Deep learning-based intelligent game confrontation evaluation method and device, and medium

Abstract

The invention belongs to the technical field of tactical drilling, and discloses an intelligent game confrontation evaluation method and device based on deep learning and a medium, and the method comprises the steps: obtaining the input data and combat result data of intelligent game confrontation; constructing multi-dimensional evaluation features based on the input data and the combat result data; based on the combat result data and a preset expert database, generating a combat effectiveness label corresponding to the multi-dimensional evaluation feature; constructing a standard sample set based on the multi-dimensional evaluation features and the combat effectiveness labels; and training a pre-constructed evaluation model based on the standard sample set to obtain a trained evaluation model, the trained evaluation model being used for outputting a combat level evaluation result by taking the real-time data of the combat result of the intelligent game confrontation as an input. According to the method, the combat level can be evaluated, and multi-dimensional and high-complexity decision-making requirements in a dynamic battlefield environment are met.

Description

Technical Field

[0001] The present invention belongs to the technical field of tactical exercises, and specifically relates to an evaluation method, device and medium for intelligent game confrontation based on deep learning. Background Art

[0002] A mathematical model of tactical confrontation is constructed based on game theory, covering complete information games, incomplete information games and multi-agent games. Among them, the coordination and dynamic game of multiple agents can target the non-round game characteristics of the battlefield (such as continuous time evolution), decouple the tactical layer (troop deployment) and the campaign layer (resource scheduling), and improve the scalability of the system.

[0003] Through the collaborative and dynamic game of multiple intelligent agents, a technologically advanced and practical helicopter squadron tactical simulation training system has been established by constructing a virtual simulation environment and necessary action models. This system enables tactical simulation training for single helicopters, pairs (triple helicopters), helicopter companies, and mixed aerial formations. Furthermore, through the application of artificial intelligence and machine learning technologies, command and decision-making behaviors and combat processes are realistically reflected, providing a new and realistic intelligent opponent, and creating a platform-supported and knowledge-based empowerment system.

[0004] The performance evaluation of multi-agent collaboration and dynamic game confrontation has become a core link in improving combat effectiveness. However, traditional evaluation methods rely on expert experience and static indicators, which are difficult to adapt to the multi-dimensional and highly complex decision-making needs in a dynamic battlefield environment. Summary of the Invention

[0005] The purpose of the present invention is to provide an evaluation method, device and medium for intelligent game confrontation based on deep learning, so as to solve the problem that existing evaluation methods rely on expert experience and static indicators and are difficult to adapt to the multi-dimensional and highly complex decision-making needs in dynamic battlefield environments.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an evaluation method for intelligent game confrontation based on deep learning, the method comprising: Obtain input data and combat result data for intelligent game confrontation; Construct multi-dimensional evaluation features based on input data and combat results data; Generate combat effectiveness labels corresponding to multi-dimensional evaluation features based on combat result data and a preset expert database; Construct a standard sample set based on multi-dimensional evaluation features and combat effectiveness labels; The pre-built evaluation model is trained based on the standard sample set to obtain a trained evaluation model. The trained evaluation model is used to take the real-time data of the combat results of the intelligent game confrontation as input and output the combat level evaluation results.

[0007] Preferably, the method further comprises: Obtain real-time data on the combat results of intelligent game confrontation; Pre-process the real-time data of combat results that can be used for game confrontation to obtain the combat characteristics to be evaluated; The combat characteristics to be evaluated are input into a trained evaluation model, and the trained evaluation model outputs a combat level evaluation result.

[0008] Preferably, the evaluation model is constructed using a convolutional neural network and a long short-term memory network, and the evaluation model includes: two convolutional layers, one LSTM layer, two fully connected layers and one output layer.

[0009] Preferably, based on the input data and operational result data, a multi-dimensional evaluation feature is constructed, including: Preprocess the input data and combat result data to obtain input features and combat result features; Based on the preset compression algorithm, the input features are compressed and reduced in dimension to obtain compressed input features; Based on the compressed input features and combat result features, multi-dimensional evaluation features are constructed.

