Multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO
Through the multi-air air combat maneuver decision-making method of entropy weight TOPSIS and PPO, the problem that the PPO method is difficult to adapt to the changes in the number of enemy-to-we-drone drones and the complex air combat environment in the multi-air air combat environment is solved, and the effective target evaluation and maneuvering of the drone is achieved Decision-making improves the accuracy and adaptability of independent decision-making.
Patent Information
- Application Number
- CN202411259054.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2024-11-26
AI Technical Summary
The existing PPO methods are difficult to adapt to the high dynamic changes in the number of enemy-to-we-drone drones and complex air combat environments in multi-aircraft air combat environments, and it is difficult to guide drones to make effective target selection and maneuver decisions.
Using multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO, the air combat indicator data is classified, forwarded and standardized, and the information entropy and entropy weights are calculated to find the optimal and worst solutions of the solution matrix. Calculate the comprehensive distance score of the drone target, and build the state space and action space of the PPO algorithm to realize maneuver decision-making.
Effectively adapt to the highly dynamic mission environment of multi-aircraft air combat and dynamically select enemy targets, improving the adaptability and decision-making accuracy of PPO algorithms in multi-aircraft air combat, and being able to gain an advantage in autonomous decision-making of drones.
Smart Images

Figure CN119026031A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of multi-aircraft decision making, and more specifically to a multi-aircraft air combat maneuver decision making method based on entropy weight TOPSIS and PPO. Background Art
[0002] Multi-UAV air combat is a typical combat mode in which air combat is developing towards intelligence. Each UAV is equipped with an AI agent with autonomous decision-making capabilities. It collects air combat information based on onboard sensors to complete situation assessment and maneuvering decisions, creating conditions for weapon launch and air combat victory. Deep reinforcement learning is a cutting-edge method for building AI agents. For example, the proximal policy optimization (PPO) method is used to construct a neural network mapping from air combat state information to maneuvering actions, and the neural network is trained through a reward mechanism and air combat data sets. The trained neural network can realize autonomous decision-making in the UAV air combat maneuvering process.
[0003] At present, the PPO method constructs a neural network mapping of air combat state information to maneuvering actions in an end-to-end manner, which makes the air combat state space dimension fixed and difficult to adapt to the multi-aircraft air combat environment with highly dynamic changes in the number of enemy and friendly drones. At the same time, when facing a complex air combat environment, such as different types of enemy drones, different attack postures, and different comprehensive air combat capabilities, the PPO method is difficult to guide drones to make effective target selection and maneuvering decisions.
[0004] Therefore, how to provide a multi-aircraft air combat maneuver decision-making method for threat assessment and ranking of multiple enemy aircraft and guiding the PPO algorithm to make maneuver decisions is an urgent problem that technicians in this field need to solve. Summary of the invention
[0005] In view of this, the present invention provides a multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO to solve the problems in the background technology.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] The present invention discloses a multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, comprising:
[0008] The air combat index data is classified, and the decision matrix Z is obtained by forward and normalization processing respectively;
[0009] According to the decision matrix Z, the information entropy of the air combat index data is calculated and normalized to obtain the index entropy weight w j ;
[0010] Solve the optimal solution and the worst solution of the decision matrix Z according to the entropy weight wj Weighted calculation of the optimal distance and the worst distance of each enemy drone target;
[0011] Calculate the comprehensive distance score of each enemy drone target and sort the score results;
[0012] Construct the state space and action space of the PPO algorithm;
[0013] Construct a PPO algorithm strategy neural network mapping model based on state space and action space;
[0014] According to the comprehensive distance score sorting results, the enemy and friendly status information is input into the PPO algorithm strategy neural network mapping model to complete the maneuver decision.
[0015] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO, the classification of air combat index data includes: extremely large index data, extremely small index data, intermediate index data, and interval index data according to the different directions in which the index data affects the final scoring result.
