provides a tactical control distance calculation model training method in an air combat simulation environment, as shown in the following formula (1), which can include the following steps: Figure 2

[0066] S1, by constructing a three-dimensional motion model of a carrier aircraft and an air-to-air missile in a super-visual distance air combat and a constraint model thereof, and setting a simulation limit condition, an offline calculation model of the tactical control distance is obtained.

[0067] The offline calculation model of the embodiment of the application is a simulation environment built by using the three-dimensional motion model and the constraint model thereof and the simulation limit condition set, and is used for calculating the tactical control distance.

[0068] The three-dimensional motion model and the constraint model thereof include:

[0069] The three-dimensional motion model and the constraint model thereof include:

[0070] Specifically:

[0071] 1) The three-dimensional motion model of the carrier aircraft is as follows:

[0072]

[0073] Wherein, t represents the flight time, also represents the simulation time; (x, y, z) respectively represent three coordinate components of the carrier aircraft; respectively represent the carrier aircraft velocity, the flight path angle and the flight path angle; (n x ,n y ,n z ) respectively represent three overload components of the carrier aircraft; g represents the gravity acceleration, the value is 9.8 m / s 2 . ​

[0074] In the calculation of tactical control distance, it is assumed that the aircraft first performs a tail-to-tail maneuver in the horizontal plane to turn away from the enemy aircraft, and then maintains a constant speed to quickly disengage from the battlefield. The overload control model for the tail-to-tail maneuver is: n x =0,n y =1,n z =N zmax , where N zmax This represents the maximum lateral overload of the aircraft, and its sign depends on the direction with the smaller turning angle.

[0075] 2) The missile's three-degree-of-freedom motion model is as follows:

[0076]

[0077] Where t represents the flight time, and also the simulation time; (x m ,y m ,z m These represent the three coordinate components of the missile; These represent missile velocity, trajectory inclination angle, and trajectory deflection angle, respectively; T m D m The table shows the thrust and drag forces acting on the missile, t≥t push Time T m =T mmax , t <t push Time T m =0; t push T represents the duration of missile thrust. mmax Indicates the missile's maximum thrust; m m Indicates the mass of the missile; (n my ,n mz ) represent the two normal overload components of the missile.

[0078] 3) Relative motion model and guidance law model of missile-carrier aircraft

[0079]

[0080] In Equation 3), the first two formulas represent the relative motion model between the missile and the carrier aircraft, or the overload constraint model between the missile and the carrier aircraft, and the last formula represents the guidance law model; where r represents the missile-target distance; q represents the target line-of-sight angle; and σ represents the missile trajectory angle. and These represent the derivatives with respect to r, q, and σ, respectively; v represents the aircraft velocity; v m Indicates missile velocity; η and η m represents the lead angle of the target and missile velocity vectors, respectively; k is the proportional guidance coefficient.

[0081] In this embodiment of the invention, the simulation constraints include:

[0082] ①When the missile-target distance r is less than the maximum damage radius r max of the missile and the time limit condition and the speed limit condition are not triggered, it is determined that the missile successfully hits the target, otherwise it is determined that the missile attack fails;

[0083] ②The time limit condition is: when the flight time t of the missile is greater than the controllable flight time t max of the missile, the missile energy is exhausted and cannot hit the target;

[0084] ③The speed limit condition is: when the speed v m of the missile is less than the minimum flight speed v min of the missile, the maneuvering performance of the missile is reduced and cannot hit the target.

[0085] In the embodiment of the present application, according to the three-degree-of-freedom model of the carrier aircraft and the missile, the relative motion model of the missile-carrier aircraft and the guidance law model, and the simulation limit condition, an offline calculation model can be built. Those skilled in the art can understand that the offline calculation model is an offline calculation simulation environment for a tactical control distance, and the simulation needs to use the parameters of the carrier aircraft, the missile and the air combat simulation environment, so simulation parameter setting is needed before the offline calculation model is used. Specifically, it can include: the maximum lateral overload N zmax of the carrier aircraft, the controllable flight time t max of the missile, the missile thrust action time t push , the minimum flight speed v min of the missile, the maximum damage radius r max of the missile, the mass m m of the missile, the maximum thrust T mmax of the missile and the proportional guidance coefficient k. The simulation parameter setting is shown in Table 1.

[0086] Table 1 Simulation parameter setting table

[0087]

[0088]

[0089] The purpose of building the offline calculation model in the embodiment of the present application is to give a numerical calculation method of the tactical control distance, so as to build a sample database as a real-time calculation model training sample based on it in the subsequent.

[0090] S2, for each set of obtained sample motion data, using the set of sample motion data, the maximum initial search range of the preset tactical control distance boundary and the offline calculation model, the motion simulation of the enemy missile and the carrier aircraft is carried out, and the tactical control distance value corresponding to the set of sample motion data is obtained.

[0091] wherein each set of sample motion data represents the motion state of the carrier aircraft and the enemy aircraft, including the speed of the carrier aircraft, the height of the carrier aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, the target approach angle, the target azimuth angle and the carrier aircraft maneuvering overload. Each parameter in a set of sample motion data can be represented by v, h, v t ,h t ,q m ,q f ,n f , respectively.

[0092] The embodiment of the present application can pre-acquire a plurality of sets of sample motion data, and the process can be as follows:

[0093] A, for each of the speed of the carrier aircraft, the height of the carrier aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, the target approach angle, the target azimuth angle and the carrier aircraft maneuvering overload, a corresponding parameter search range and a search step are pre-set.

[0094] wherein for the above seven parameters, the parameter search range corresponding to the speed of the carrier aircraft and the speed of the enemy aircraft can be the same, such as 150-450 m / s, or 100-500 m / s, etc. The parameter search range corresponding to the height of the carrier aircraft and the height of the enemy aircraft can be the same, such as 2000-20000 meters, or 1500-15000 meters, etc. The parameter search range corresponding to the target approach angle can be -90°-90°, etc. The parameter search range corresponding to the target azimuth angle can be -45°-45°, etc. The parameter search range corresponding to the carrier aircraft maneuvering overload can be 0-9g, etc., with the unit being g.