[0010] Preferably, the preset compression algorithm is a sparse autoencoding algorithm, which includes two sparse autoencoding layers and a classification layer, and each of the two sparse autoencoding layers includes a sparse autoencoder.

[0011] Preferably, based on a preset compression algorithm, the input features are compressed and dimensionally reduced to obtain compressed input features, including: Input the input features to the first sparse autoencoder layer to compress the input features and obtain the first-order feature representation; The first-order feature representation is input into the second sparse autoencoder layer to compress the first-order feature representation to obtain the second-order feature representation; Input the second-order feature representation into the classification layer and determine the loss value between the second-order feature representation and the input feature; When the loss value between the second-order feature representation and the input feature reaches a preset value, the second-order feature representation is used as the compressed input feature.

[0012] Preferably, the pre-built evaluation model is trained based on the standard sample set to obtain a trained evaluation model, including: Divide the standard sample set into N subsets; where N is a positive integer; The nth subset is used as the test set, and the remaining subsets are used as the training set. The evaluation model is trained N times. In each training, the training set is used to train the evaluation model, and the test set is used to calculate the accuracy of the evaluation model after this training. Where n = 1, 2, ..., N; Based on the accuracy of the evaluation model after N training times, the trained evaluation model is determined.

[0013] In a second aspect, the present invention provides an evaluation device for intelligent game confrontation based on deep learning, which is used to implement the above-mentioned evaluation method for intelligent game confrontation based on deep learning. The device includes: Data acquisition module, used to obtain input data and combat result data of intelligent game confrontation; Feature construction module, used to construct multi-dimensional evaluation features based on input data and combat result data; The label generation module is used to generate combat effectiveness labels corresponding to multi-dimensional evaluation features based on combat result data and a preset expert database; A sample construction module is used to construct a standard sample set based on multi-dimensional evaluation features and combat effectiveness labels; The model training module is used to train the pre-built evaluation model based on the standard sample set to obtain a trained evaluation model. The trained evaluation model is used to take the real-time data of the combat results of the intelligent game confrontation as input and output the combat level evaluation results.

[0014] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned deep learning-based intelligent game confrontation evaluation method when executing the computer program.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned evaluation method for intelligent game confrontation based on deep learning.

[0016] Beneficial effects: The present invention constructs multi-dimensional evaluation features through the input data and combat result data of intelligent game confrontation, and then combines the expert database to generate combat effectiveness labels corresponding to the multi-dimensional evaluation features to train the evaluation model. The trained evaluation model can be deployed to realize the evaluation of combat level and adapt to the multi-dimensional and highly complex decision-making needs in a dynamic battlefield environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings: Figure 1 This is a flowchart of an evaluation method for intelligent game confrontation based on deep learning provided by one embodiment of the present invention; Figure 2 This is a block diagram of an evaluation device for intelligent game confrontation based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0019] Example 1 Figure 1 This is a flow chart of an evaluation method for intelligent game confrontation based on deep learning provided by an embodiment of the present invention. Figure 1 As shown, this embodiment provides an evaluation method for intelligent game confrontation based on deep learning, the method comprising: Step S10: Obtain input data and combat result data of the intelligent game confrontation.

[0020] In this embodiment, the multi-agents in the intelligent game confrontation include but are not limited to: an environment agent, an action agent, and a game confrontation agent.

[0021] Among them, the environmental intelligent agent uses virtual simulation technology to build a realistic battlefield environment, including terrain, weather, and deployment of enemy and friendly forces.

[0022] Among them, the action agent is used to establish an action model to simulate the actions of helicopter teams on the battlefield, including tactical actions such as single-machine, double / triple-machine, helicopter company and air mixed formation.

[0023] Among them, game-themed intelligent agents are used to simulate the combat action styles and real-time command and control modes of powerful enemies through the drive of knowledge base.

[0024] Therefore, the input data of intelligent game confrontation includes but is not limited to command data such as combat entity platforms, aircraft / shipborne equipment, weapon effectiveness, combat missions, etc. These command data are various combat commands made by intelligent agents based on changes in combat situation, such as air reconnaissance commands, early warning detection commands, firepower strike commands, and electronic jamming commands.