[0016] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, the steps of respectively performing forward processing on the extremely large index data, the extremely small index data, the intermediate index data, and the interval index data to obtain the intermediate matrix X are as follows:
[0017] The forward calculation of the very small indicator is as follows:
[0018]
[0019] Where x i It represents the data of a certain indicator of the drone numbered i, and max is the maximum value of the indicator data of all drones. is the indicator data after positive conversion;
[0020] The forward calculation of the intermediate indicator data is as follows:
[0021] M = max{|xi-xbest|};
[0022]
[0023] Where x best is the optimal value of a certain indicator data of the drone, M is the distance x among all drones best The farthest value.
[0024] The forward calculation of interval indicator data is as follows:
[0025] M=max{a-min{x i},max{xi}-b};
[0026]
[0027] Where [a, b] is the optimal interval of the indicator data, and M is the value of the indicator that is farthest from the interval for all drones;
[0028] Extremely large indicator data does not need to be positively processed;
[0029] All the forward processed data are composed into the intermediate matrix X as follows:
[0030]
[0031] Where n represents the number of drones, and i = 1, 2, ..., n; m represents the number of drone indicators, and j = 1, 2, ..., m.
[0032] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, each element of the intermediate matrix X is standardized as follows:
[0033] x ij are the elements of the intermediate matrix X;
[0034] Where z ij is the standardized element, which is constructed into the decision matrix Z as follows:
[0035]
[0036] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, the information entropy of the air combat index data is calculated according to the decision matrix Z and normalized to obtain the index entropy weight w j The specific steps are as follows: Calculate each element z in the decision matrix Z ij The probability p of the same indicator j ij And the probability matrix P is as follows:
[0037]
[0038] The information entropy of each air combat index data is calculated according to the probability matrix P as follows:
[0039]
[0040] In the formula, if an element p ij Take the value as 0, let p ij ln(p ij ) is set to 0 to avoid calculation errors;
[0041] Calculate the utility value d of each air combat index data information j And normalized to get the index entropy weight w j as follows:
[0042] d j =1-e j ;
[0043]
[0044] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, the optimal solution and the worst solution of the decision matrix Z are solved, and the optimal solution and the worst solution are calculated according to the entropy weight w j The specific steps for weighted calculation of the optimal distance and the worst distance of each enemy drone target are as follows:
[0045] According to the decision matrix Z, the optimal solution z of each air combat index data is calculated respectively + and the worst solution z - as follows:
[0046]
[0047] Calculate the optimal distance of each drone based on the decision matrix Z and the worst distance as follows:
[0048]
[0049] Among them, w j is the index entropy weight, Z ij are the elements of the decision matrix Z.
[0050] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, the specific steps of calculating the comprehensive distance score of each enemy UAV target and sorting the score results are as follows:
[0051] Calculate the comprehensive distance score S of each drone i And the normalized results as follows:
[0052]
[0053] in, is the optimal distance for each drone, is the worst distance for each drone;
[0054] The comprehensive distance scores of each drone are sorted from large to small as follows:
[0055]
[0056] Where r is the number of the drone according to the sorting result, and n is the number of the drone with the smallest sorting result.
[0057] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, the state space of the PPO algorithm includes:
[0058] The state space of PPO is divided into the UAV's own state information and the enemy-friend relative situation information. The state space matrix after division is as follows:
[0059] S=[S1,S 11 ,S 12 ,...,S 1r ];
[0060] Where S1 is the state information of the drone itself, S 1r is the relative situation information between the UAV and the target enemy aircraft numbered r;
[0061] The drone's own status information includes its own altitude, roll angle and pitch angle, its ground coordinate axis velocity and scalar total velocity; the relative situation information includes its relative altitude, relative speed, aspect angle, antenna array angle and distance to the enemy aircraft.
[0062] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, the action space of the PPO algorithm includes:
[0063] The action space of PPO is defined as four-dimensional discrete actions [a1, a2, a3, a4], which represent the control commands of aileron, elevator, rudder and throttle respectively; the discrete actions are normalized and used as the control command input of the UAV;
[0064]
[0065] Where a represents the discrete action value, which is an integer; N and B represent the normalized weight and bias. Indicates the normalized action value.
