Data knowledge dual-drive situation assessment method, model, medium and electronic equipment

Abstract

The invention discloses a data knowledge dual-drive situation assessment method and model, a medium and electronic equipment. The method comprises the following steps: carrying out formalized coding on a current frame situation and a preorder frame situation, and correspondingly generating a current single-frame coding situation and a preorder single-frame coding situation; calculating an information value of the preorder single-frame coding situation, and performing threshold comparison and sampling to generate a multi-frame sequence coding situation with the current single-frame coding situation; executing residual text convolution, and outputting an instant revenue value; deducing a current frame situation random game to a final game to generate a game tree; executing decentralized evaluation, and generating an instant evaluation value of each stage of the game tree; executing a Bellman evaluation process, and calculating a deduction evaluation value; and calculating a comprehensive evaluation value of the current frame situation according to the instant evaluation value and the deduction evaluation value. According to the method, the accuracy of intelligent agent situation judgment in war game deduction is improved, and the autonomy, reliability and flexibility of intelligent agent game decision making in a team-level tactical confrontation environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of situation assessment, and in particular to a data-knowledge dual-driven situation assessment method, a data-knowledge dual-driven situation assessment model, a computer-readable storage medium, and an electronic device. Background Art

[0002] Battlefield situation assessment is the core link of combat command activities and the premise and basis for the effective implementation of combat command decisions. With the widespread application of a large number of early warning detection equipment and sensors, the large amount of battlefield situation data obtained has far exceeded the commander's analysis and processing capabilities. The fast pace of informationized warfare has put forward higher requirements for the timeliness and accuracy of situation assessment. Therefore, it is necessary to assist commanders in completing situation assessment through intelligent means. Intelligent battlefield situation assessment is the basis and key to realizing intelligent combat command. Intelligent situation assessment can not only effectively improve the speed of commanders in assessing the situation, but also make up for the limitations of commanders' subjective judgment and enhance the accuracy of situation assessment. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a situation assessment method driven by both data and knowledge, which is conducive to improving the accuracy of intelligent agent situation judgment in war game simulation and enhancing the autonomy, reliability and flexibility of intelligent agent game decision-making in a team-level tactical confrontation environment.

[0004] The present invention also proposes a situation assessment model driven by both data and knowledge.

[0005] The present invention also proposes a computer-readable storage medium for implementing the data-knowledge dual-driven situation assessment method.

[0006] The present invention also proposes an electronic device for implementing the data-knowledge dual-driven situation assessment method.

[0007] According to the data-knowledge dual-driven situation assessment method of the first aspect of the present invention, the method includes an immediate assessment of the immediate benefits of the action situation and a deduction assessment of the future benefits of the action situation, wherein the immediate assessment includes performing situation encoding and performing residual text convolution, and the deduction assessment includes performing a multi-stage game deduction process and performing a Bellman evaluation process. The method includes the following steps:

[0008] Formal encoding is performed on a current frame situation and a plurality of preceding frame situations that are temporally preceding the current frame situation, and a current single-frame encoding situation and a set including a plurality of the preceding single-frame encoding situations are correspondingly generated;

[0009] Calculating the information value of the preceding single-frame coding situation in the set, and performing threshold comparison and sampling to jointly generate a multi-frame sequence coding situation with the current single-frame coding situation;

[0010] Performing the residual text convolution on the multi-frame sequence encoding situation and outputting an immediate benefit value;

[0011] Deducing the current frame situation random game to the end game to generate a game tree;

[0012] Perform decentralized evaluation, use evaluation factors to calculate multi-factor evaluation values ​​at each stage of the game tree, and use adjacent superiority entropy weight method to weight the multi-factor evaluation values ​​to generate instant evaluation values ​​at each stage of the game tree;

[0013] Executing the Bellman evaluation process, and calculating the derivation evaluation value of the current frame situation according to the decentralized evaluation and the Bellman evaluation process;

[0014] According to the instant evaluation value and the deduced evaluation value, the comprehensive evaluation value of the current frame situation is weightedly calculated using the adjacent superiority entropy weight method.

[0015] According to the data-knowledge dual-driven situation assessment method of the embodiment of the present invention, an action data assessment model for calculating immediate benefits and a simulation deduction assessment method for calculating future benefits are established. The advantages and disadvantages of the battlefield situation are comprehensively evaluated through immediate benefits and future benefits, which provides a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and helps to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0016] In addition, the data-knowledge dual-driven situation assessment method according to the embodiment of the present invention also has the following additional technical features:

[0017] According to some embodiments of the present invention, the formal coding includes coding entity attributes, environmental attributes, entity position distribution, and target distribution. The coding of the entity attributes includes coding the entity camp, equipment type, force status, and current action category. The coding of the environmental attributes includes elevation coding of the combat area grid. The entity position distribution Pos is encoded according to the following formula:

[0018]

[0019] In the formula, [xc,yc]|c is the coordinate of the distribution center c of this type of entity, is the distance between the entity and the center of the distribution, yes The distance along the x-axis; yes The distance along the y-axis; the target distribution GoPos is encoded according to the following formula:

[0020] GoPos={Pos,Fac}

[0021] Where Pos is the target position; Fac is the target camp.

[0022] According to some embodiments of the present invention, the threshold comparison and sampling include:

[0023] When the information value of the preceding single-frame coding situation is greater than the set threshold value 0.12, the preceding single-frame coding situation is sampled to generate the multi-frame sequence coding situation; the information value Infor of the j-th preceding single-frame coding situation is calculated according to the following formula:

[0024]

[0025] In the formula, C j is the sum of the coding numbers of the jth preceding single-frame coding states; C j-1 is the sum of the coding numbers of the (j-1)th preceding single-frame coding situation; the C is calculated according to the following formula j :

[0026]

[0027] Where n j is the length of the jth preceding single frame encoding state; s j [k] is the jth preceding single frame coding state s j The k-th code number at time j.