[0025] The combat result data includes but is not limited to the result data generated by the intelligent agent during the operation process, such as: status data of each unit and equipment, and firepower usage data.

[0026] In this embodiment, for the result data generated by the intelligent agent during the operation process, a fixed step time period is set to obtain the instantaneous combat result data generated by the intelligent agent in each time period. The combat result data is sorted in time, and all result data sets with timestamps are selected according to the step time as the combat result data.

[0027] For example: when you choose to extract a command in the game confrontation, the input data in the battle to the end stage and the force results after the game confrontation: time start time (combat entity platform, aircraft / shipborne equipment, weapon combat effectiveness, combat mission), time end time (combat entity, aircraft / shipborne equipment, weapon combat effectiveness), etc.

[0028] Step S20: Construct a multi-dimensional evaluation feature based on the input data and the combat result data. In this embodiment, the multi-dimensional evaluation feature is constructed by selectively extracting the input data of the battle from the time when a command is issued to the end of the game confrontation and the force results after the game confrontation: the time start time (combat entity platform, aircraft / shipborne equipment, weapon combat effectiveness, combat mission), time end time (combat entity, aircraft / shipborne equipment, weapon combat effectiveness), and other data.

[0029] Specifically, based on input data and operational results data, multi-dimensional evaluation features are constructed, including: Step S201: pre-process the input data and combat result data to obtain input features and combat result features; the pre-processing in this example includes: data normalization and data consistency verification, etc.

[0030] Step S202: Based on a preset compression algorithm, compress and reduce the dimension of the input features to obtain compressed input features.

[0031] In this embodiment, since the dimension of the input data is large, the dimension of the input features is also large, and the input features need to be compressed and reduced in dimension to remove some redundant information.

[0032] Among them, the preset compression algorithm is a sparse autoencoding algorithm, which includes two sparse autoencoding layers and a classification layer. Both sparse autoencoding layers contain a sparse autoencoder. The sparse autoencoder is an unsupervised learning algorithm used to compress high-dimensional data into a low-dimensional space while retaining the main features of the data.

[0033] Therefore, based on the preset compression algorithm, the input features are compressed and reduced in dimension to obtain the compressed input features, including: Step a10: Input the input features to the first sparse autoencoder layer, compress the input features, and obtain a first-order feature representation.

[0034] Step a20: Input the first-order feature representation to the second sparse autoencoder layer, compress the first-order feature representation, and obtain the second-order feature representation.

[0035] Step a30: Input the second-order feature representation into the classification layer and determine the loss value between the second-order feature representation and the input feature. The classification layer is a softmax classifier. Input the second-order feature representation into the softmax classifier. The softmax classifier outputs a classification vector. According to the classification vector and the input feature, a loss function is established. The loss function can be used to calculate the loss value between the second-order feature representation and the input feature.

[0036] Step a40: When the loss value between the second-order feature representation and the input feature reaches a preset value, the second-order feature representation is used as the compressed input feature.

[0037] Step S203: Construct multi-dimensional evaluation features based on the compressed input features and combat result features.

[0038] Therefore, this embodiment reduces the data dimension and improves the evaluation efficiency through steps S201 to S203.

[0039] Step S30: Based on the combat result data and the preset expert library, generate combat effectiveness labels corresponding to the multi-dimensional evaluation features.

[0040] In this embodiment, the combat effectiveness labels are divided into multiple levels by experts according to the combat effectiveness (excellent to poor), for example, six levels: excellent, good, medium, qualified, unqualified, and poor. The combat effectiveness can be quantified and weighed according to the success criteria of the combat mission in the task list based on the specific combat action.

[0041] In this embodiment, the action data generated during the game confrontation process is huge, and it is unlikely to rely entirely on expert judgment to determine the combat effectiveness. A deep learning model is used to classify unlabeled game confrontation actions and obtain combat effectiveness labels with more dimensions.