[0095] The search step corresponding to the speed of the carrier aircraft and the speed of the enemy aircraft can be the same, such as 50 m / s, or 100 m / s, etc. The search step corresponding to the height of the carrier aircraft and the height of the enemy aircraft can be the same, such as 500 meters, or 1000 meters, etc. The search step corresponding to the target approach angle can be 5°, 10°, etc. The search step corresponding to the target azimuth angle can also be 5°, 10°, etc. The search step corresponding to the carrier aircraft maneuvering overload can be 1g, etc.

[0096] It can be understood that the parameter search range corresponding to the above seven parameters is an empirical value obtained according to actual measurement data, and the search step can be reasonably designed according to simulation accuracy and the number of sets of sample motion data, etc.

[0097] B, for each parameter, the parameter search range corresponding to the parameter is searched multiple times with its own search step to obtain the search value of the parameter in each search.

[0098] For example, if the search step of the target entry angle is 10°, the search values obtained in the search range [-90°, +90°] corresponding to the target entry angle by using the search step are -90°, -80°, -70°, 0°, 10°, 70°, 80° and 90° in sequence.

[0099] C. The search values of the parameters obtained in all the searches are combined to form a plurality of sets of sample motion data.

[0100] The sample motion data can represent the motion states of the carrier and the enemy aircraft. Each set of sample motion data contains the search values of seven parameters.

[0101] Specifically, all the search values of the parameters obtained in the searches can be randomly combined to form a plurality of sets of sample motion data. The search value of a parameter obtained in a search can be repeatedly used in the plurality of sets of sample motion data.

[0102] Alternatively, all the search values of the parameters obtained in the searches can be combined according to a certain order to form a plurality of sets of sample motion data. For example, the search values of the parameters in the same search order form a set of sample motion data, which is reasonable.

[0103] Moreover, the embodiment of the present application presets a maximum initial search range for the tactical control distance boundary, and subsequent iterative searches are performed in the range to obtain the tactical control distance corresponding to each set of sample motion data. Specifically, the maximum initial search range of the tactical control distance boundary can be [0km, 200km] or the like.

[0104] Hereinafter, any set of sample motion data in the plurality of sets of sample motion data is taken as an example to illustrate how to obtain the tactical control distance value corresponding to the set of sample motion data by using the set of sample motion data, the preset maximum initial search range of the tactical control distance boundary and the offline solving model.

[0105] In an optional implementation, the set of sample motion data, the preset maximum initial search range of the tactical control distance boundary and the offline solving model are used to obtain the tactical control distance value corresponding to the set of sample motion data by performing motion simulation of the enemy missile and the carrier, including the following steps:

[0106] a1. The bisection point of the search range of the tactical control distance boundary used in the current iteration is calculated.

[0107] In the embodiment of the present application, each iteration is implemented by using the search range of the corresponding tactical control distance boundary. When the search range of the tactical control distance boundary changes, the iteration number also changes.

[0108] For the convenience of understanding, let i represent the iteration number, where i is a natural number greater than 0. The search range of the tactical control distance boundary used in the current iteration, which is the ith iteration, can be represented as [R n,i ,R f,i ].

[0109] Wherein, the search range of the tactical control distance boundary used in the first iteration is the maximum initial search range of the tactical control distance boundary; that is, for i = 1, [R n,1 ,R f,1 ] R n,1 = 0km; R f,1 = 200km.

[0110] The bisection point calculated from the search range of the tactical control distance boundary used in the current iteration can be represented as R g,i , which is referred to as the bisection point corresponding to the current iteration for the sake of simplicity, wherein R g,i is calculated according to the following formula:

[0111]

[0112] a2, input the bisection point corresponding to the current iteration and the sample motion data into the offline solving model, and output a miss distance value.

[0113] Specifically, input R g,i and the sample motion data (v, h, v t , h t , q m , q f , n f ) into the offline solving model to simulate the motion of the enemy missile and the aircraft, wherein the offline solving model has completed simulation parameter setting before, and is constrained by simulation constraints during the simulation process. After the simulation process is completed, a miss distance value is output.

[0114] a3, determine whether the miss distance value indicates that the aircraft is not hit by the missile, and based on different determination results of whether the miss distance value indicates that the aircraft is not hit by the missile, correspondingly reduce the search range of the tactical control distance boundary used in the current iteration.

[0115] Wherein, the way to determine whether the miss distance value indicates that the aircraft is not hit by the missile can be:

[0116] The miss distance value is a first preset value, indicating that the aircraft is not hit by the missile; the miss distance value is a second preset value, indicating that the aircraft is hit by the missile;

[0117] For example, the first preset value and the second preset value can be 0 and 1 respectively, and of course other numbers can also be used to distinguish whether the carrier is not hit by the missile.

[0118] In the embodiment of the application, the carrier is hit by the missile or the carrier is not hit by the missile, and finally the search range of the tactical control distance boundary used in the current iteration is reduced, but the reduction methods are different. It can be understood that the embodiment of the application searches for the tactical control distance value corresponding to the set of sample motion data by continuously reducing the search range of the tactical control distance boundary in the simulation iteration.

[0119] Specifically, based on different determination results of whether the miss distance value indicates that the carrier is not hit by the missile, the search range of the tactical control distance boundary used in the current iteration is reduced accordingly, including:

[0120] b1, if the miss distance value indicates that the carrier is hit by the missile, the lower limit value in the search range of the tactical control distance boundary used in the current iteration is replaced by the bisection point corresponding to the current iteration.

[0121] For the ith iteration, if the miss distance value output by the offline calculation model after the simulation process ends indicates that the carrier is hit by the missile, the search range of the tactical control distance boundary used in the current iteration can be directly reduced in a corresponding manner, and the search range of the tactical control distance boundary obtained after the reduction in this manner is [R g,i ,R f,i ].

[0122] b2, if the miss distance value indicates that the carrier is not hit by the missile, it is determined whether at least one of the time limit condition and the speed limit condition is satisfied.