[0028] According to some embodiments of the present invention, the multi-frame sequence coding situation constructs an ordered two-dimensional situation coding matrix with a length of Len and a width of 3 with a sampling step of 3, and the ordered two-dimensional situation coding matrix is ​​arranged in chronological order, wherein Len represents the length of a single-frame coding situation,

[0029] When the number of the single-frame coding situations in the multi-frame sequence coding situation is less than 3, the vacant positions of the ordered two-dimensional situation coding matrix are filled with all-0 single-frame coding situations with a length of Len to unify the output form of the multi-frame sequence coding situation.

[0030] According to some embodiments of the present invention, performing the residual text convolution on the multi-frame sequence encoding state includes:

[0031] Extracting long-range features of the multi-frame sequence encoding situation using a residual channel, wherein the residual channel includes a first convolutional layer and a first batch of normalization layers;

[0032] Extracting semantic features of the multi-frame sequence encoding situation using a text convolution channel, wherein the text convolution channel includes a plurality of convolution modules connected in series and a first random inactivation layer, and the convolution module includes a second convolution layer, a second batch normalization layer, a first linear rectification function layer, and a maximum pooling layer;

[0033] Fusing the features extracted by the residual channel and the text convolution channel;

[0034] Perform secondary feature extraction using a multilayer perceptron, wherein the multilayer perceptron includes a first linear layer, a second linear rectifier function layer, a second random inactivation layer, and a second linear layer;

[0035] The instantaneous return value is output using a normalized exponential function layer mapping.

[0036] According to some embodiments of the present invention, the random game of the current frame situation is deduced to the final game, including:

[0037] Obtain the current frame situation s and the corresponding action sequence a;

[0038] Initialize the number of deductions n and the number of stages n in the game deduction process s ;

[0039] For the i-th deduction, perform the following operations:

[0040] For the game deduction process described in stage k, the following operations are performed:

[0041] The Red side of the confrontation generates the current action situation Planning issues under P file;

[0042] The Red Team is based on the Randomly select strategy π and generate planning domain D file;

[0043] Call the local map planner to solve the P file under the D file and generate the The following sequence of actions

[0044] The blue side of the opposing team follows the and the blue side's action library {π e}, generate corresponding action sequence through conditional triggering

[0045] According to the With the Generate action situations through action mapping model deduction

[0046] Using the Update the Complete the k-stage deduction;

[0047] Generate the game tree and store the game tree in a database.

[0048] According to some embodiments of the present invention, the decentralized assessment includes evaluating resource efficiency V1, system value V2, and target damage V3, and V1 is calculated according to the following formula:

[0049]

[0050] In the formula, δ1, δ2, β j is a constant weight coefficient; α1 and α2 are the ammunition delivery efficiency and ammunition action efficiency respectively; cost(i) is the sum of resource consumption of action i in a; c a is the total number of actions in a; f e (α j ) is the utility function and is calculated according to the following formula:

[0051]

[0052] The α1 and α2 are calculated according to the following formula:

[0053]

[0054] Where n t is the number of tasks; c1 and c2 are constant coefficients; time(i) is the time consumption of task i; ammo(i) is the ammunition consumption of task i; D(i) is the target damage value in task i, which is a non-negative number; V2 is calculated according to the following formula:

[0055]

[0056] The first term of the formula is the situation control index, which is used to indicate the perception and control ability of the situation area; the second term of the formula is the tactical efficiency index, which is used to indicate the completion of the task; δ3 and δ4 are weight coefficients; m is the total number of situation areas, α K is a constant coefficient, p K (i) represents the capability status of situation area i, sets capability code K for situation control capability, and when situation area i has the capability represented by capability code K, p K (i) is 1, otherwise p K (i) is 0; n g is the target number of the game deduction process described in the kth stage; r kj is the achievement status of the jth goal in the game deduction process described in the kth stage, r kj When the j-th goal is achieved, it takes 1, otherwise it takes 0; the V3 is calculated according to the following formula:

[0057]

[0058] The right column of the formula is, from top to bottom, recognition accuracy, hit probability, and damage capability. The recognition accuracy is used to indicate the accuracy of identifying the target, the hit probability is used to indicate the accuracy of the weapon, and the damage capability is used to indicate the striking power of the ammunition; δ5, δ6, and δ7 are weight coefficients; γ i is a constant coefficient; r1 is a mission load parameter; r2 is a target parameter; r3 is a target environment parameter; η i is a constant coefficient; p1 is an ammunition accuracy parameter; p2 is an aiming accuracy parameter; ε1, ε2, ε3 are constant coefficients; ο1 is ammunition power; ο2 is active protection; ο3 is passive protection; the deduction evaluation value is calculated according to the following Bellman evaluation formula:

[0059]

[0060] In the formula, λ k is the discount factor, λ k ∈[0,1]; φ i V is the index weight determined by the adjacent superiority entropy weight method; i (s k ,a k ) According to the value of i, the action situation s at the current stage k and action sequence a k of the V1, the V2 and the V3.

