Intelligent model training method based on finger simulation calculation integrated architecture

Abstract

The invention discloses an intelligent model training method based on finger-simulation-calculation integrated architecture design, which comprises the following steps: integrating command control, a simulation system, an AI middle station and a data middle station through a message bus according to a standard protocol, establishing a training environment of an intelligent model integrated with finger-simulation-calculation based on confrontation deduction, and adopting an architecture mode of enlarging a region in a small region. A distributed and flexibly expanded hardware support environment is formed, an intelligent model operation environment and an intelligent algorithm training support environment are separated by adopting a mode of separately deploying an intelligent algorithm and a model, an original command and control system application process is not changed, and rapid training of the intelligent model is realized in a virtual battlefield environment created by a simulation system. Compared with the prior art, the method has the advantages that the time cost and the trial and error cost of applying the command and control system to a battlefield are reduced, the efficiency is improved, and great convenience is brought to application of the intelligent command and control system and intelligent equipment to which the intelligent command and control system belongs.

Description

Technical Field

[0001] The present invention relates to the field of intelligent model training methods for command and control systems and equipment systems, and specifically to an intelligent model training method based on a command-simulation-computing integrated architecture design. Background Art

[0002] With the rapid development of artificial intelligence technology, command and control systems and their equipment are deeply integrated with artificial intelligence. The situational awareness, situational cognition, task planning, collaborative decision-making, action control and effectiveness evaluation of command and control systems and their equipment are deeply integrated with elements of artificial intelligence algorithms. Command and control systems and their units can be considered as a multi-agent system. Before the integration of artificial intelligence elements, command and control systems and their equipment use traditional algorithms to process various functions. The processing of traditional algorithms is basically deterministic input and deterministic output. During development, testing and application, their functions can basically be exhaustively verified. However, after the integration of artificial intelligence elements, based on the characteristics of artificial intelligence algorithms, more types of data input are needed to provide a basic training environment for intelligent model learning and optimization, so that the intelligent model can continuously adjust internal parameters to achieve higher efficiency. The use of digital simulation methods to construct an intelligent model confrontation deduction training environment is an inevitable requirement and solution for the integrated development of artificial intelligence, command and control systems and their equipment. According to the currently available information on building simulation training environments, most of the constructed intelligent model training environments have relatively single functions. They are basically targeted verification environments for single intelligent models. The intelligent model and the test environment are tightly coupled, and the training environment is not combined with the application environment. It is impossible to form a rapid application of the "training and application" of the intelligent model, which greatly limits the effectiveness of artificial intelligence in command and control systems and their equipment. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the above technical defects and provide a method for training intelligent models for command and control systems and equipment systems. It adopts an integrated design architecture of command and control, simulation and deduction, AI computing and data management, and integrates command and control processes, equipment models, simulation scenarios and AI computing frameworks to train and verify the application of intelligent models in an adversarial manner.

[0004] In order to solve the above problems, the technical solution of the present invention is: an intelligent model training method based on finger-simulation-computation integrated architecture design, comprising the following steps:

[0005] Step 1: System architecture design. The system can be divided into command and control systems, simulation systems, AI middle platforms and data middle platforms according to logic. It consists of hardware and software. The hardware provides computing services and human-computer interaction functions. The software includes basic platform software and application software. The hardware includes command and control servers, simulation servers, AI training and push servers, data middle platform servers, high-performance terminal computers, large-screen displays, switches and related ancillary equipment. The basic platform software includes operating systems, cloud deployment and management environments, AI computing frameworks and databases. The application software includes command and control systems, simulation and deduction systems and data governance systems.

[0006] The system adopts a layered architecture, including a platform support layer, a functional service layer, and a system application layer. The basic support layer includes a theoretical and standard part, a resource part, and a software environment part. The theoretical and standard part is a theory and standard for simulation, command and control, and AI. The resource part includes hardware resources, model data, environmental data, hypothetical cases, regulations, organization and military deployment information, as well as intelligent perception, intelligent cognition, and intelligent decision-making. Through digital modeling, the mapping of physical space and digital space is realized. The software environment part includes operating system, cloud platform management, AI computing framework, and database management.