[0123] For the ith iteration, if the miss distance value output by the offline calculation model after the simulation process ends indicates that the carrier is not hit by the missile, it is necessary to further determine the reason why the carrier is not hit by the missile to determine whether the search range of the tactical control distance boundary used in the current iteration can be directly reduced in a corresponding manner.

[0124] At this time, it is necessary to determine whether at least one of the time limit condition and the speed limit condition is satisfied, and the time limit condition and the speed limit condition are contents of the simulation limit condition. Please understand in combination with the foregoing, and no detailed description is made here.

[0125] b3, if not, return to the step of inputting the bisection point corresponding to the current iteration and the set of sample motion data into the offline calculation model to simulate again with the original input data.

[0126] If the time limit condition is not met, and the speed limit condition is not met, that is, the reason why the carrier is not hit by the missile is not due to the missile performance factors such as energy depletion or performance decline of the missile. At this time, it is possible that it is caused by some data delay and the like, and therefore, it is necessary to re-simulate using the original simulation input data, to output the miss distance value again, and to determine whether the miss distance value output again can indicate that the carrier is not hit by the missile. If it is determined again that the carrier is hit by the missile, the step b1 is executed. If it is determined again that the carrier is not hit by the missile, the step b2 is executed.

[0127] b4, if met, replace the upper limit value in the search range of the tactical control distance boundary used in the current iteration with the bisection point corresponding to the current iteration.

[0128] If the time limit condition is met, or the speed limit condition is met, that is, the reason why the carrier is not hit by the missile is due to the missile performance factors such as energy depletion or performance decline of the missile. At this time, it is not necessary to simulate again, and it is possible to directly reduce the search range of the tactical control distance boundary used in the current iteration in a corresponding manner. The search range of the tactical control distance boundary obtained after the reduction in this manner is [R n,i ,R g,i ].

[0129] Through the above steps, after outputting a miss distance value from the offline calculation model for the input bisection point corresponding to the current iteration and the set of sample motion data, regardless of the judgment result of the miss distance value for whether the carrier is not hit by the missile, or whether it is re-simulated, finally, a reduced search range of the tactical control distance boundary will be obtained, which is [R g,i ,R f,i ] or [R n,i ,R g,i ].

[0130] At the same time, it needs to be emphasized that, in the above process, even if re-simulation occurs, since the same search range of the tactical control distance boundary is used as the simulation input data, it still belongs to the same iteration.

[0131] a4, determine whether the reduced search range of the tactical control distance boundary meets the preset error requirement.

[0132] Regardless of whether the reduced search range of the tactical control distance boundary is [R g,i ,R f,i ] or [R n,i ,R g,i ], it is necessary to further determine whether it meets the preset error requirement.

[0133] In an optional embodiment, the step of judging whether the search range of the tactical control distance boundary after the reduction meets the preset error requirement comprises:

[0134] judging whether the search range of the tactical control distance boundary after the reduction meets the condition that the absolute value of the difference between the upper limit and the lower limit is less than the preset boundary precision constant.

[0135] That is, for [R g,i ,R f,i ], it is judged whether |R g,i -R f,i |<ε exists; for [R n,i ,R g,i ], it is judged whether |R n,i -R g,i |<ε exists. Wherein, ε represents the preset boundary precision constant, and its value is determined according to experience value, such as 500 meters, 1000 meters, etc.

[0136] If no, a5 is executed, the search range of the tactical control distance boundary after the reduction is used as the search range of the tactical control distance boundary used in the next iteration, and the step of calculating the bisection point of the search range of the tactical control distance boundary used in the current iteration is returned.

[0137] If the search range of the tactical control distance boundary after the reduction does not meet the preset error requirement, the iteration cannot be stopped, the search range of the tactical control distance boundary after the reduction is used as the search range of the tactical control distance boundary used in the next iteration, and the step a1 is returned to continue the iteration.

[0138] That is, if the search range of the tactical control distance boundary after the reduction is [R g,i ,R f,i ], it is determined that [R n,i+1 ,R f,i+1 ]=[R g,i ,R f,i ] and the step a1 is returned.

[0139] If the search range of the tactical control distance boundary after the reduction is [R n,i ,R g,i ], it is determined that [R n,i+1 ,R f,i+1 ]=[R n,i ,R g,i ] and the step a1 is returned.

[0140] If yes, a6 is executed, the bisection point of the search range of the tactical control distance boundary after the reduction is calculated and used as the tactical control distance value corresponding to the group of sample motion data.

[0141] If the search range of the reduced TCR boundary meets the preset error requirement, the iteration is stopped, the bisection point of the search range of the reduced TCR boundary is calculated according to formula (4), and the calculated bisection point is determined as the TCR value corresponding to the set of sample motion data. Wherein, the TCR value can be represented as D TCR .

[0142] The process of obtaining the corresponding TCR value for a set of sample motion data can be understood with reference to Figure 3 . Figure 3 The flowchart of the search process of the TCR for a set of sample motion data in the embodiment of the present application is shown. Please understand in combination with the foregoing description, wherein it needs to be explained that for the current iteration, if the miss distance value output by the offline calculation model indicates that the aircraft is not hit by the missile, and it is judged that the time limit condition and the speed limit condition are not met, t=t+dt indicates that one simulation period is added, and the current iteration is simulated again using the original input data, which means that the bisection point corresponding to the current iteration and the set of sample motion data are input into the offline calculation model for simulation again, but the iteration number is unchanged in the simulation again.

[0143] It can be understood that through the TCR boundary search of S2 step, the corresponding TCR value can be obtained for each set of sample motion data.

[0144] S3, each set of sample motion data and the corresponding TCR value are combined as a sample data set, and a sample database is obtained based on all sample data sets.

[0145] This step can be understood as a data arrangement and normalization process. Specifically, the sample database can be in a table form, wherein each row is a sample data set, which is composed of a set of sample motion data and a corresponding TCR value. And the data in the sample database is normalized to eliminate the influence of dimension on the data. The specific normalization method can be the maximum and minimum normalization method, etc. Therefore, it can be understood that the value range in the sample database is mapped to [0, 1].