[0066] Preferably, in the above-mentioned multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, a PPO algorithm strategy neural network mapping model is constructed, specifically:
[0067] Construct a multilayer perceptron module. The input layer of the multilayer perceptron module is the state space information, the middle layer is the hidden layer, the layers are connected by the linear function Linear, the ReLU activation function and the normalization function LayerNorm, and the last layer is the gated recurrent unit;
[0068] Construct an action output module. The action output module passes through two hidden layers and one action output layer. The hidden layers are composed of linear functions Linear, ReLU activation function and normalization function LayerNorm. There is a linear function Linear between the hidden layer and the action output layer. The action output is the probability value of the discrete action, and the discrete action with the largest probability is taken as the final output.
[0069] It can be seen from the above technical solution that compared with the prior art, the present invention discloses a multi-aircraft air combat maneuvering decision method based on entropy weight TOPSIS and PPO to effectively adapt to the high dynamic mission environment of multi-aircraft air combat. In air combat, with the increase (reinforcement) and decrease (battle damage) in the number of enemy and our fighters, the entropy weight TOPSIS method is used to dynamically select enemy targets, solving the problem of fixed input dimension of the PPO algorithm. In air combat, the model, performance, and air combat situation of enemy fighters all increase the complexity of air combat. Through the entropy weight TOPSIS method, various combat factors of enemy fighters can be comprehensively considered, threat assessment and ranking can be performed for enemy fighters, and suitable targets can be selected as the basis for our aircraft maneuvering decision. Multi-aircraft air combat maneuvering decision tasks can be effectively and autonomously executed. The neural network mapping of air combat situation to maneuvering action instructions is realized through the PPO algorithm, and maneuvering decisions are completed according to air combat situation information without pilot / human intervention, and an advantageous situation is occupied during the maneuvering process, creating conditions for weapon launch and air combat victory. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0071] Figure 1 It is the entropy weight TOPSIS and PPO multi-aircraft air combat maneuver decision flow chart;
[0072] Figure 2 Schematic diagram of state space and action space;
[0073] Figure 3 It is a schematic diagram of the multi-layer perceptron module;
[0074] Figure 4 This is a schematic diagram of the action output module. DETAILED DESCRIPTION
[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0076] The embodiment of the present invention discloses a multi-aircraft air combat maneuver decision method based on entropy weight TOPSIS and PPO, such as Figure 1 As shown, the following steps are included:
[0077] Step 1: Classify the air combat index data, perform forward and standardization processing respectively to obtain the decision matrix Z;
[0078] Step 2: Calculate the information entropy of each indicator based on the decision matrix Z and normalize it to obtain the indicator entropy weight w j ;
[0079] Step 3: Solve the optimal solution and the worst solution of the decision matrix Z according to the entropy weight w j Weighted calculation of the optimal distance and the worst distance of each enemy drone target;
[0080] Step 4: Calculate the comprehensive distance score of each enemy UAV target and sort the score results;
[0081] Step 5: Construct the state space and action space of the PPO algorithm;
[0082] Step 6: Construct a PPO algorithm strategy neural network mapping model based on the state space and action space;
[0083] Step 7: Based on the comprehensive distance score sorting results, input the enemy and friendly status information into the PPO algorithm strategy neural network mapping model to complete the maneuver decision.
[0084] In step 1, each index data of the air combat is forwarded and standardized to obtain the decision matrix Z. Assume that the position of a certain drone of our side at a certain moment of the air combat is the coordinate origin [0,0,0], and the heading is [0,0,1]. The serial numbers and air combat data of the five enemy drones are shown in Table 1. Since the various index data contained in the air combat affect the final scoring result in different directions, some indicators play a positive role in the scoring, that is, when other indicators are the same, the larger the index value, the larger the score of the entire scheme. Some indicators play a reverse role in the scoring, that is, when other indicators are the same, the larger the index value, the smaller the score of the entire scheme.
