中國軍隊第五代指揮資訊系統及其智慧化技術
現代英語：
Modern war presents the explosive growth of battlefield information and new combat style. With the continuous emergence of new technologies such as artificial intelligence and edge computing, a new generation of command information system is coming. Based on the international fourth generation command information system, this paper imagines the overall architecture of the fifth generation command information system, expounds the technical characteristics of its knowledge center, intelligent enabling, cloud edge integration, independent evolution and resilience adaptation, analyze its key technologies, continuously improves the battlefield information advantage, and transforms to the battlefield cognitive advantage, decision-making advantage and action advantage.
Abstract
Modern war presents the explosive growth of battlefield information and new combat style. With the continuous emergence of new technologies such as artificial intelligence and edge computing, a new generation of command information system is coming. Based on the international fourth generation command information system, this paper imagines the overall architecture of the fifth generation command information system, expounds the technical characteristics of its knowledge center, intelligent enabling, cloud edge integration, independent evolution and resilience adaptation, analyze its key technologies, continuously improves the battlefield information advantage, and transforms to the battlefield cognitive advantage, decision-making advantage and action advantage.
Key words
command information system ; artificial intelligence ; edge computing ; situation processing ; planning and decision ; action control
Cite this article
Download CitationsZHANG Zhi-hua , WANG Fan . The Fifth Generation Command Information System and Its Intelligent Technology. Command Control and Simulation . 2021, 43(5): 1-7 https://doi.org/10.3969/j.issn.1673-3819.2021.05.001
Previous Article Next Article In his report to the 19th CPC National Congress, President Xi Jinping clearly pointed out that “we should accelerate the development of military intelligence and improve the joint combat capability and all-domain combat capability based on network information systems”
[ 1 ] . This statement indicates that future wars will be based on networked and intelligent system operations. The fifth-generation command information system will focus on intelligence, strengthen battlefield information advantages, and strive for battlefield cognitive advantages, decision-making advantages, and action advantages. According to relevant reports, the international command information system has gone through four stages of development
[ 2 ] and is evolving towards the fifth-generation command information system. The system architecture is developing towards intelligence, knowledge, cloud edge, and service. The fourth-generation system in the world mainly uses networking, service, and cloud to build an overall coordinated command information system
[ 2 ] , which meets the needs of coordinated operations to a certain extent and achieves information advantages. However, with the explosive growth of battlefield information, it is difficult to transform the system information advantage into the commander’s cognitive and decision-making advantages. With the emergence of new combat styles such as unmanned combat and cyber warfare, in order to adapt to the complexity and nonlinear characteristics of combat command, the command information system must break through cognitive technology and provide accurate battlefield situation cognition and planning and decision-making capabilities. The fifth-generation command information system is envisioned to be centered on artificial intelligence, edge computing, and cloud brain technology to enhance battlefield cognitive advantages, decision-making advantages, and action advantages, support combat command to move from the information domain to the cognitive domain, and realize capabilities such as information knowledge, intelligent decision-making, agile command and control, multi-domain collaboration, and edge services.