[0146] The sample database reflects the mapping relationship between the sample motion data and the TCR value. Specifically, the TCR can be represented as the following nonlinear function relationship:

[0147] D TCR =D(v,h,v t ,h t ,q m ,q f ,n f ) (5)

[0148] Wherein, v, h are the speed and height of the carrier; v t , h t are the speed and height of the enemy aircraft; q m is the target entry angle; q f is the target azimuth angle; n f is the carrier maneuvering overload, D TCR is the fitting value of the tactical control distance, D is the mapping relationship of the input parameters (v, h, v t , h t , q m , q f , n f to the output of the tactical control distance.

[0149] In addition, the sample database can be further divided into a training set and a test set according to a certain proportion for subsequent use, for example, 30% of the sample data groups are randomly selected from the sample database to form the test set, and the rest are used as the training set, and the like.

[0150] S4, an initial real-time calculation model of the tactical control distance is constructed based on a sparse self-encoding network, and the initial real-time calculation model is trained by using the sample database to obtain a target real-time calculation model after training, so as to output the corresponding tactical control distance value for any group of input motion data to be measured.

[0151] The initial real-time calculation model of the embodiment of the application is constructed by using a sparse self-encoding network (SAE), and a TCR fitting network based on SAE is realized. The abstract representation of the sparse self-encoding network is as shown in Figure 4 .

[0152] In an optional implementation, the structure of the initial real-time calculation model comprises:

[0153] an input layer, four sparse self-encoding networks and an output layer; wherein each sparse self-encoding network only contains one hidden layer;

[0154] Wherein, the number of nodes of the input layer is 7; the number of nodes of the hidden layers of the four sparse self-encoding networks is 14, 16, 10 and 8 respectively; the number of nodes of the output layer is 1; the number of nodes of the hidden layers of the four sparse self-encoding networks is determined through network performance influence evaluation optimization.

[0155] Specifically, the embodiment of the present application can set different node numbers for the hidden layers of the four SAEs after building a network structure of an input layer, four SAEs and an output layer, thereby obtaining a plurality of network structures after matching the input layer and the output layer, then evaluating the performance of each obtained network, the evaluation indexes can include root mean square error (RMSE), mean absolute percentage error (MAPE) and mean absolute error (MAE), etc., then selecting a network with the optimal performance and determining the structure of the network as the structure of the initial real-time calculation model. The network performance evaluation process is not described in detail here.

[0156] It is known to those skilled in the art that an auto-encoder network (AE) is a kind of unsupervised learning deep neural network, which can be used to extract high-dimensional features of sample data. The core purpose of the auto-encoder network is not to reconstruct the input by the output, but to learn the intrinsic features of the original data according to the mapping relationship from the input layer to the hidden layer. Since the number of input layer neurons of the initial real-time calculation model of the embodiment of the present application is 7, which is relatively small, and the number of nodes of the determined SAE hidden layer is more than the number of input layer nodes, it is difficult to learn the deep features of the sample by using the traditional auto-encoder network. Therefore, in view of this problem, the embodiment of the present application establishes a sparse auto-encoder network model. The sparse auto-encoder network adds a sparsity constraint on the basis of the traditional auto-encoder network, which can effectively improve the generalization ability of network feature learning, so the network feature learning ability is stronger than that of the traditional auto-encoder network. Specifically, a penalty factor for controlling sparsity is added to the loss function of the traditional auto-encoder network.

[0157] The loss function of the traditional auto-encoder network is represented as:

[0158]

[0159] The loss function of the sparse auto-encoder network is represented as:

[0160]

[0161] wherein m represents the input layer dimension; h w,b (x (i) ) represents the network output corresponding to the input x (i) ; β represents the weight factor for controlling sparsity; n represents the number of hidden layer neurons; represents the KL distance, which is used to represent the difference between two functions; ρ represents the expected value of the average activation; represents the average activation of the jth neuron.

[0162] In an optional embodiment, the sample database is used to train the initial real-time calculation model to obtain a trained target real-time calculation model, including the following steps:

[0163] c1, pre-train each sparse autoencoder network in the initial real-time solution model individually to obtain the corresponding pre-trained network weights.

[0164] Specifically, this step is the network pre-training process, which requires training each SAE individually to obtain the corresponding optimal network parameters. The hidden layer of the nth SAE is represented as L. n The SAE pre-training process is unsupervised; for TCR sample data, only the input matrix is ​​needed for training. For ease of description, let X be the input matrix of the nth SAE. (n) The output matrix is ​​denoted as Y. (n) .

[0165] Specifically, given the input matrix X of the first SAE (1) and output matrix Y (1) =X (1) Randomly initialize network parameter values ​​and train the first SAE using a loss function incorporating KL distance. SAE (W,b) The network weights are updated using the backpropagation algorithm for error until the output Y meets the convergence error or the training reaches the maximum simulation step size. The weights W1 from the input layer to the hidden layer L1 are recorded, completing the training of the first SAE. The first sparse autoencoder network is pre-trained using a sample database; specifically, the input matrix X for the first SAE pre-training... (1) It is composed of motion data from all groups of samples in the sample database.

[0166] Starting with the second sparse autoencoder network, the input during pre-training is the activation of each neuron in the hidden layer of the previous sparse autoencoder network after pre-training. Specifically, the activation of each neuron in hidden layer L1 is used as the input matrix of the second SAE, i.e., X. (2) =(h(x1),h(x2),...,h(x L1 )) T Output matrix Y (2) =X (2) =(h(x1),h(x2),...,h(x L1 )) T Similar to the training process of the first SAE, the second SAE uses the activation of the hidden layer L1 of the first SAE as input for training, reconstructs the activation in the output layer, until the convergence error is satisfied or the maximum simulation step size is reached, and then records the weights W2 from the input layer to the hidden layer L2, thus completing the training of the second SAE.

[0167] Then the excitation of each neuron of the second SAE's hidden layer L2 is taken as the input matrix of the third SAE, and the training of the third SAE is continued according to the above-mentioned mode until the pre-training process of all hidden layers L1, L2,..., L n is completed, and the pre-trained network weights W1, W2,..., W n are obtained, that is, the pre-training of the initial real-time calculation model is completed.