[0061] According to the data and knowledge dual-driven situation assessment model of the second aspect of the present invention, it includes: an input module, the input module is used to input situation data; an instant evaluation model for evaluating the immediate benefits of the action situation, the instant evaluation model includes a situation encoding module and a residual text convolutional network; a deduction evaluation model for evaluating the future benefits of the action situation, the deduction evaluation model includes a game deduction module and a Bellman evaluation model, the game deduction module is used to execute a multi-stage game deduction process, and the Bellman evaluation model is used to execute the Bellman evaluation process, wherein the situation encoding module is used to formally encode the current frame situation and the previous situation that is temporally ahead of the current frame situation, and correspondingly generates a current single-frame encoded situation and a set including multiple previous single-frame encoded situations; and is used to calculate the information value of the previous single-frame encoded situation in the set, and perform thresholding. The invention relates to a method for comparing and sampling values ​​to generate a multi-frame sequence coding situation together with the current single-frame coding situation; the residual text convolution network is used to perform the residual text convolution on the multi-frame sequence coding situation and output an immediate benefit value; the game deduction module is used to deduce the current frame situation to the final game by random game to generate a game tree; the Bellman evaluation model is used to perform decentralized evaluation, evaluate the factors to calculate the multi-factor evaluation value of each stage of the game tree, and use the adjacent superiority entropy weight method to weight the multi-factor evaluation value to generate an immediate evaluation value of each stage of the game tree; it is used to perform the Bellman evaluation process, and calculate the deduced evaluation value of the current frame situation according to the decentralized evaluation and the Bellman evaluation process; it also includes: an output module, which uses the adjacent superiority entropy weight method to weight the comprehensive evaluation value of the current frame situation according to the immediate evaluation value and the deduced evaluation value.

[0062] According to the data-knowledge dual-driven situation assessment model of the embodiment of the present invention, an action data assessment model for calculating immediate benefits and a simulation deduction assessment method for calculating future benefits are established. The advantages and disadvantages of the battlefield situation are comprehensively evaluated through immediate benefits and future benefits, which provides a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and helps to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0063] According to the computer-readable storage medium of the third aspect embodiment of the present invention, a computer program is stored thereon, and when the computer program is executed by a processor, the data-knowledge dual-driven situation assessment method as described in the first aspect embodiment of the present invention is implemented.

[0064] According to the computer-readable storage medium of the embodiment of the present invention, it is possible to establish an action data evaluation model for calculating immediate benefits and a simulation deduction evaluation method for calculating future benefits, and comprehensively evaluate the advantages and disadvantages of the battlefield situation through immediate benefits and future benefits, thereby providing a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and helping to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0065] According to an embodiment of the fourth aspect of the present invention, an electronic device includes a processor and a memory, wherein the processor is connected to the memory, and the memory is used to store a computer program. When the computer program is executed by the processor, the data and knowledge dual-driven situation assessment method as described in the embodiment of the first aspect of the present invention is implemented.

[0066] The electronic device according to the embodiment of the present invention can establish an action data evaluation model for calculating immediate benefits and a simulation deduction evaluation method for calculating future benefits. It can comprehensively evaluate the advantages and disadvantages of the battlefield situation through immediate benefits and future benefits, provide a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and help to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0067] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 is a structural diagram of a situation assessment method driven by both data and knowledge according to an embodiment of the present invention;

[0069] Figure 2 is a schematic diagram of formal coding according to an embodiment of the present invention;

[0070] Figure 3 is a specific structural diagram of the formal coding according to an embodiment of the present invention;

[0071] Figure 4 is a schematic diagram of a residual text convolutional network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0073] The following describes the data and knowledge dual-driven situation assessment method according to an embodiment of the first aspect of the present invention with reference to the accompanying drawings.

[0074] like Figure 1-Figure 4 As shown, the data-knowledge dual-driven situation assessment method according to an embodiment of the present invention includes an immediate assessment of the immediate benefits of the action situation and a deduction assessment of the future benefits of the action situation. The immediate assessment includes performing situation encoding and performing residual text convolution. The deduction assessment includes performing a multi-stage game deduction process and performing a Bellman evaluation process. The method includes the following steps:

[0075] The current frame situation and the multiple previous frame situations that are temporally preceding the current frame situation are formally coded, and the current single-frame coded situation and the set of multiple previous single-frame coded situations are correspondingly generated. For example, it can be the situation data of the war game competition obtained through training. Figure 2 The time T is the current frame state, and three key frame states T-1, T-n' and T-m' are sampled for encoding. Each key frame state includes an environment coding layer and an entity coding layer.

[0076] Calculate the information value of the previous single-frame coding situation in the set, and perform threshold comparison and sampling to generate a multi-frame sequence coding situation together with the current single-frame coding situation;

[0077] Perform residual text convolution on the multi-frame sequence encoding situation and output the immediate benefit value;

[0078] Deducing the current frame situation random game to the end game, generating a game tree;

[0079] Perform decentralized evaluation, use evaluation factors to calculate the multi-factor evaluation value of each stage of the game tree, and use the adjacent superiority entropy weight method to weight the multi-factor evaluation value to generate an instant evaluation value for each stage of the game tree;

[0080] Execute the Bellman evaluation process, and calculate the derivation evaluation value of the current frame situation according to the decentralized evaluation and Bellman evaluation process;

[0081] According to the immediate evaluation value and the deduced evaluation value, the comprehensive evaluation value of the current frame situation is weighted and calculated using the adjacent superiority entropy weight method.

[0082] Situation assessment is a quantitative evaluation of the implementation effect of the planning scheme, which is the basis for scheme decision-making. As a component factor of the selection formula, the situation assessment value can guide the pruning search direction of the decision tree, improve the efficiency of planning decision-making, and effectively improve the accuracy of decision-making.

[0083] Therefore, according to the data-knowledge dual-driven situation assessment method of the embodiment of the present invention, an action data assessment model for calculating immediate benefits and a simulation deduction assessment method for calculating future benefits are established. The advantages and disadvantages of the battlefield situation are comprehensively evaluated through immediate benefits and future benefits, which provides a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and helps to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0084] According to some embodiments of the present invention, Figure 3 As shown, the formal coding includes coding entity attributes, environmental attributes, entity position distribution, and target distribution. The coding of entity attributes includes coding entity camps, equipment types, force status, and current action categories. The coding of environmental attributes includes elevation coding of combat area grids. The entity position distribution Pos is coded according to the following formula:

[0085]

[0086] In the formula, [x c ,y c ]|c is the coordinate of the distribution center c of this type of entity, is the distance between the entity and the center of the distribution, yes The distance along the x-axis; yes The distance along the y-axis; the target distribution GoPos is encoded according to the following formula:

[0087] GoPos={Pos,Fac} (2)

[0088] Where Pos is the target position; Fac is the target camp.