[0007] Step 2: Design the system hardware deployment architecture. The basic hardware environment includes the command and control server, command and control terminal, simulation server, simulation terminal, AI training and pushing server, data server, switch, display and other equipment. The system is deployed in a block-based centralized manner. The command and control hardware, simulation hardware, AI computing hardware and data middle platform hardware are logically and independently distributed in small areas. Each independent area is centrally connected through a firewall to build a flexible and variable large-area distributed combination basic hardware environment. The system built with this architecture can expand resources at any time as needed. When the AI ​​computing power needs to be increased, add AI training and pushing servers. When the command and control services increase, add command servers. When the simulation services increase, add simulation servers. Each corresponding terminal is also adaptively expanded as needed;

[0008] Step 3: System workflow design, design intelligent algorithms for intelligent models. After the intelligent algorithms are designed, they are dynamically loaded and deployed to the AI ​​platform to develop simulation model components, assemble components, and form equipment simulation models. The simulation system sets up scenarios according to the training requirements of the intelligent model, and the command and control system sets up equipment grouping and tasks. Then, the task planning content is sent to the simulation system to control the simulation system model to run according to the plan. At the same time, the simulation scenario situation information is sent to the AI ​​platform to provide action input for the intelligent algorithm. After intelligent calculation, the action output is transmitted to the equipment model of the simulation system to control the movement of the model. The entire system works continuously to complete the algorithm training of the intelligent model. In the confrontation simulation, the data platform collects the data generated by the simulation in real time for storage, and analyzes and processes the data when necessary.

[0009] Step 4: Design of system information interaction relationship. During joint simulation training of intelligent algorithms, the AI ​​algorithm to be trained is loaded into the AI ​​middle platform, and interconnected through the middleware and simulation platform equipment model or command and control. With the support of various virtual construction environments of the simulation platform, the action information and behavior feedback information are interacted with the simulation platform to realize the training simulation of intelligent equipment and intelligent algorithms in various application situations. During the command and control application display and operation training, the command and control system is connected to the simulation system as a unit, and the equipment model is loaded in the simulation system according to the system requirements to build a virtual battlefield situation. The first-party (red party) equipment of the simulation system works under the command of the command and control system, and conducts combat confrontation with the virtual second-party (blue party) forces of the simulation platform, realizing the full function and full process work display of the command and control system in a laboratory environment.

[0010] Further: Based on the system information interaction relationship design in step 4, the following steps are included:

[0011] 1) Information transmission between the simulation platform and the AI ​​middle station. When conducting intelligent algorithm training, the simulation platform and the AI ​​middle station are connected through the data middleware after adaptation. The information sent by the simulation platform to the AI ​​middle station is the equipment operation status and environment interaction status, and the information sent by the AI ​​middle station to the simulation platform is the equipment movement and behavior control information;

[0012] 2) Information transmission between the simulation platform and the command and control platform. When evaluating the command and control plan, the command and control platform sends the combat plan to the simulation platform. The combat plan is converted into a combat scenario on the simulation platform. The simulation platform performs simulation deduction according to the combat scenario to complete the combat plan evaluation. At the same time, the command and control system can send control instructions to the equipment model during simulation to adjust the working status of the equipment model in real time.

[0013] 3) Information transmission between the simulation platform and the data center. The data transmitted between the simulation platform and the data center include equipment model data, geographical environment data, organization and deployment data, and intelligence data;

[0014] 4) Information transmission between the command and control system and the data center. The data transmitted by the command and control system and the data center include equipment data parameters, organization and deployment data, and intelligence data;

[0015] 5) Information transmission between the command and control system and the AI ​​center station. The data transmitted between the command and control system and the AI ​​center station are equipment action control information and equipment operation status information;

[0016] 6) Information transmission between the data middle platform and the AI ​​middle platform. The information transmitted between the data middle platform and the AI ​​middle platform is the algorithm.