[0168] c2, all the pre-trained sparse auto-encoding networks are stacked, and the input layer and the output layer are added, and then the global optimization of the network weights of the whole network structure is performed by using the sample database to obtain the trained target real-time calculation model.

[0169] The pre-training of the step c1 only optimizes the parameters of each hidden layer, and the network parameters of the hidden layer obtained are only locally optimal. Since the pre-training is performed, the initial weights W of the whole network will be in a relatively optimal position, and then in the step c2, the input layer and the output layer are added, and the whole network is fine-tuned from the global perspective by using the labeled data, so the step c2 can be understood as a network global optimization process.

[0170] In the formula, all the pre-trained hidden layers are stacked, and the preliminary structure of the deep network model finally obtained by stacking is shown in Figure 5 . Figure 5 It is to be understood that the diagram only contains all the pre-trained hidden layers, and does not contain the input layer and the output layer.

[0171] Specifically, the network global optimization is a supervised learning process, the input matrix is composed of all group sample motion data in the sample database, the output matrix is composed of all tactical control distance values in the sample database, and the global optimization of the weights of the whole deep network is performed by substituting the loss function J SAE (W, b) into the error back propagation algorithm, until the error convergence condition is met or the maximum simulation step is reached, and all the trained network parameters are recorded, that is, the global training of the sparse auto-encoding network is completed. The training process of the sparse auto-encoding network is shown in Figure 6 . In the formula, the SAE unit represents each sparse auto-encoding network; the decoder represents the decoder; and the encoder represents the encoder.

[0172] It is to be understood that after the network training of the initial real-time calculation model is completed, the trained target real-time calculation model can be obtained.

[0173] In actual use, any group of to-be-tested motion data (v, h, v th t ,q m ,q f ,n f )input target real-time solving model, can output corresponding tactical control distance value D TCR .

[0174] Further, a plurality of tactical control distance values can be obtained by using a plurality of groups of required motion data to be measured, so as to obtain a tactical control distance boundary, which is a curve connected by the tactical control distance values. The specific process is described below.

[0175] For the processing procedure of the method for training the tactical control distance solving model in the air combat simulation environment provided by the embodiment of the present application, please refer to the foregoing description and Figure 7 understanding, Figure 7 The algorithm structure diagram of the method for training the tactical control distance solving model in the air combat simulation environment provided by the embodiment of the present application. The processing procedure of the method for training the tactical control distance solving model in the air combat simulation environment provided by the embodiment of the present application mainly includes the following two parts: ①TCR offline solving mathematical model (offline solving model); ②TCR precise fitting network based on sparse auto-encoding network (real-time solving model). Among them, "initialization of simulation parameters and conditions" corresponds to the simulation parameter setting required before using the offline solving model in the simulation environment in S1 step; "building simulation environment" corresponds to building the offline solving model according to the three-degree-of-freedom model of the aircraft and the missile, the relative motion model of the missile-aircraft and the guidance law model, and the simulation limiting conditions in S1 step; "TCR boundary search" corresponds to S2 step; "data preprocessing" corresponds to S3 step; "TCR sample database" is the sample database obtained in S3 step; "real-time solving model" represents the process from the initial real-time solving model to the target real-time solving model after training; among them, "constructing network model" corresponds to constructing the initial real-time solving model of the tactical control distance based on the sparse auto-encoding network in S4 step; "network depth setting" refers to the process of determining the best node number of each hidden layer after evaluating the influence of different node numbers of each hidden layer in the initial real-time solving model on the network performance; "sparsity restriction" is a penalty factor for controlling sparsity added in the loss function in SAE; "network parameter training" represents the network training of the initial real-time solving model, and "pre-training" and "global optimization" correspond to c1 step and c2 step, respectively. For specific contents, please refer to the related description in the foregoing description, which will not be repeated here.

[0176] The embodiment of the application proposes a tactical control distance (TCR) concept for decision information support in an over-the-horizon air combat simulation confrontation environment, and proposes a quantitative representation method of the tactical control distance (TCR). In the provided tactical control distance calculation model training method in the air combat simulation environment, firstly, an offline calculation model of the tactical control distance is established. The offline calculation model can provide a high-accuracy motion simulation environment of the enemy missile and the aircraft because it fully combines the three-dimensional motion model and the constraint model of the aircraft and the air-to-air missile in the over-the-horizon air combat, and adds simulation constraints, and simulates more accurate TCR values for each group of sample motion data. Next, a sample database is constructed based on the offline calculation model, to provide training samples with high accuracy and high real matching degree of the air combat simulation environment for the real-time calculation model. Then, an initial real-time calculation model based on a sparse auto-encoding network is designed according to the characteristics of the data in the sample database. The initial real-time calculation model adds a sparsity constraint on the basis of the traditional auto-encoding network, which can effectively improve the network feature learning ability and improve the training effect. The network training of the initial real-time calculation model is performed by using the "pre-training + global fine-tuning" strategy, so that the simulation result of the target real-time calculation model after training can meet the requirements of calculation accuracy and timeliness, and the accuracy and real-time performance are high when used for fitting and calculating the TCR.

[0177] In the second aspect, based on the tactical control distance calculation model training method in the air combat simulation environment provided in the first aspect, the embodiment of the application further provides a tactical control distance real-time calculation method in an air combat simulation environment, which is described with reference to Figure 8 It is understood that the method comprises:

[0178] For any group of acquired motion data in the air combat simulation environment, the group of motion data to be measured is input into the pre-trained target real-time calculation model to obtain the tactical control distance value corresponding to the group of motion data to be measured.

[0179] Any group of motion data to be measured represents the motion state of the aircraft and the enemy aircraft, including the speed of the aircraft, the height of the aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, the target entry angle, the target azimuth angle and the aircraft maneuvering overload; and the target real-time calculation model is trained according to the air combat simulation environment tactical control distance calculation model training method of the first aspect.