[0089] For example, entity number I, faction code F, type T Z , state S, action set A and distribution center c, distance DI from c, x-axis distance from c The y-axis distance from c Encoded as a 1×14 entity code, the target number E, distribution center C, distance DE from C, x-axis distance from c The y-axis distance from c The camp Fac is encoded, and a 4-bit module separator is added to form a target code of size 1×12. After encoding N entities and numbering M targets, an entity coding layer of size N×26 is formed. The environment is encoded and an environment coding layer of size N×26 is formed. If the number of encoded objects is insufficient, it is padded with -1. The entity coding layer and the environment coding layer form a 2-channel code, which increases the coding dimension to N×26×2, thus forming a multi-frame sequence coding situation of N×26×2l.

[0090] It can be understood that the 4-bit module separator is used to separate the entity code and the target code, and the 4-bit module separator is placed between the entity code and the target code, so the 4-bit module separator can also be added to the end of the entity code. When the number of targets encoded is different, the encoding sequence can be padded with 0.

[0091] As a result, complex and massive situational data can be simplified and abstracted into computer-recognizable coded data, which helps to improve the intelligent agent's perception and understanding of key elements such as combat entity attributes, location distribution, and mission environment.

[0092] According to some embodiments of the present invention, threshold comparison and sampling include:

[0093] When the information value of the previous single-frame coding situation is greater than the set threshold value 0.12, the previous single-frame coding situation is sampled to generate a multi-frame sequence coding situation; the information value Infor of the jth previous single-frame coding situation is calculated according to the following formula:

[0094]

[0095] In the formula, C j is the sum of the coding numbers of the jth previous single frame coding state; C j-1 is the sum of the coding numbers of the (j-1)th previous single-frame coding situation; C is calculated according to the following formula j :

[0096]

[0097] Where n j is the length of the jth preceding single-frame coding state; s j [k] is the jth previous single frame coding state s j The k-th code number at time j.

[0098] Therefore, by measuring the importance of a certain frame coding state through the information function, the importance of the previous single frame coding state can be judged, laying the foundation for the construction of multi-frame sequence coding state.

[0099] According to some embodiments of the present invention, a multi-frame sequence coding situation constructs an ordered two-dimensional situation coding matrix with a length of Len and a width of 3 with a sampling step of 3, and the ordered two-dimensional situation coding matrix is ​​arranged in chronological order, wherein Len represents the length of the single-frame coding situation. When the number of single-frame coding situations in the multi-frame sequence coding situation is less than 3, a single-frame coding situation with a length of Len and a full value of 0 is used to fill the vacant position of the ordered two-dimensional situation coding matrix to unify the output form of the multi-frame sequence coding situation.

[0100] By fusing the current single-frame coding state with multiple previous single-frame coding states as time-series input, the data received by the residual text convolutional neural network contains current and historical information, which can improve the feature information expression of the coding state and help improve the information level of the coding method.

[0101] According to some embodiments of the present invention, Figure 4 As shown, performing residual text convolution on multi-frame sequence encoding posture includes:

[0102] The coding dimension of the multi-frame sequence coding situation of size N×26×2l is adjusted to H×H×2l, H=32;

[0103] The residual channel is used to extract long-range features of the multi-frame sequence encoding situation. The residual channel includes the first convolution layer with a convolution kernel of 7×7 and the first batch of normalization layers of (H-6)×(H-6)×2l;

[0104] The text convolution channel is used to extract the semantic features of the multi-frame sequence encoding situation. The text convolution channel includes two convolution modules in series and a first random dropout layer of (H-6)×(H-6)×2l. The convolution module includes a second convolution layer with a convolution kernel of 3×3, a second batch normalization layer, a first linear rectification function layer, and a maximum pooling layer of size 2×2.

[0105] Fuse the features extracted by the residual channel and the text convolution channel;

[0106] A three-layer multilayer perceptron is used for secondary feature extraction, where the multilayer perceptron includes a first linear layer, a second linear rectifier function layer, a second random inactivation layer, and a second linear layer;

[0107] Use a normalized exponential function layer mapping to output the immediate return value.

[0108] The residual text convolutional neural network achieves full extraction of coding state features by constructing two parallel residual channels and text convolution channels, which helps to solve the degradation problems of feature information loss and gradient vanishing in deep neural networks, and also helps to improve the speed and accuracy of instant evaluation of coding state.

[0109] According to some embodiments of the present invention, the random game of the current frame situation is deduced to the final game, including:

[0110] Get the current frame situation s and the corresponding action sequence a;

[0111] Initialization number of deductions n and the number of stages n in the game deduction process s ;

[0112] For the i-th deduction, perform the following operations:

[0113] For the k-th stage game deduction process, the following operations are performed:

[0114] The Red side of the confrontation generates the current action situation Planning issues under P file;

[0115] Red side based on Randomly select strategy π and generate planning domain D file;

[0116] Call the local map planner to solve the P file under the D file and generate the strategy The following sequence of actions

[0117] The blue side in the confrontation is based on and the blue side's action library {π e}, generate corresponding action sequence through conditional triggering

[0118] according to and Generate action situations through action mapping model deduction

[0119] use renew Complete the k-stage deduction;

[0120] Generate a game tree and store it in the database. Thus, a game tree from the current frame situation to the final situation is constructed, realizing the random sequential deduction of the two opposing sides from the perspective of the own side, and providing support for the Bellman evaluation of the current single-frame coding situation.