[0085] Table 1
[0086]
[0087] According to the direction of influencing the score, air combat indicators are divided into four categories: extremely large indicators, extremely small indicators, intermediate indicators, and interval indicators. Extremely large indicators are indicators with larger data, the better, and can also be called benefit indicators. In the evaluation process, the larger the value of the indicator, the greater the benefit, and the better the performance of the corresponding target scheme participating in the evaluation in terms of this indicator, and the higher the score ranking. Extremely small indicators are indicators with smaller data, the better, and can also be called cost indicators. In the scoring process, the smaller the value of the indicator, the greater the benefit, and the better the performance of the corresponding target scheme participating in the evaluation in terms of this indicator, and the higher the score ranking. Intermediate indicators are evaluation indicators with an optimal value in the middle of the data, and can also be called target indicators. The highest evaluation state of this indicator is the intermediate optimal value. The smaller the deviation from the optimal value, the better the performance of the evaluated target scheme in terms of this indicator, and the higher the score ranking. The greater the deviation from the optimal value, the worse the performance of the evaluated target scheme in terms of this indicator, and the lower the score ranking. An interval indicator is an evaluation indicator that has an optimal interval in the middle of the data. The indicator value is optimal within the interval. Outside the optimal interval, the smaller the gap between the data of this type of indicator and the interval boundary, the better the performance of the target solution being evaluated in terms of the indicator, and the higher the score ranking. The larger the gap with the interval boundary, the worse the performance of the target solution being evaluated in terms of the indicator, and the lower the score ranking.
[0088] Data positive processing is to process the above four types of data into extremely large indicators to make the data positively unified.
[0089] In order to eliminate the influence of different indicator dimensions, it is necessary to standardize the data obtained after normalization. Standardization is a commonly used data preprocessing method, also known as feature scaling. The purpose is to process data of different magnitudes, units and distributions to facilitate comparison or unified calculation.
[0090] Step 1.1: The forward calculation of the extremely small indicator data is as follows:
[0091]
[0092] Where x i It represents the data of a certain indicator of the drone numbered i, and max is the maximum value of the indicator data of all drones. It is the indicator data after positive conversion.
[0093] The forward calculation of the intermediate indicator data is as follows:
[0094] M=max{|x i -x best |};
[0095]
[0096] Where x best is the optimal value of a certain indicator data of the drone, M is the distance x among all drones best The farthest value.
[0097] The forward calculation of interval indicator data is as follows:
[0098] M=max{a-min{x i},max{x i}-b};
[0099]
[0100] Where [a, b] is the optimal interval of the indicator data, and M is the value of the indicator that is farthest from the interval for all drones.
[0101] Very large indicator data does not need to be normalized.
[0102] All the data after forward processing are composed into matrix X as follows:
[0103]
[0104] Where n represents the number of drones, and i = 1, 2, ..., n; m represents the number of drone indicators, and j = 1, 2, ..., m.
[0105] The speed of the drone, the type of drone and the target air combat capability in the air combat are taken as the large-scale indicators, and the distance, target azimuth, our azimuth and relative heading angle are taken as the small-scale indicators. The air combat data is forward-transformed to obtain the matrix X:
[0106]
[0107] Step 1.2: Standardize each element of matrix X as follows:
[0108]
[0109] Where z ij is the standardized element, which is constructed into the decision matrix Z as follows:
[0110]
[0111] The decision matrix Z is obtained by standardizing the forward matrix X:
[0112]
[0113] In step 2, the information entropy of each indicator is calculated according to the decision matrix Z and normalized to obtain the indicator entropy weight wj . Information entropy is a physical quantity used to measure the complexity of random signals. It mainly expresses the uncertainty or confusion of random signals in numerical form. The higher the information entropy value of an indicator, the higher the degree of information confusion of the indicator, the more concentrated the probability distribution of the information, and the smaller the impact on the overall plan. The information utility value is the positive and normalized processing of the information entropy to make the weight difference between indicators more obvious. The higher the information utility value, the smaller the information entropy, and the greater the weight that should be assigned to the indicator. Specifically:
[0114] Step 2.1: Calculate the elements z in the decision matrix Z ij The probability p of the same indicator j ij And the probability matrix P is as follows:
[0115]
[0116] Calculate the elements z in the decision matrix Z ij The probability p of the same indicator j ij And the probability matrix P is:
[0117]
[0118] Step 2.2: Calculate the information entropy of each indicator based on the probability matrix P as follows:
[0119]
[0120] In the formula, if an element p ij Take the value as 0, let p ij ln(p ij ) is set to 0 to avoid calculation errors.