[0180] Specifically, any group of motion data to be measured can be represented as X=(v,h,v t ,h t ,q m ,q,n f ), which is input into the target real-time calculation model trained according to the air combat simulation environment tactical control distance calculation model training method provided in the first aspect, and the target real-time calculation model can output the tactical control distance D in the over-the-horizon air combat simulation environment under the situationTCR In this process, a set of motion data corresponds to a situation, and different situations are mainly reflected by the target approach angle. The process of obtaining the real-time target calculation model is detailed in the first section and will not be repeated here.

[0181] Optionally, after obtaining the corresponding tactical control distance value for any set of motion data to be tested, the method further includes:

[0182] By sequentially concatenating the tactical control distance values ​​obtained from various sets of measured motion data under the same air combat simulation environment and with continuous situational awareness, the tactical control distance boundary is obtained. For detailed procedures, please refer to [link to relevant documentation]. Figure 9 understand.

[0183] Among them, in the same air combat simulation environment and with continuous situation, the only difference between the target entry angles in each group of motion data is that the target entry angles change with a preset step size.

[0184] Specifically, the motion data to be tested in each group (v, h, v) t ,h t ,q m ,q,n f In the target entry angle q only m Unlike other methods, the target approach angle changes with a preset step size from the first set of motion data to the last set. The preset step size can be, for example, 5°, 10°, or 20°, and can be selected as needed.

[0185] For example, the target entry angle typically ranges from [-90°, 90°], and varies with a preset step size. Each discrete target entry angle corresponds to a situation. The tactical control distance value D calculated by the real-time target solution model for situations ranging from -90° to +90° is then used. TCR By connecting them sequentially, the tactical control distance boundary can be obtained.

[0186] Those skilled in the art will understand that, for air combat simulation, obtaining the tactical control distance value corresponding to a single set of motion data to be measured only fulfills a basic requirement, while obtaining the tactical control distance boundary is more practically significant. By utilizing the tactical control distance boundary, the air combat situation can be accurately judged, and offensive and defensive tactical strategies can be adjusted in a timely manner.

[0187] The tactical control distance is a key dynamic node information in the beyond-visual-range air combat confrontation exercise, and is an important reference standard for the maneuver user to execute corresponding tactics in the air combat simulation confrontation environment. The method for real-time solving of the tactical control distance in the air combat simulation environment provided by the embodiment of the application is realized by a target real-time solving model constructed based on a sparse self-encoding network, and the target real-time solving model is obtained by using the training method for the tactical control distance solving model in the air combat simulation environment proposed by the embodiment of the application. In the training method, the offline solving model is combined with the three-dimensional motion model of the aircraft and the air-to-air missile in the beyond-visual-range air combat and the constraint model thereof, and is subjected to simulation limitation, so that a high-accuracy motion simulation environment of the enemy missile and the aircraft is provided, and a more accurate TCR value is simulated for each group of sample motion data. Then, a sample database is constructed based on the offline solving model, and the training sample with high accuracy and high real matching degree of the air combat simulation environment is provided for the real-time solving model. Then, an initial real-time solving model based on the sparse self-encoding network is designed according to the characteristics of the data in the sample database, the sparse limitation is added to the initial real-time solving model based on the traditional self-encoding network, the network feature learning ability is effectively improved, and the training effect is improved. The network training of the initial real-time solving model is performed by using the strategy of "pre-training + global fine-tuning", so that the simulation result of the target real-time solving model after the training can meet the requirements of the calculation accuracy and the timeliness, and the accuracy and the real-time performance are high when the target real-time solving model is used to fit and solve the TCR.

[0188] Therefore, when the real-time solving of the tactical control distance is performed, any group of to-be-tested motion data in the air combat simulation environment is input into the target real-time solving model which has been pre-trained, and the target real-time solving model is used to fit and solve the TCR by introducing the sparse self-encoding network, the sample primary features are extracted by the sparse self-encoding network, the high-dimensional feature quantity representing the nonlinear law of the sample database is obtained, and the deep network model is used to fit the high-dimensional feature quantity, so that the accurate value of the TCR under the corresponding situation can be quickly obtained.

[0189] The embodiment of the application fully considers the decision requirement of the fighter aircraft in the beyond-visual-range air combat simulation environment in consideration of the safety and the task completion degree, establishes a tactical control distance (TCR) theoretical model based on the analysis of the air combat process, and designs a TCR real-time solving method based on the sparse self-encoding network. The method fully mines the potential law of the data sample, overcomes the gradient dispersion problem of the deep neural network, and effectively improves the fitting solving accuracy and the solving timeliness.

[0190] Existing research work only analyzes air combat tactics from the perspective of attack or defense, and it is difficult to comprehensively reflect the situation changes of both combat parties. The embodiment of the present application establishes a theoretical model of the tactical control distance based on the commonly used maneuver control model in air combat simulation, and designs a corresponding quantification method. The embodiment of the present application provides a real-time calculation method for the tactical control distance in the air combat simulation environment based on the sparse self-encoding network, which provides theoretical guidance for air combat decision-making. The real-time calculation method for the tactical control distance in the air combat simulation environment of the embodiment of the present application fully considers the characteristics of high dynamic, strong real-time and uncertain air combat simulation system, and the conflict between attack and defense, and comprehensively balances the decision-making conflict between approaching and turning out, which can effectively make up for the loopholes in the situation information perception in the beyond-visual-range air combat simulation confrontation, better guide the user (pilot) to make real-time decision selection, thereby improving the combat effectiveness of the fighter in the air combat simulation system, and has strong engineering practicability and effectiveness.

[0191] In order to verify the effectiveness of the tactical control distance calculation model training method in the air combat simulation environment and the real-time calculation method for the tactical control distance in the air combat simulation environment proposed in the embodiment of the present application, the following will be described in combination with experimental data.

[0192] (I) Generating a sample database

[0193] According to the simulation parameter setting in Table 1, the initial height of the enemy aircraft and the aircraft is 7000m, the simulation step is 0.01s, the step of the target entry angle is 10°, and the target entry angle is searched from -90° to +90°. The simulation results of the tactical control distance boundary of the offline calculation model are shown in FIG. 2. Figure 10 Figure 10 FIG. 3 is a simulation result diagram of the real-time calculation of the tactical control distance in the embodiment of the present application. As can be seen from FIG. 3, Figure 10 it can be seen that the tactical control distance is relatively larger under the head-on situation of the target entry angle of 0°, and the aircraft needs to approach the enemy aircraft while bearing a larger threat; the tactical control distance is shortened when the enemy is approached laterally, and the boundary is reduced by about 23% when the target entry angle is 90° compared with the head-on engagement.