[0017] The advantages of the present invention compared with the prior art are:

[0018] The present invention adopts an integrated architecture of command and control, simulation system, AI middle platform and data middle platform, and uses digital simulation means to construct a training environment for intelligent models based on confrontational simulation, which opens up the entire process of intelligent model training and application, forming a rapid application of "training and application" of intelligent models in one, reducing the time cost and trial-and-error cost of applying command and control systems to the battlefield, improving efficiency, and bringing great convenience to the application of intelligent command and control systems and their corresponding intelligent equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a system diagram of an intelligent model training method based on an integrated finger-simulation-computation architecture design of the present invention.

[0020] Figure 2 It is a system connection relationship diagram of an intelligent model training method based on the finger-simulation-computation integrated architecture design of the present invention.

[0021] Figure 3 It is a system architecture diagram of an intelligent model training method based on an integrated finger-simulation-computation architecture design according to the present invention.

[0022] Figure 4 It is a system flow chart of an intelligent model training method based on the finger-simulation-computation integrated architecture design of the present invention.

[0023] Figure 5 It is a system interaction relationship diagram of an intelligent model training method based on the finger-simulation-computation integrated architecture design of the present invention. DETAILED DESCRIPTION

[0024] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings, wherein the same components are represented by the same reference numerals.

[0025] It should be noted that the words "front", "rear", "left", "right", "up" and "down" used in the following description refer to directions in the drawings, and the words "inside" and "outside" refer to directions toward or away from the geometric center of a specific component, respectively.

[0026] In order to make the contents of the present invention more clearly understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0027] like Figures 1 to 5 As shown,

[0028] An intelligent model training method based on finger-simulation-computation integrated architecture design includes the following steps:

[0029] Step 1: System architecture design. The system can be divided into command and control systems, simulation systems, AI middle platforms and data middle platforms according to logic. It consists of hardware and software. The hardware provides computing services and human-computer interaction functions. The software includes basic platform software and application software. The hardware includes command and control servers, simulation servers, AI training and push servers, data middle platform servers, high-performance terminal computers, large-screen displays, switches and related ancillary equipment. The basic platform software includes operating systems, cloud deployment and management environments, AI computing frameworks and databases. The application software includes command and control systems, simulation and deduction systems and data governance systems.

[0030] The system adopts a layered architecture, including a platform support layer, a functional service layer, and a system application layer. The basic support layer includes a theoretical and standard part, a resource part, and a software environment part. The theoretical and standard part is a theory and standard for simulation, command and control, and AI. The resource part includes hardware resources, model data, environmental data, hypothetical cases, regulations, organization and military deployment information, as well as intelligent perception, intelligent cognition, and intelligent decision-making. Through digital modeling, the mapping of physical space and digital space is realized. The software environment part includes operating system, cloud platform management, AI computing framework, and database management.

[0031] Step 2: Design the system hardware deployment architecture. The basic hardware environment includes the command and control server, command and control terminal, simulation server, simulation terminal, AI training and pushing server, data server, switch, display and other equipment. The system is deployed in a block-based centralized manner. The command and control hardware, simulation hardware, AI computing hardware and data middle platform hardware are logically and independently distributed in small areas. Each independent area is centrally connected through a firewall to build a flexible and variable large-area distributed combination basic hardware environment. The system built with this architecture can expand resources at any time as needed. When the AI ​​computing power needs to be increased, add AI training and pushing servers. When the command and control services increase, add command servers. When the simulation services increase, add simulation servers. Each corresponding terminal is also adaptively expanded as needed;

[0032] Step 3: System workflow design, design intelligent algorithms for intelligent models. After the intelligent algorithms are designed, they are dynamically loaded and deployed to the AI ​​platform to develop simulation model components, assemble components, and form equipment simulation models. The simulation system sets up scenarios according to the training requirements of the intelligent model, and the command and control system sets up equipment grouping and tasks. Then, the task planning content is sent to the simulation system to control the simulation system model to run according to the plan. At the same time, the simulation scenario situation information is sent to the AI ​​platform to provide action input for the intelligent algorithm. After intelligent calculation, the action output is transmitted to the equipment model of the simulation system to control the movement of the model. The entire system works continuously to complete the algorithm training of the intelligent model. In the confrontation simulation, the data platform collects the data generated by the simulation in real time for storage, and analyzes and processes the data when necessary.