[0121] According to some embodiments of the present invention, decentralized assessment includes evaluating resource efficiency V1, system value V2, and target damage V3. Resource efficiency evaluates the process consumption of tasks. Resource efficiency can reflect the rationality of the planning scheme. A planning scheme with high resource efficiency not only saves equipment resources, but also makes the task allocation of heterogeneous platforms more reasonable and reduces the risk of tasks. V1 is calculated according to the following formula:

[0122]

[0123] In the formula, δ1, δ2, β j is a constant weight coefficient; α1 and α2 are the ammunition delivery efficiency and ammunition action efficiency respectively; cost(i) is the sum of resource consumption of action i in a; c a is the total number of actions in a; f e (α j ) is the utility function and is calculated according to the following formula:

[0124]

[0125] Delivery efficiency is defined as the delivery efficiency per target per base number of ammunition, that is, the ratio of the average delivery time per target to the base number of ammunition delivered. The smaller the ratio, the shorter the action time of the ammunition, the more timely the attack effect, and the higher the delivery efficiency. Action efficiency is defined as the damage per target per base number of ammunition, that is, the ratio of the average damage per target to the base number of ammunition consumed. The higher the ratio, the higher the utilization rate of ammunition resources and the higher the action efficiency. Calculate α1 and α2 according to the following formula:

[0126]

[0127] Where n t is the number of tasks; c1 and c2 are constant coefficients; time(i) is the time consumption of task i; ammo(i) is the ammunition consumption of task i; D(i) is the target damage value in task i, which is a non-negative number.

[0128] The system value evaluates the top-level changes in the battlefield situation before and after the implementation of the task, focusing on the battlefield situation and tactical efficiency. According to some embodiments of the present invention, the system value includes situation control indicators and tactical efficiency indicators, and V2 is calculated according to the following formula:

[0129]

[0130] The first term of the formula is the situation control index, which is used to indicate the perception and control ability of the situation area; the second term of the formula is the tactical efficiency index, which is used to indicate the completion of the task; δ3 and δ4 are weight coefficients; m is the total number of situation areas, α K is a constant coefficient, p K (i) represents the capability status of situation area i, and sets capability code K for situation control capability. When situation area i has the capability represented by capability code K, p K (i) is 1, otherwise p K (i) is 0; n g is the target number of the game deduction process at the kth stage; r kj is the state of achieving the jth goal in the kth stage game deduction process, r kjWhen the jth goal is achieved, it takes 1, otherwise it takes 0. For example, when the state state(i) of situation area i in the situation data exists in its capability encoding K set {code}, p K (i) A value of 1 indicates that it has the capability. For example, when p = 1, {code} = {1-occupied state, 2-combat state}; when p = 0, {code} = {0-unknown state, 2-opponent occupied state}.

[0131] The target damage calculation is the implementation result of the strike mission. According to the process of identification, aiming and striking, the target damage calculation is divided into three parts: identification accuracy, hit probability and damage capability. V3 is calculated according to the following formula:

[0132]

[0133] The right column of the formula from top to bottom is recognition accuracy, hit probability, and damage capability. Recognition accuracy is used to indicate the accuracy of target identification, hit probability is used to indicate the accuracy of weapons, and damage capability is used to indicate the impact power of ammunition; δ5, δ6, and δ7 are weight coefficients; γ i is a constant coefficient; r1 is a mission load parameter; r2 is a target parameter; r3 is a target environment parameter; η i is a constant coefficient; p1 is the ammunition accuracy parameter; p2 is the aiming accuracy parameter; ε1, ε2, ε3 are constant coefficients; ο1 is the ammunition power; ο2 is active protection; ο3 is passive protection; the deduced evaluation value is calculated according to the following Bellman evaluation formula:

[0134]

[0135] In the formula, λ k is the discount factor, λ k ∈[0,1]; φ i is the index weight determined by the adjacent superiority entropy weight method; V i (s k ,a k ) According to the value of i, the action situation s at the current stage k and action sequence a k V1, V2 and V3.

[0136] The decentralized evaluation constructs an evaluation system that includes indicators such as resource efficiency, system value, and target damage, and connects the multi-stage evaluation of the game tree in series through the Bellman equation, realizing the calculation of future benefits of the current frame situation, further improving the accuracy and rationality of the comprehensive situation evaluation.

[0137] The weight measures the importance of each component indicator. Reasonable indicator weights make the evaluation model both systematic and accurate. The weight analysis method is divided into subjective method and objective method. The subjective method relies on human experience to make subjective assignments. It represents the external explicit relationship of the data. This method lacks objective understanding of the internal indicators. The objective method uses sample data to explore the weight relationship between data. It represents the inherent relationship within the data, but is not suitable for data with less correlation.

[0138] Adjacent Possibility Method (APM) is a subjective weight assignment method based on limited binary comparison. The method follows the steps of adjacent indicator superiority assignment and global indicator superiority assignment. The adjacent indicator superiority assignment is completed by expert judgment and fuzzy language scaling. Expert judgment provides the basis for fuzzy language calibration. The fuzzy language calibration value table is shown in Table 1.

[0139] Table 1 Fuzzy language calibration values

[0140]

[0141] The superiority of the global indicator is calculated by recursively calculating the adjacent superiority.

[0142] According to some embodiments of the present invention, the comprehensive weight is calculated according to the following formula:

[0143]

[0144] In the formula, is the comprehensive weight of the jth indicator, such as nz is the index number; ω j is the subjective weight of the jth indicator; θ j is the objective weight of the jth indicator.