[0121] According to the probability matrix P, the information entropy of each indicator is calculated as follows: [0.7020.8560.7100.7980.9950.9590.878];
[0123] Step 2.3: Calculate the information utility value d of each indicator j And normalized, get the index entropy weight w j as follows:
[0124] d j =1-e j ;
[0125]
[0126] Calculate the information utility value d of each indicator j And normalized, the index entropy weight is obtained as: [0.271 0.130 0.263 0.183 0.004 0.037 0.111];
[0128] The step 3 calculates the optimal solution and the worst solution of the overall solution according to the decision matrix Z, and calculates the optimal solution and the worst solution according to the entropy weight w j The optimal distance and the worst distance of each enemy drone target are calculated by weighted calculation. Specifically:
[0129] Step 3.1: Calculate the optimal solution z for each indicator based on the decision matrix Z + and the worst solution z - as follows:
[0130]
[0131] According to the decision matrix Z, calculate the optimal solution z for each indicator + and the worst solution z - for:
[0132] z + =[0.782 0.572 0.740 0.743 0.520 0.657 0.557];
[0133] z - =[0 0000.3600.2630.014];
[0134] Step 3.2: Calculate the optimal distance of each drone based on the decision matrix Z and the worst distance as follows:
[0135]
[0136] Calculate the optimal distance d of each UAV according to the decision matrix Z + and the worst distance d - for:
[0137] d + =[0.6200.5170.4130.4180.388];
[0138] d - =[0.2610.4720.4610.5300.439];
[0139] The step 4 calculates the comprehensive distance score of each enemy UAV target and sorts the score results, specifically:
[0140] Step 4.1: Calculate the comprehensive distance score S of each drone i And the normalized results as follows:
[0141]
[0142]
[0143] Calculate the comprehensive distance score S of each drone i And the normalized results for:
[0144] S=[0.272 0.477 0.527 0.559 0.531];
[0145]
[0146] Step 4.2: Sort the comprehensive distance scores of each drone from largest to smallest as follows:
[0147]
[0148] Where r is the number of the drone according to the sorting result.
[0149] The comprehensive distance scores of each drone are sorted from large to small as follows:
[0150]
[0151] The larger the comprehensive distance score of a drone, the greater the threat it poses to our aircraft, and our aircraft should pay more attention to its behavior and actions to make maneuvering decisions.
[0152] The step 5 constructs the state space and action space of the PPO algorithm, specifically:
[0153] Step 5.1: Divide the state space of PPO into the UAV's own state information and the enemy-friend relative situation information. The state space matrix after division is as follows:
[0154] S=[S1,S 11 ,S 12 ,...,S 1r ];
[0155] Where S1 is the state information of the drone itself, S 1r It is the relative situation information between the UAV and the target enemy aircraft numbered r.
[0156] If the PPO algorithm only focuses on the flight status of a certain enemy drone, the state space should contain the state information S1 of our drone and the relative situation information S between our drone and the target enemy drone. 11 In this example, S1 is constructed as a matrix Among them, H θ are the height, roll angle and pitch angle of our aircraft respectively, V x ,V y,V z ,V c They are the ground coordinate axis velocity and scalar total velocity of our aircraft respectively. 11 Constructed as [H r ,V r ,AA,ATA,R], respectively represent the relative height, relative speed, aspect angle, antenna array angle and distance between our aircraft and the enemy aircraft. The dimension of the state space is 14, such as Figure 2 shown.