[0033] Step 4: Design of system information interaction relationship. During joint simulation training of intelligent algorithms, the AI ​​algorithm to be trained is loaded into the AI ​​middle platform, and interconnected through the middleware and simulation platform equipment model or command and control. With the support of various virtual construction environments of the simulation platform, the action information and behavior feedback information are interacted with the simulation platform to realize the training simulation of intelligent equipment and intelligent algorithms in various application situations. During the command and control application display and operation training, the command and control system is connected to the simulation system as a unit, and the equipment model is loaded in the simulation system according to the system requirements to build a virtual battlefield situation. The red equipment of the simulation system works under the command of the command and control system, and conducts combat confrontation with the virtual blue forces of the simulation platform, realizing the full function and full process work display of the command and control system in a laboratory environment.

[0034] The system information interaction relationship design based on step 4 includes the following steps:

[0035] 1) Information transmission between the simulation platform and the AI ​​middle station. When conducting intelligent algorithm training, the simulation platform and the AI ​​middle station are connected through the data middleware after adaptation. The information sent by the simulation platform to the AI ​​middle station is the equipment operation status and environment interaction status, and the information sent by the AI ​​middle station to the simulation platform is the equipment movement and behavior control information;

[0036] 2) Information transmission between the simulation platform and the command and control platform. When evaluating the command and control plan, the command and control platform sends the combat plan to the simulation platform. The combat plan is converted into a combat scenario on the simulation platform. The simulation platform performs simulation deduction according to the combat scenario to complete the combat plan evaluation. At the same time, the command and control system can send control instructions to the equipment model during simulation to adjust the working status of the equipment model in real time.

[0037] 3) Information transmission between the simulation platform and the data center. The data transmitted between the simulation platform and the data center include equipment model data, geographical environment data, organization and deployment data, and intelligence data;

[0038] 4) Information transmission between the command and control system and the data center. The data transmitted by the command and control system and the data center include equipment data parameters, organization and deployment data, and intelligence data;

[0039] 5) Information transmission between the command and control system and the AI ​​center station. The data transmitted between the command and control system and the AI ​​center station are equipment action control information and equipment operation status information;

[0040] 6) Information transmission between the data middle platform and the AI ​​middle platform. The information transmitted between the data middle platform and the AI ​​middle platform is the algorithm.

[0041] Among them, the basic supporting software includes operating system, cloud platform and intelligent computing framework, which work together to form the underlying operating environment of intelligent model adversarial simulation.

[0042] Operating system selection: The operating system uses the "Taihang" and "Kunlun" operating systems of the Ordnance Industry Group on the terminal and server sides respectively, supporting the localization construction from the bottom of the operating system. The Taihang operating system enhances the upper-layer communication capabilities by adding automatic link optimization and multi-mode redundant modules to ensure the quality of communication services, and shields other application details at the communication level, making it easier for users to achieve communication functions. The Kunlun operating system introduces an AI reasoning framework to support AI computing from the bottom layer, and optimizes container components to achieve real-time scheduling at the container kernel level. Through the selection of the "Taihang" and "Kunlun" localized operating systems, the training of intelligent models is supported under the localized operating system.

[0043] The cloud management platform is selected to flexibly adjust the requirements based on computing power and software deployment. The cloud management platform is used on the operating system to unify the management and control of computing resources and scheduling resources. Under the principle of decentralization, efficient data synchronization, distributed computing, and flexible and fast dynamic reorganization are achieved on the domestic operating system, providing basic support for the flexible and on-demand expansion of hardware and software resources for intelligent model training. The command and control system server, simulation system server, AI computing, and data middle-end server of this system are deployed on the cloud management platform and included in the cloud management platform for management and control. Through the cloud platform orchestration scheduling management engine, container engine, consistency component, and remote communication component, various pooled computing resources are flexibly used to simplify the construction of basic training conditions for intelligent models.