[0145] The subjective weight is calculated according to the following formula:

[0146]

[0147] In the formula, β j，l is the fuzzy superiority of the jth indicator to the lth indicator, and β is calculated according to the following formula j，l :

[0148] β j，l =β j，l-1 +2·(1-β j，l-1 )·(β l-1,l -0.5) (14)

[0149] In the formula, when j = l, β j，l =β l，j=0.5; In addition, there is β j，l =1-β l，j The relative superiority matrix β of all indicators can be obtained through this recursive formula:

[0150]

[0151] Each row of the relative superiority matrix β represents an indicator, and the columns are the relative superiority values ​​of the indicator and other indicators. The subjective weight of the indicator in the row can be obtained by summing and normalizing each row.

[0152] The entropy weight method determines the objective weight by comparing the information entropy of the indicator data. In information theory, information entropy is defined as a measure of the degree of disorder of the system. The smaller the entropy value, the greater the amount of information in the system and the greater the impact on the whole, that is, its weight should also be greater. The objective weight is calculated according to the following formula:

[0153]

[0154] In the formula, H j is the entropy value of the jth indicator. The calculation of entropy weight is based on data normalization. The samples need to be normalized and H is calculated according to the following formula: j :

[0155]

[0156] Where n h is the number of existing samples of the jth indicator; z′ ju is the normalized value of the u-th sample of the j-th indicator. The normalization methods of different types of indicators are different. When the indicator is a benefit indicator, z′ is calculated according to the following formula ju :

[0157]

[0158] In the formula, z j is the sample set of the jth indicator; z ju is the original value of the i-th sample of the j-th indicator.

[0159] The comprehensive weight of the integrated adjacent superiority and entropy weight takes into account both the external performance of the data and the internal connection of the data, improves the applicability and effectiveness of the weight assignment, and can be applied to complex weight calculations under multiple indicators.

[0160] In order to solve the problem of weight assignment of constituent indicators in the action situation assessment model, the weights of the primary indicators and secondary indicators are calculated respectively. Table 2 shows the initial weight data of the secondary indicators. Expert scoring is adopted for indicators without reference, such as system value, remaining force, etc., and then all indicators are normalized.

[0161] Table 2 Initial data of constituent indicators under the action situation assessment model

[0162]

[0163] First, the adjacent weights of the secondary indicators are calculated through expert judgment. For the secondary indicators under the system value indicators, experts judge that tactical efficiency is slightly more important than situation control, that is, the relative superiority β 21 =0.58, according to the definition, the weight set ω1 = (0.42, 0.58). Similarly, for other secondary indicators, ammunition resource efficiency is more important than operation resource efficiency, in this case ω2 = (0.67, 0.33). For the target damage capability indicator, damage power is obviously more important than hit probability, and hit probability is slightly more important than recognition accuracy, that is, β 32 =0.75, β 21 =0.58, we can deduce β 13 =0.13, then the relative superiority matrix of the secondary target damage index is:

[0164]

[0165] Thus, we can obtain the weight set ω3 = (0.233, 0.296, 0.471). According to the normalization formula, we can obtain the normalized matrix:

[0166]

[0167] At this time, the secondary indicator entropy value, entropy weight (objective weight), and comprehensive weight can be calculated as shown in Table 3.

[0168] Similarly, the same method can be used to calculate the weight of the first-level indicators. According to professional evaluation opinions, the target damage indicator is slightly more important than the resource efficiency indicator, the resource efficiency indicator is slightly more important than the remaining force indicator, and the remaining force indicator is equally important as the system value indicator. At this time, the comprehensive weight of the second-level indicator is used as the objective weight data source of the first-level indicator, and the objective weight, subjective weight and comprehensive weight of the first-level indicator can be calculated, as shown in Table 4.

[0169] Table 3 Comprehensive weights of secondary indicators

[0170]

[0171] Table 4 Comprehensive weights of first-level indicators

[0172]

[0173] The data-knowledge dual-driven situation assessment model according to the second aspect of the embodiment of the present invention includes: an input module, an instant assessment model, a deduction assessment model and an output module.

[0174] Specifically, the input module is used to input situation data. The immediate evaluation model is used to evaluate the immediate benefits of the action situation, and the immediate evaluation model includes a situation encoding module and a residual text convolutional network. The deduction evaluation model is used to evaluate the future benefits of the action situation, and the deduction evaluation model includes a game deduction module and a Bellman evaluation model. The game deduction module is used to execute multiple stages of game deduction processes, and the Bellman evaluation model is used to execute the Bellman evaluation process.

[0175] Among them, the situation encoding module is used to formally encode the current frame situation and the previous situation that is located in front of the current frame situation in time, and correspondingly generates the current single-frame coding situation and a set including multiple previous single-frame coding situations; and is used to calculate the information value of the previous single-frame coding situation in the set, and perform threshold comparison and sampling to jointly generate a multi-frame sequence coding situation with the current single-frame coding situation. The residual text convolution network is used to perform residual text convolution on the multi-frame sequence coding situation and output the immediate benefit value. The game deduction module is used to deduce the current frame situation to the end of the random game and generate a game tree. The Bellman evaluation model is used to perform decentralized evaluation, evaluate the elements to calculate the multi-factor evaluation value of each stage of the game tree, and use the adjacent superiority entropy weight method to weight the multi-factor evaluation value to generate the instant evaluation value of each stage of the game tree; it is used to perform the Bellman evaluation process, and calculate the deduced evaluation value of the current frame situation according to the decentralized evaluation and Bellman evaluation process. The output module uses the adjacent superiority entropy weight method to weight the comprehensive evaluation value of the current frame situation based on the immediate evaluation value and the deduced evaluation value.

[0176] According to the data-knowledge dual-driven situation assessment model of the embodiment of the present invention, an action data assessment model for calculating immediate benefits and a simulation deduction assessment method for calculating future benefits are established. The advantages and disadvantages of the battlefield situation are comprehensively evaluated through immediate benefits and future benefits, which provides a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and helps to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0177] According to the computer-readable storage medium of the third aspect embodiment of the present invention, a computer program is stored thereon, and when the computer program is executed by a processor, the data-knowledge dual-driven situation assessment method as described in the first aspect embodiment of the present invention is implemented.