[0042] Step S40: Construct a standard sample set based on multi-dimensional evaluation features and combat effectiveness labels.

[0043] Step S50: The pre-built evaluation model is trained based on the standard sample set to obtain a trained evaluation model. The trained evaluation model is used to take the real-time data of the combat results of the intelligent game confrontation as input and output the combat level evaluation results.

[0044] In this embodiment, the evaluation model is constructed using a convolutional neural network (CNN) and a long short-term memory network (LSTM). The evaluation model includes two convolutional layers, one LSTM layer, two fully connected layers, and one output layer. The two convolutional layers are used to extract features from the input data (standard sample set), and the LSTM can effectively process time series data, such as the action data of the agent at different time steps. The LSTM can learn the dependencies between the agents at different time steps, thereby more accurately evaluating their combat level.

[0045] As a further optimization of this embodiment, the pre-built evaluation model is trained based on the standard sample set to obtain a trained evaluation model, including: Step S501: Divide the standard sample set into N subsets; where N is a positive integer; Step S502: Use the nth subset as the test set and the remaining subsets as the training sets to train the evaluation model N times. In each training, the training set is used to train the evaluation model, and the test set is used to calculate the accuracy of the evaluation model after this training; where n = 1, 2, ..., N; Step S503: Determine a trained evaluation model based on the accuracy of the evaluation model after N trainings.

[0046] In this embodiment, during the training process, the standard sample set is divided into multiple training sets and test sets using the n-fold method, and the average accuracy of multiple test sets is taken as the final accuracy. Only the evaluation model with an accuracy of more than 90% is selected as the trained evaluation model.

[0047] As a further optimization of this embodiment, the method further includes: Step b10: Obtaining real-time data on the results of the intelligent game confrontation; Step b20: Preprocess the real-time data of combat results that can be used for game confrontation to obtain combat characteristics to be evaluated; Step b30: Input the combat characteristics to be evaluated into the trained evaluation model, and the trained evaluation model outputs the combat level evaluation result.

[0048] The present invention constructs multi-dimensional evaluation features through the input data and combat result data of intelligent game confrontation, and then combines the expert database to generate combat effectiveness labels corresponding to the multi-dimensional evaluation features to train the evaluation model. The trained evaluation model can be deployed to realize the evaluation of combat level and adapt to the multi-dimensional and highly complex decision-making needs in a dynamic battlefield environment.

[0049] Example 2 Figure 2This is a block diagram of an evaluation device for intelligent game confrontation based on deep learning provided by an embodiment of the present invention. Figure 2 As shown, this embodiment provides an evaluation device for intelligent game confrontation based on deep learning, which is used to implement the evaluation method for intelligent game confrontation based on deep learning in Example 1. The device includes: Data acquisition module, used to obtain input data and combat result data of intelligent game confrontation; Feature construction module, used to construct multi-dimensional evaluation features based on input data and combat result data; The label generation module is used to generate combat effectiveness labels corresponding to multi-dimensional evaluation features based on combat result data and a preset expert database; A sample construction module is used to construct a standard sample set based on multi-dimensional evaluation features and combat effectiveness labels; The model training module is used to train the pre-built evaluation model based on the standard sample set to obtain a trained evaluation model. The trained evaluation model is used to take the real-time data of the combat results of the intelligent game confrontation as input and output the combat level evaluation results.

[0050] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the evaluation method for intelligent game confrontation based on deep learning in the first embodiment is implemented.

[0051] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the evaluation method of intelligent game confrontation based on deep learning in embodiment one is implemented.

[0052] The present invention constructs multi-dimensional evaluation features through the input data and combat result data of intelligent game confrontation, and then combines the expert database to generate combat effectiveness labels corresponding to the multi-dimensional evaluation features to train the evaluation model. The trained evaluation model can be deployed to realize the evaluation of combat level and adapt to the multi-dimensional and highly complex decision-making needs in a dynamic battlefield environment.