[0194] Based on the above offline calculation model of TCR, offline simulation is carried out under the following conditions: the height range of the aircraft and the enemy is 2000-20000m, the speed range of the aircraft and the target is 150-450m / s, the target azimuth angle range is -45°-45°, the target entry angle range is -90°-90°, the aircraft maneuvering overload range is 0-9g, and other simulation parameters are shown in Table 1. T=1000 sample data are obtained by simulation, and part of the sample data is shown in Table 2. Each row is a sample data group, and all sample data groups are normalized and integrated to form a sample database. ​

[0195] Table 2 simulation sample data table

[0196]

[0197] (ii) determine the network structure and network training process

[0198] To evaluate the fitting effect, the network performance indicators are selected as the mean relative error MAPE, the mean absolute error MAE, and the square root mean square error RMSE, which are defined as follows:

[0199]

[0200] Wherein, TCR k , TCR k are the network output value and the theoretical value of the tactical control distance value, respectively, and T is the number of samples.

[0201] The initial real-time solution model is used as the TCR fitting network, the number of input layer nodes is 7, the number of output layer nodes is 1, the activation function of the network is selected as the commonly used sigmoid function, the sparsity parameter is 0.05, the learning rate is 0.01, and the maximum number of training is 10000 times. Considering the network performance and system burden, four SAE networks are used in the embodiment of the application. The performance indicators of the entire network training under different numbers of hidden layer nodes of SAE are shown in Table 3.

[0202] Table 3 Influence of different network structures on SAE network performance

[0203]

[0204] In Table 3, the six values in the first column represent the number of input layer nodes, the number of hidden layer nodes of the four SAE networks, and the number of output layer nodes. It can be seen that the number of input layer nodes is 7 and the number of output layer nodes is 1, which is unchanged, and the number of hidden layer nodes of the four SAEs is changed for evaluation.

[0205] Overall, when the network structure is 7-14-16-10-8-1, the convergence efficiency and solution accuracy of the method are obviously improved, and the overall evaluation index is more superior, and the network structure is finally selected as 7-14-16-10-8-1. The network trained in layers is used as a whole for global optimization, and the entire training process of the deep fitting network based on SAE is completed. In order to better show the model fitting performance, the network after global optimization is tested for fitting performance on the test set, and the network performance indicators are counted, as shown in Table 4.

[0206] Table 4 Performance index statistics of real-time solution model based on SAE

[0207]

[0208] As can be seen from Table 4, compared with only pre-training, the network error after global optimization is greatly reduced, and the network fitting performance is significantly improved, which shows that the network after global optimization can better fit the TCR boundary.

[0209] (III) Comparison of offline simulation and real-time solving performance

[0210] Further investigate the fitting effect of the SAE solving model, taking the TCR solving under the situation shown in Table 1 as an example, compare the fitting results of the deep network corresponding to the target real-time solving model with the offline simulation results of the offline solving model, and the results are shown in Figure 11 Figure 11 is a comparison chart of the real-time solving and offline simulation results of the tactical control distance in the embodiment of the application. In order to more obviously highlight the simulation comparison effect, the offline simulation results and the fitting results are distinguished by solid lines and asterisks respectively. Figure 11 In the figure, the solid line and the asterisk correspond to the TCR offline simulation results of the offline solving model and the TCR real-time fitting simulation results of the target real-time solving model respectively. As can be seen from Figure 11 , there is an inconsistent fitting result in the area where the target entry angle is 0°, and the reason is that the fitting values at 0° and 360° are not exactly equal. Overall, the solving results of the deep fitting network based on SAE are highly consistent with the offline simulation results, and the fitted TCR can closely follow the change trend of the offline simulation value, better reflect the boundary characteristics of TCR, and the overall fitting effect of the network is excellent.

[0211] In order to analyze the timeliness of the model solving, 100 samples are randomly selected from the test set for testing, and the offline simulation and real-time simulation time is counted as shown in the attached Figure 12 . All solving processes are carried out on a computer with Windows 10 system, main frequency 3.6GHz, and memory 3.96GB. As can be seen from Figure 12 , the average time of real-time solving one data sample is controlled within 12ms. As can be seen, the real-time solving method of the tactical control distance proposed in the embodiment of the application has fast solving speed and can meet the timeliness requirement. In combination with Figure 11 and Figure 12 , compared with the offline simulation model, the TCR real-time solving model based on SAE greatly improves the solving timeliness on the basis of realizing similar solving accuracy, and the overall performance evaluation index is better.

[0212] ​To sum up, the embodiment of the present application uses the tactical control range (TCR) for decision information support of an air combat simulation system, the proposed tactical control range calculation model training method in an air combat simulation environment and the real-time tactical control range calculation method in the air combat simulation environment have excellent situation expression forms, can meet the demand of decision information support in a high dynamic and strong real-time simulation environment, effectively make up for the lack of air combat node information, and have important significance for improving the beyond-visual-range air combat efficiency of fighter aircrafts in air combat confrontation games and air combat simulation systems.