[0178] According to the computer-readable storage medium of the embodiment of the present invention, it is possible to establish an action data evaluation model for calculating immediate benefits and a simulation deduction evaluation method for calculating future benefits, and comprehensively evaluate the advantages and disadvantages of the battlefield situation through immediate benefits and future benefits, thereby providing a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and helping to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0179] According to an embodiment of the fourth aspect of the present invention, the electronic device includes a processor and a memory, the processor and the memory are connected, the memory is used to store a computer program, and when the computer program is executed by the processor, the data and knowledge dual-driven situation assessment method as described in the embodiment of the first aspect of the present invention is implemented.

[0180] The electronic device according to the embodiment of the present invention can establish an action data evaluation model for calculating immediate benefits and a simulation deduction evaluation method for calculating future benefits. It can comprehensively evaluate the advantages and disadvantages of the battlefield situation through immediate benefits and future benefits, provide a decision-making basis for the situation judgment of the intelligent agent in the war game simulation, and help to improve the autonomy, reliability and flexibility of the intelligent agent's game decision-making in the team-level tactical confrontation environment.

[0181] Other structures and operations of the electronic device according to the embodiment of the present invention are known to those skilled in the art and will not be described in detail here.

[0182] In the description of the present invention, it is to be understood that the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0183] In the description of the present invention, "a first feature" or "a second feature" may include one or more of the features. A first feature "above" or "below" a second feature may include direct contact between the first and second features, or may include contact between the first and second features through another feature between them instead of direct contact. A first feature "above", "above" or "above" a second feature may include the first feature being directly above or obliquely above the second feature, or may simply mean that the first feature is higher in level than the second feature.

[0184] It should be noted that, in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "specific embodiments", "examples" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0185] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A data-knowledge dual-driven situation assessment method, characterized in that: The method includes performing an immediate evaluation on the immediate benefits of the action situation and performing a deduction evaluation on the future benefits of the action situation, wherein the immediate evaluation includes performing situation encoding and performing residual text convolution, and the deduction evaluation includes performing a multi-stage game deduction process and performing a Bellman evaluation process. The method includes the following steps: Formal encoding is performed on a current frame situation and a plurality of preceding frame situations that are temporally preceding the current frame situation, and a current single-frame encoding situation and a set including a plurality of the preceding single-frame encoding situations are correspondingly generated; Calculating the information value of the preceding single-frame coding situation in the set, and performing threshold comparison and sampling to jointly generate a multi-frame sequence coding situation with the current single-frame coding situation; Performing the residual text convolution on the multi-frame sequence encoding situation and outputting an immediate benefit value; Deducing the current frame situation random game to the end game to generate a game tree; Perform decentralized evaluation, use evaluation factors to calculate multi-factor evaluation values ​​at each stage of the game tree, and use adjacent superiority entropy weight method to weight the multi-factor evaluation values ​​to generate instant evaluation values ​​at each stage of the game tree; Executing the Bellman evaluation process, and calculating the derivation evaluation value of the current frame situation according to the decentralized evaluation and the Bellman evaluation process; According to the instant evaluation value and the deduced evaluation value, the comprehensive evaluation value of the current frame situation is weightedly calculated using the adjacent superiority entropy weight method.

2. The data-knowledge dual-driven situation assessment method according to claim 1 is characterized in that: The formal coding includes coding entity attributes, environmental attributes, entity position distribution, and target distribution. The coding of the entity attributes includes coding entity camps, equipment types, force status, and current action categories. The coding of the environmental attributes includes elevation coding of the combat area grid. The entity position distribution Pos is encoded according to the following formula: In the formula, [x c ,y c ]|c is the coordinate of the distribution center c of this type of entity, is the distance between the entity and the center of the distribution, yes distance along the x-axis; yes The distance along the y-axis; the target distribution GoPos is encoded according to the following formula: GoPos={Pos,Fac} Where Pos is the target position; Fac is the target camp.

3. The data-knowledge dual-driven situation assessment method according to claim 1 is characterized in that: The threshold comparison and sampling include: When the information value of the preceding single-frame coding situation is greater than the set threshold value 0.12, the preceding single-frame coding situation is sampled to generate the multi-frame sequence coding situation; the information value Infor of the j-th preceding single-frame coding situation is calculated according to the following formula: In the formula, C j is the sum of the coding numbers of the jth preceding single-frame coding states; C j-1 is the sum of the coding numbers of the (j-1)th preceding single-frame coding situation; the C is calculated according to the following formula j : Where n j is the length of the jth preceding single frame encoding state; s j [k] is the jth preceding single frame coding state s j The k-th code number at time j.

4. The data-knowledge dual-driven situation assessment method according to claim 1 is characterized in that: The multi-frame sequence coding situation constructs an ordered two-dimensional situation coding matrix with a length of Len and a width of 3 with a sampling step number of 3. The ordered two-dimensional situation coding matrix is ​​arranged in chronological order, wherein Len represents the length of a single-frame coding situation. When the number of the single-frame coding situations in the multi-frame sequence coding situation is less than 3, the vacant positions of the ordered two-dimensional situation coding matrix are filled with all-0 single-frame coding situations with a length of Len to unify the output form of the multi-frame sequence coding situation.