[0053] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0055] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for evaluating intelligent game confrontation based on deep learning, characterized in that: The method comprises: Obtain input data and combat result data for intelligent game confrontation; Construct multi-dimensional evaluation features based on input data and combat results data; Generate combat effectiveness labels corresponding to multi-dimensional evaluation features based on combat result data and a preset expert database; Construct a standard sample set based on multi-dimensional evaluation features and combat effectiveness labels; The pre-built evaluation model is trained based on the standard sample set to obtain a trained evaluation model. The trained evaluation model is used to take the real-time data of the combat results of the intelligent game confrontation as input and output the combat level evaluation results.

2. The evaluation method for intelligent game confrontation based on deep learning according to claim 1 is characterized in that: The method further comprises: Obtain real-time data on the combat results of intelligent game confrontation; Pre-process the real-time data of combat results that can be used for game confrontation to obtain the combat characteristics to be evaluated; The combat characteristics to be evaluated are input into a trained evaluation model, and the trained evaluation model outputs a combat level evaluation result.

3. The evaluation method for intelligent game confrontation based on deep learning according to claim 1 is characterized in that: The evaluation model is constructed using a convolutional neural network and a long short-term memory network. The evaluation model includes: two convolutional layers, one LSTM layer, two fully connected layers and one output layer.

4. The evaluation method for intelligent game confrontation based on deep learning according to claim 1 is characterized in that: Based on input data and operational results data, multi-dimensional evaluation features are constructed, including: Preprocess the input data and combat result data to obtain input features and combat result features; Based on the preset compression algorithm, the input features are compressed and reduced in dimension to obtain compressed input features; Based on the compressed input features and combat result features, multi-dimensional evaluation features are constructed.

5. The evaluation method for intelligent game confrontation based on deep learning according to claim 4 is characterized in that: The preset compression algorithm is a sparse autoencoding algorithm, which includes two sparse autoencoding layers and a classification layer, and each of the two sparse autoencoding layers includes a sparse autoencoder.

6. The evaluation method for intelligent game confrontation based on deep learning according to claim 5 is characterized in that: Based on the preset compression algorithm, the input features are compressed and reduced in dimension to obtain the compressed input features, including: Input the input features to the first sparse autoencoder layer to compress the input features and obtain the first-order feature representation; The first-order feature representation is input into the second sparse autoencoder layer to compress the first-order feature representation to obtain the second-order feature representation; Input the second-order feature representation into the classification layer and determine the loss value between the second-order feature representation and the input feature; When the loss value between the second-order feature representation and the input feature reaches a preset value, the second-order feature representation is used as the compressed input feature.

7. The evaluation method for intelligent game confrontation based on deep learning according to any one of claims 1 to 6, characterized in that: The pre-built evaluation model is trained based on the standard sample set to obtain a trained evaluation model, including: Divide the standard sample set into N subsets; where N is a positive integer; The nth subset is used as the test set, and the remaining subsets are used as the training set. The evaluation model is trained N times. In each training, the training set is used to train the evaluation model, and the test set is used to calculate the accuracy of the evaluation model after this training. Where n = 1, 2, ..., N; Based on the accuracy of the evaluation model after N training times, the trained evaluation model is determined.

8. An evaluation device for intelligent game confrontation based on deep learning, used to implement the evaluation method for intelligent game confrontation based on deep learning according to any one of claims 1 to 7, characterized in that: The device comprises: Data acquisition module, used to obtain input data and combat result data of intelligent game confrontation; Feature construction module, used to construct multi-dimensional evaluation features based on input data and combat result data; The label generation module is used to generate combat effectiveness labels corresponding to multi-dimensional evaluation features based on combat result data and a preset expert database; A sample construction module is used to construct a standard sample set based on multi-dimensional evaluation features and combat effectiveness labels; The model training module is used to train the pre-built evaluation model based on the standard sample set to obtain a trained evaluation model. The trained evaluation model is used to take the real-time data of the combat results of the intelligent game confrontation as input and output the combat level evaluation results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the evaluation method for intelligent game confrontation based on deep learning according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the program implements the evaluation method for intelligent game confrontation based on deep learning as described in any one of claims 1 to 7.

Images