[0213] In the description of the present application, the terms "first", "second" are only used for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0214] The above only describes the preferred embodiments of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A training method for a tactical control distance calculation model in an air combat simulation environment, characterized in that, The application relates to a method for calculating a tactical control distance of an air-to-air missile, and belongs to the field of air combat simulation. The method comprises the following steps: An offline calculation model of the tactical control distance is obtained by constructing a three-dimensional motion model and a constraint model of a carrier aircraft and an air-to-air missile in an over-the-horizon air combat, and setting a simulation limit condition; For each set of obtained sample motion data, a tactical control distance value corresponding to the set of sample motion data is obtained by performing motion simulation of an enemy missile and the carrier aircraft by using the set of sample motion data, a preset maximum initial search range of a boundary of the tactical control distance, and the offline calculation model; wherein the set of sample motion data represents the motion state of the carrier aircraft and the enemy aircraft, and comprises the speed of the carrier aircraft, the height of the carrier aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, a target entry angle, a target azimuth angle and a carrier aircraft maneuvering overload; Each set of sample motion data and the corresponding tactical control distance value are combined as a sample data set, and a sample database is obtained based on all the sample data sets; An initial real-time calculation model of the tactical control distance is constructed based on a sparse self-encoding network, and the initial real-time calculation model is trained by using the sample database, so that a target real-time calculation model trained is obtained, which is used for outputting a corresponding tactical control distance value for any set of inputted to-be-tested motion data; The method comprises the following steps: For the search range of the boundary of the tactical control distance used in the current iteration, a bisection point thereof is calculated; wherein the search range of the boundary of the tactical control distance used in the first iteration is the maximum initial search range of the boundary of the tactical control distance; The bisection point corresponding to the current iteration and the set of sample motion data are inputted into the offline calculation model, and a miss distance value is outputted; It is judged whether the miss distance value indicates that the carrier aircraft is not hit by the missile, and the search range of the boundary of the tactical control distance used in the current iteration is correspondingly reduced based on different judgment results of whether the miss distance value indicates that the carrier aircraft is not hit by the missile; It is judged whether the search range of the reduced boundary of the tactical control distance meets a preset error requirement; If not, the search range of the reduced boundary of the tactical control distance is taken as the search range of the boundary of the tactical control distance used in the next iteration, and the step of calculating the bisection point of the search range of the boundary of the tactical control distance used in the current iteration is returned; 2. The method of claim 1, wherein the method is performed in a simulated air combat environment. If yes, the bisection point of the search range of the reduced boundary of the tactical control distance is calculated and taken as the tactical control distance value corresponding to the set of sample motion data. The three-dimensional motion model and the constraint model thereof comprise:

3. The method of claim 1, wherein the method is performed in a simulated air combat environment. A three-degree-of-freedom motion model of the carrier aircraft, a three-degree-of-freedom motion model of the missile, a relative motion model of the missile-carrier aircraft and a guidance law model. The simulation limit condition comprises: When the missile-target distance is smaller than the maximum damage radius of the missile and the time limit condition and the speed limit condition are not triggered, the missile is determined to successfully hit the target, otherwise the missile attack is determined to fail. The time limit condition is that when the missile flight time is greater than the controllable flight time of the missile, the missile energy is exhausted and cannot hit the target; The speed limit condition is that when the missile speed is less than the minimum flight speed of the missile, the maneuverability of the missile is reduced and cannot hit the target.

4. The method of claim 3, wherein the method further comprises: Based on whether the miss distance value indicates that the aircraft is not hit by the missile, the search range of the tactical control distance boundary used in the current iteration is correspondingly narrowed, including: If the miss distance value indicates that the aircraft is hit by the missile, the lower limit value in the search range of the tactical control distance boundary used in the current iteration is replaced by the bisection point corresponding to the current iteration; If the miss distance value indicates that the aircraft is not hit by the missile, it is judged whether at least one of the time limit condition and the speed limit condition is met; If not, return to the step of inputting the bisection point corresponding to the current iteration and the set of sample motion data into the offline solving model to simulate with the original input data; If it is met, the upper limit value in the search range of the tactical control distance boundary used in the current iteration is replaced by the bisection point corresponding to the current iteration.

5. The method of claim 3 or 4, wherein the method further comprises: The search range of the narrowed tactical control distance boundary is judged whether it meets the preset error requirement, including: Judge whether the search range of the narrowed tactical control distance boundary meets the condition that the absolute value of the upper and lower limits is less than the preset boundary precision constant.

6. The method of claim 1, wherein, The structure of the initial real-time solving model includes: An input layer, four sparse auto-encoding networks and an output layer; wherein each sparse auto-encoding network only contains one hidden layer; Wherein, the number of nodes of the input layer is 7; the number of nodes of the hidden layer of the four sparse auto-encoding networks is 14, 16, 10 and 8 respectively; the number of nodes of the output layer is 1; the number of nodes of the hidden layer of the four sparse auto-encoding networks is determined by network performance influence evaluation optimization.

7. The method of claim 6, wherein the method further comprises: The network training of the initial real-time solving model using the sample database is performed to obtain a trained target real-time solving model, including: Each sparse auto-encoding network in the initial real-time solving model is pre-trained separately to obtain the corresponding pre-trained network weight; wherein the first sparse auto-encoding network is pre-trained using the sample database; from the second sparse auto-encoding network, the input for pre-training is the excitation of each neuron in the hidden layer of the previous sparse auto-encoding network after pre-training; After superimposing all pre-trained sparse auto-encoding networks and adding the input layer and the output layer, the global optimization of network weights of the overall network structure is performed using the sample database to obtain a trained target real-time solving model.

8. A method for real-time calculation of tactical control distance in an air combat simulation environment, characterized in that, Including: For any set of acquired motion data in the air combat simulation environment, the set of motion data to be tested is input into the pre-trained target real-time solving model to obtain the corresponding tactical control distance value of the set of motion data to be tested; wherein any set of motion data to be tested represents the motion state of the aircraft and the enemy aircraft, including the speed of the aircraft, the height of the aircraft, the speed of the enemy aircraft, the height of the enemy aircraft, the target entry angle, the target azimuth angle and the aircraft maneuvering overload; the target real-time solving model is trained according to the air combat simulation environment tactical control distance solving model training method of any one of claims 1 to 7.

9. The method of claim 8, wherein the method further comprises: After obtaining the corresponding tactical control distance value for any set of motion data to be tested, the method further comprises: The tactical control distance boundary is obtained by sequentially connecting the tactical control distance values obtained by each set of motion data to be tested under the same air combat simulation environment and in a continuous situation; wherein only the target entry angle is different in each set of motion data to be tested under the same air combat simulation environment and in a continuous situation, and each target entry angle changes by a preset step.

Images