5. The data-knowledge dual-driven situation assessment method according to claim 1 is characterized in that: Performing the residual text convolution on the multi-frame sequence encoding state includes: Extracting long-range features of the multi-frame sequence encoding situation using a residual channel, wherein the residual channel includes a first convolutional layer and a first batch of normalization layers; Extracting semantic features of the multi-frame sequence encoding situation using a text convolution channel, wherein the text convolution channel includes a plurality of convolution modules connected in series and a first random inactivation layer, and the convolution module includes a second convolution layer, a second batch normalization layer, a first linear rectification function layer, and a maximum pooling layer; Fusing the features extracted by the residual channel and the text convolution channel; Perform secondary feature extraction using a multilayer perceptron, wherein the multilayer perceptron includes a first linear layer, a second linear rectifier function layer, a second random inactivation layer, and a second linear layer; The instantaneous return value is output using a normalized exponential function layer mapping.

6. The data-knowledge dual-driven situation assessment method according to claim 1 is characterized in that: The random game of the current frame situation is deduced to the final game, including: Obtain the current frame situation s and the corresponding action sequence a; Initialize the number of deductions n and the number of stages n in the game deduction process s ; For the i-th deduction, perform the following operations: For the game deduction process described in stage k, the following operations are performed: The Red side of the confrontation generates the current action situation Planning issues under P file; The Red Team is based on the Randomly select strategy π and generate planning domain D file; Call the local map planner to solve the P file under the D file and generate the The following sequence of actions The blue side of the opposing team follows the and the blue side's action library {π e }, generate corresponding action sequence through conditional triggering According to the With the Generate action situations through action mapping model deduction Using the Update the Complete the k-stage deduction; Generate the game tree and store the game tree in a database.

7. The data-knowledge dual-driven situation assessment method according to claim 1 is characterized in that: The decentralized assessment includes evaluating resource efficiency V1, system value V2 and target damage V3, and V1 is calculated according to the following formula: In the formula, δ1, δ2, β j is a constant weight coefficient; α1 and α2 are the ammunition delivery efficiency and ammunition action efficiency respectively; cost(i) is the sum of resource consumption of action i in a; c a is the total number of actions in a; f e (α j ) is the utility function and is calculated according to the following formula: The α1 and α2 are calculated according to the following formula: Where n t is the number of tasks; c1 and c2 are constant coefficients; time(i) is the time consumption of task i; ammo(i) is the ammunition consumption of task i; D(i) is the target damage value in task i, which is a non-negative number; V2 is calculated according to the following formula: The first term of the formula is the situation control index, which is used to indicate the perception and control ability of the situation area; the second term of the formula is the tactical efficiency index, which is used to indicate the completion of the task; δ3 and δ4 are weight coefficients; m is the total number of situation areas, α K is a constant coefficient, p K (i) represents the capability status of situation area i, sets capability code K for situation control capability, and when situation area i has the capability represented by capability code K, p K (i) is 1, otherwise p K (i) is 0; n g is the target number of the game deduction process described in the kth stage; r kj is the achievement status of the jth goal in the game deduction process described in the kth stage, r kj When the j-th goal is achieved, it takes 1, otherwise it takes 0; the V3 is calculated according to the following formula: The right column of the formula is, from top to bottom, recognition accuracy, hit probability, and damage capability. The recognition accuracy is used to indicate the accuracy of identifying the target, the hit probability is used to indicate the accuracy of the weapon, and the damage capability is used to indicate the striking power of the ammunition; δ5, δ6, and δ7 are weight coefficients; γ i is a constant coefficient; r1 is a mission load parameter; r2 is a target parameter; r3 is a target environment parameter; η i is a constant coefficient; p1 is an ammunition accuracy parameter; p2 is an aiming accuracy parameter; ε1, ε2, ε3 are constant coefficients; ο1 is ammunition power; ο2 is active protection; ο3 is passive protection; the deduction evaluation value is calculated according to the following Bellman evaluation formula: In the formula, λ k is the discount factor, λ k ∈[0,1]; φ i V is the index weight determined by the adjacent superiority entropy weight method; i (s k ,a k ) According to the value of i, the action situation s at the current stage k and action sequence a k of the V1, the V2 and the V3.

8. A data-knowledge dual-driven situation assessment model, characterized in that: include: An input module, wherein the input module is used to input situation data; An instant evaluation model for evaluating the instant benefits of an action situation, the instant evaluation model comprising a situation encoding module and a residual text convolutional network; A deduction and evaluation model for evaluating the future benefits of the action situation, the deduction and evaluation model includes a game deduction module and a Bellman evaluation model, the game deduction module is used to execute a multi-stage game deduction process, and the Bellman evaluation model is used to execute the Bellman evaluation process, wherein: The situation encoding module is used to formally encode the current frame situation and the previous situation that is located in front of the current frame situation in time, and correspondingly generate the current single-frame coding situation and a set including multiple previous single-frame coding situations; and is used to calculate the information value of the previous single-frame coding situation in the set, and perform threshold comparison and sampling to jointly generate a multi-frame sequence coding situation with the current single-frame coding situation; The residual text convolution network is used to perform the residual text convolution on the multi-frame sequence encoding situation and output an immediate benefit value; The game deduction module is used to deduce the current frame situation random game to the end game and generate a game tree; The Bellman evaluation model is used to perform decentralized evaluation, evaluate factors and calculate multi-factor evaluation values ​​at each stage of the game tree, and use the adjacent superiority entropy weight method to weight the multi-factor evaluation values ​​to generate an instant evaluation value at each stage of the game tree; and is used to perform the Bellman evaluation process, and calculate the deduced evaluation value of the current frame situation according to the decentralized evaluation and the Bellman evaluation process; Also includes: An output module, wherein the output module calculates the comprehensive evaluation value of the current frame situation by weighting using the adjacent superiority entropy weight method according to the instant evaluation value and the deduced evaluation value.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the data-knowledge dual-driven situation assessment method as described in any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory, the processor is connected to the memory, the memory is used to store a computer program, and when the computer program is executed by the processor, the data-knowledge dual-driven situation assessment method as described in any one of claims 1-7 is implemented.

Images