[0157] Step 5.2: Define the action space of PPO as four-dimensional discrete actions [a1, a2, a3, a4], which represent the control commands of aileron, elevator, rudder and throttle respectively. The discrete actions are normalized and used as the control command input of the UAV.
[0158]
[0159] Where a represents the discrete action value, which is an integer. N and B represent the normalized weight and bias. Indicates the normalized action value.
[0160] Let the discrete action values of actions a1, a2, and a3 be 0 to 40, and the discrete action value of a4 be 0 to 29, respectively, according to the formula Japanese style Normalize a1, a2, and a3 to [-1, 1], and normalize a4 to [0.4, 0.9], thereby forming the control instructions for the aileron, elevator, rudder, and throttle.
[0161] The step 6 constructs a PPO algorithm strategy neural network mapping model, specifically:
[0162] Step 6.1: Construct a multi-layer perceptron module. The input layer of the module is state space information with a dimension of 14. The middle layer is a hidden layer. The layers are connected by linear functions Linear, ReLU activation functions, and normalization functions LayerNorm. The last layer is a gated recurrent unit. The dimensions of the middle and last layers are both 128. Figure 3 shown.
[0163] Step 6.2: Construct an action output module. The module has two hidden layers and an action output layer. The hidden layers are connected by linear functions, ReLU activation functions, and LayerNorm. The dimension is 128. The hidden layer and the action output layer are connected by linear functions. The dimension of the action output layer is 4. The action output is the probability value of the discrete action, and the discrete action with the largest probability is taken as the final output, such as Figure 4 shown.
[0164] In step 7, according to the comprehensive distance score sorting result, the enemy and friendly state information is input into the neural network model to complete the maneuver decision. In this example, the PPO algorithm only focuses on the flight status of a certain enemy drone, so the state information of the drone ranked first is used as input, and the action output is calculated by the neural network model in step 6 to realize the maneuver action decision of our aircraft.
[0165] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0166] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. Multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO, It is characterized in that include: The air combat index data is classified, and the decision matrix Z is obtained by forward and normalization processing respectively; According to the decision matrix Z, the information entropy of the air combat index data is calculated and normalized to obtain the index entropy weight w j ; Solve the optimal solution and the worst solution of the decision matrix Z according to the entropy weight w j Weighted calculation of the optimal distance and the worst distance of each enemy drone target; Calculate the comprehensive distance score of each enemy drone target and sort the score results; Construct the state space and action space of the PPO algorithm; Construct a PPO algorithm strategy neural network mapping model based on state space and action space; According to the comprehensive distance score sorting results, the enemy and friendly status information is input into the PPO algorithm strategy neural network mapping model to complete the maneuver decision.
2. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that The classification of air combat index data includes: extremely large index data, extremely small index data, intermediate index data, and interval index data according to the different directions in which the index data affects the final scoring results.
3. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that The steps for forward processing of extremely large index data, extremely small index data, intermediate index data, and interval index data to obtain the intermediate matrix X are as follows: The forward calculation of the very small indicator is as follows: Where x i It represents the data of a certain indicator of the drone numbered i, and max is the maximum value of the indicator data of all drones. is the indicator data after positive conversion; The forward calculation of the intermediate indicator data is as follows: M = max{|xi-xbest|}; Where x best is the optimal value of a certain indicator data of the drone, M is the distance x among all drones best The farthest value; The forward calculation of interval indicator data is as follows: M=max{a-min{x i },max{x i }-b}; Where [a, b] is the optimal interval of the indicator data, and M is the value of the indicator that is farthest from the interval for all drones; Extremely large indicator data does not need to be positively processed; All the forward processed data are composed into the intermediate matrix X as follows: Where n represents the number of drones, and i = 1, 2, ..., n; m represents the number of drone indicators, and j = 1, 2, ..., m.
4. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 3, It is characterized in that The elements of the intermediate matrix X are standardized as follows: x ij are the elements of the intermediate matrix X; Where z ij is the standardized element, which is constructed into the decision matrix Z as follows:
5. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that According to the decision matrix Z, the information entropy of the air combat index data is calculated and normalized to obtain the index entropy weight w j The specific steps are as follows: Calculate each element z in the decision matrix Z ij The probability p of the same indicator j ij And the probability matrix P is as follows: The information entropy of each air combat index data is calculated according to the probability matrix P as follows: In the formula, if an element p ij Take the value as 0, let p ij ln(p ij ) is set to 0 to avoid calculation errors; Calculate the utility value d of each air combat index data information j And normalized, get the index entropy weight w j as follows: d j =1-e j ; 6. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that Solve the optimal solution and the worst solution of the decision matrix Z according to the entropy weight w j The specific steps for weighted calculation of the optimal distance and the worst distance of each enemy drone target are as follows: According to the decision matrix Z, the optimal solution z of each air combat index data is calculated respectively + and the worst solution z - as follows: Calculate the optimal distance of each drone based on the decision matrix Z and the worst distance as follows: Among them, w j is the index entropy weight, Z ij are the elements of the decision matrix Z.
7. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that The specific steps for calculating the comprehensive distance score of each enemy drone target and sorting the score results are as follows: Calculate the comprehensive distance score S of each drone i And the normalized results as follows: in, is the optimal distance for each drone, is the worst distance for each drone; The comprehensive distance scores of each drone are sorted from large to small as follows: Where r is the number of the drone according to the sorting result, and n is the number of the drone with the smallest sorting result.
8. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that The state space of the PPO algorithm includes: The state space of PPO is divided into the UAV's own state information and the enemy-friend relative situation information. The state space matrix after division is as follows: S=[S 1 ,S 11 ,S 12 ,...,S 1r ]; Where S 1 is the drone’s own status information, S 1r is the relative situation information between the UAV and the target enemy aircraft numbered r; The drone's own status information includes its own altitude, roll angle and pitch angle, its ground coordinate axis velocity and scalar total velocity; the relative situation information includes its relative altitude, relative speed, aspect angle, antenna array angle and distance to the enemy aircraft.
9. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that The action space of the PPO algorithm includes: The action space of PPO is defined as a four-dimensional discrete action [a 1 ,a 2 ,a 3 ,a 4 ], which represent the control commands of aileron, elevator, rudder and throttle respectively; the discrete actions are normalized and used as the control command input of the UAV; Where a represents the discrete action value, which is an integer; N and B represent the normalized weight and bias. Indicates the normalized action value.
10. The multi-aircraft air combat maneuver decision-making method based on entropy weight TOPSIS and PPO according to claim 1, It is characterized in that Construct the PPO algorithm strategy neural network mapping model, specifically: Construct a multi-layer perceptron module. The input layer of the multi-layer perceptron module is the state space information, the middle layer is the hidden layer, the layers are connected by the linear function Linear, the ReLU activation function and the normalization function LayerNorm, and the last layer is the gated recurrent unit; Construct an action output module. The action output module passes through two hidden layers and one action output layer. The hidden layers are composed of linear functions Linear, ReLU activation function and normalization function LayerNorm. There is a linear function Linear between the hidden layer and the action output layer. The action output is the probability value of the discrete action, and the discrete action with the largest probability is taken as the final output.
Citation Information
Patent Citations
TOPSIS-method multi-target threat ordering method based on information entropy
CN103246818A
Intelligent agent confrontation method based on artificial intelligence
CN115936055A
Design method of adaptive reinforcement learning decision model facing priori knowledge
CN117454738A
Multi-path video transmission method based on multi-agent deep reinforcement learning
CN118233671A