[0044] Selection of intelligent computing frameworks. Intelligent computing frameworks include MindSpore, TensorFlow, PyTorch, etc. The computing frameworks are deployed in containers and provide various intelligent computing service capabilities.

[0045] Improvement of the command and control system. The command and control system is improved and designed based on the existing command and control system. The standard modules include information access, intelligence processing, task planning, target allocation and status monitoring, etc. The interface is transformed and connected to the simulation system, AI middle platform and data middle platform via the network to interact with equipment status information and simulation environment information.

[0046] Simulation system improvement: The simulation system is improved and designed based on general simulation. The standard modules include component design, model assembly, simulation deduction, situation display and evaluation calculation. The interface is transformed to access the command and control system, AI middle platform and data middle platform via the network to interact with equipment status information and simulation environment information.

[0047] AI middle platform integration: In traditional intelligent model training, intelligent algorithms and computing frameworks are closely combined, and the development and computing processes are deeply bound, which cannot flexibly switch between various frameworks for different intelligent algorithms. This method uses existing kubernetes, isula, kubeedge, etc. to build an AI middle platform for the entire chain of intelligent algorithm development, and realizes the integration of multiple intelligent computing frameworks and intelligent algorithms. In terms of hardware equipment, self-controllable and excellent performance Ascend AI training cards and inference cards are selected for AI model training and AI model inference. In terms of operating system, the domestic "Kunlun" operating system is used to provide kernel-level security. In terms of heterogeneous software architecture, the domestic CANN architecture is used to enable AI chip computing power and develop and compile AI algorithms and operators. In terms of AI framework, frameworks such as MindSpore, TensorFlow, and PyTorch are selected. In terms of algorithms, single intelligent algorithms (including target recognition and obstacle avoidance) and group intelligent algorithms (task allocation, area search, path planning, and game confrontation) are integrated. Through configuration files, configurable interfaces are designed to access command and control systems and data middle platforms.

[0048] At the functional level, it provides solutions covering the entire process of artificial intelligence model training, manages computing, storage and other resources, and provides users with a convenient and friendly interactive interface, which reduces the difficulty of intelligent algorithm development and training, and reduces the cost of algorithm development and training. At the application level, it can access various existing public mature algorithms. On the one hand, it can expand the existing algorithm warehouse to cope with complex actual combat scenarios, and at the same time verify the feasibility and reliability of existing algorithms. On the other hand, it can expand the data set to provide sufficient data for model training.

[0049] The improvement of the data middle platform is based on the existing data middle platform. Through interface transformation, it connects to the command and control system, simulation system and AI middle platform via the network to interact with equipment status information and simulation environment information.

[0050] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent model training method based on finger-simulation-computation integrated architecture design, characterized by: The steps include: Step 1: System architecture design. The system can be divided into command and control systems, simulation systems, AI middle platforms and data middle platforms according to logic. It consists of hardware and software. The hardware provides computing services and human-computer interaction functions. The software includes basic platform software and application software. The hardware includes command and control servers, simulation servers, AI training and push servers, data middle platform servers, high-performance terminal computers, large-screen displays, switches and related ancillary equipment. The basic platform software includes operating systems, cloud deployment and management environments, AI computing frameworks and databases. The application software includes command and control systems, simulation and deduction systems and data governance systems. The system adopts a layered architecture, including a platform support layer, a functional service layer, and a system application layer. The basic support layer includes a theoretical and standard part, a resource part, and a software environment part. The theoretical and standard part is a theory and standard for simulation, command and control, and AI. The resource part includes hardware resources, model data, environmental data, hypothetical cases, regulations, organization and military deployment information, as well as intelligent perception, intelligent cognition, and intelligent decision-making. Through digital modeling, the mapping of physical space and digital space is realized. The software environment part includes operating system, cloud platform management, AI computing framework, and database management. Step 2: Design the system hardware deployment architecture. The basic hardware environment includes the command and control server, command and control terminal, simulation server, simulation terminal, AI training and pushing server, data server, switch, display and other equipment. The system is deployed in a block-based centralized manner. The command and control hardware, simulation hardware, AI computing hardware and data middle platform hardware are logically and independently distributed in small areas. Each independent area is centrally connected through a firewall to build a flexible and variable large-area distributed combination basic hardware environment. The system built with this architecture can expand resources at any time as needed. When the AI ​​computing power needs to be increased, add AI training and pushing servers. When the command and control services increase, add command servers. When the simulation services increase, add simulation servers. Each corresponding terminal is also adaptively expanded as needed; Step 3: System workflow design, design intelligent algorithms for intelligent models. After the intelligent algorithms are designed, they are dynamically loaded and deployed to the AI ​​platform to develop simulation model components, assemble components, and form equipment simulation models. The simulation system sets up scenarios according to the training requirements of the intelligent model, and the command and control system sets up equipment grouping and tasks. Then, the task planning content is sent to the simulation system to control the simulation system model to run according to the plan. At the same time, the simulation scenario situation information is sent to the AI ​​platform to provide action input for the intelligent algorithm. After intelligent calculation, the action output is transmitted to the equipment model of the simulation system to control the movement of the model. The entire system works continuously to complete the algorithm training of the intelligent model. In the confrontation simulation, the data platform collects the data generated by the simulation in real time for storage, and analyzes and processes the data when necessary. Step 4: Design of system information interaction relationship. During joint simulation training of intelligent algorithms, the AI ​​algorithm to be trained is loaded into the AI ​​middle platform, and interconnected through the middleware and simulation platform equipment model or command and control. With the support of various virtual construction environments of the simulation platform, the action information and behavior feedback information are interacted with the simulation platform to realize the training simulation of intelligent equipment and intelligent algorithms in various application situations. During the command and control application display and operation training, the command and control system is connected to the simulation system as a unit, and the equipment model is loaded in the simulation system according to the system requirements to build a virtual battlefield situation. The first-party equipment of the simulation system works under the command of the command and control system, and conducts combat confrontation with the virtual second-party forces of the simulation platform, realizing the full function and full process of the command and control system in the laboratory environment.

2. The intelligent model training method based on finger-simulation-computation integrated architecture design according to claim 1 is characterized by: The system information interaction relationship design based on step 4 includes the following steps: 1) Information transmission between the simulation platform and the AI ​​middle station. When conducting intelligent algorithm training, the simulation platform and the AI ​​middle station are connected through the data middleware after adaptation. The information sent by the simulation platform to the AI ​​middle station is the equipment operation status and environment interaction status, and the information sent by the AI ​​middle station to the simulation platform is the equipment movement and behavior control information; 2) Information transmission between the simulation platform and the command and control platform. When evaluating the command and control plan, the command and control platform sends the combat plan to the simulation platform. The combat plan is converted into a combat scenario on the simulation platform. The simulation platform performs simulation deduction according to the combat scenario to complete the combat plan evaluation. At the same time, the command and control system can send control instructions to the equipment model during simulation to adjust the working status of the equipment model in real time. 3) Information transmission between the simulation platform and the data center. The data transmitted between the simulation platform and the data center include equipment model data, geographical environment data, organization and deployment data, and intelligence data; 4) Information transmission between the command and control system and the data center. The data transmitted by the command and control system and the data center include equipment data parameters, organization and deployment data, and intelligence data; 5) Information transmission between the command and control system and the AI ​​center station. The data transmitted between the command and control system and the AI ​​center station are equipment action control information and equipment operation status information; 6) Information transmission between the data middle platform and the AI ​​middle platform. The information transmitted between the data middle platform and the AI ​​middle platform is the algorithm.

Images