1 New Concept of Command and Control
1.1 Intelligent command and control
Intelligent command and control is to use artificial intelligence methods to achieve the transformation from “information-based, network-centric” to “intelligent, knowledge-centric”, and assist commanders in solving perception, understanding, and cognitive problems in the command field. The system architecture and technical architecture of the command information system will change. The system will apply corresponding intelligent technologies around functional domains such as situation, command, control, and support to improve the cognitive and decision-making efficiency of combat command. Foreign militaries pay great attention to the intelligent application of combat command. Since 2007, the US DARPA has published three white papers on national and military development strategies for artificial intelligence, and has launched plans such as “Deep Green”
[ 3 ⇓ – 5 ] , “The High-Tech Holy Grail of the Third Offset Strategy”, and “Commander’s Virtual Staff”. In the field of intelligence perception and tactical decision-making, it has launched artificial intelligence projects such as “Insight”, “Xdata”, “Deep Learning”, “Deep Text Search and Filtering”, “Distributed Battlefield Management”, “Human-Machine Collaboration”, “Mind’s Eye”, “Trace”, “Human-Machine Collaboration”, “X-Plan”, “Cognitive Electronic Warfare”, and “AlphaAI Air Combat”, realizing the ability to deeply understand battlefield intelligence, predict situation cognition, and automatically generate and deduce tactical plans. Since then, the U.S. military has also set up projects such as “Autonomous Negotiation Formation”, “Big Dog”, and “Hummingbird” to improve the manned and unmanned collaborative control capabilities. Overall, the U.S. military currently has the world’s leading level of intelligent combat command. In addition, Germany, France, Russia and other countries have also conducted extensive research in intelligent information perception and processing, intelligent autonomous unmanned combat platforms, etc., and have achieved fruitful research results
1.2 Tactical Edge Command and Control
With the development of military technology, traditional large-scale cluster combat methods are gradually transformed into small-scale asymmetric combat. Combat activities at the tactical edge will play an important role in war. The tactical edge is also known as the “first tactical mile”
[ 9 ] . It is far away from the command center and has limited communication, computing, and service resources. It is usually composed of combat platforms, tactical units, and special forces. In order to gain information and decision-making advantages, command units at all levels use ubiquitous networks, micro-clouds, and other technologies to achieve information and resource sharing. Mobile computing devices at the tactical edge use fog computing methods to integrate into larger combat units and form micro-clouds under self-organizing networks. The large amount of situation information obtained by the tactical edge is calculated, stored, and shared in the tactical micro-cloud, which simplifies the scale of interaction with the command center, improves the timeliness of information interaction, and solves the problem of insufficient service capabilities at the tactical frontier in the past.
1.3 Multi-Domain Battle Command and Control
In 2016, the U.S. Army proposed the concept of “multi-domain warfare”
[ 10 ] , taking “synchronous cross-domain firepower” and “all-domain mobility” as core elements, promoting the high integration of combat elements, enhancing all-domain strike capabilities, and attempting to eliminate the “anti-access/area denial” capabilities of China, Russia and other countries. It mainly has the following three characteristics
[ 10 ] . First, the combat domain is expanded in multiple dimensions, enabling the U.S. Army to deploy forces from the ground to multiple combat domains such as sea, air, electricity, and the Internet, and has the ability to integrate with other services. Second, the combat elements are highly integrated, and the various services and combat functional domains can share information, coordinate tactics, and synchronize actions, which promotes the transformation of joint services to the integration of combat capability elements. Third, the command chain is developing in a flat direction, and the command mechanism is efficient and flexible. It is necessary to have centralized planning and decentralized execution, and to share information and instructions with various command nodes and individual soldiers, extend the tactical command chain, and realize rapid, multi-line, and multi-domain combat command.
1.4 Mosaic Combat Command and Control
In 2017, DARPA proposed the concept of “mosaic warfare”
[ 11-12 ] , which takes into account both ” threat-based” and “capability-based” equipment construction methods, and flexibly combines sensors, command and control nodes, combat platforms, and cooperative manned and unmanned systems in multiple combat domains on demand to form a mission system. System integration uses a building block approach to dynamically link dispersed fine-grained systems together to form a combat system similar to a “mosaic block”. “Mosaic warfare” uses intelligent decision-making tools to provide distributed situational awareness and adaptive planning and control, assist in combat mission planning, and implement distributed combat management. “Mosaic warfare” requires the replacement of fixed combat force composition with adaptive system reorganization, and the combat command has a resilient and adaptable information system that can customize physically dispersed mixed combat units on demand and meet various dynamic and collaborative combat requirements
2. Transformation of the Characteristics of the Fifth Generation Command Information System
1) The system shifts from network-centric to knowledge-centric. The network-centric approach brings battlefield information advantage, which is then transformed into cognitive advantage and decision-making advantage. The information sharing between systems shifts to knowledge-centric intelligence sharing, which promotes the transformation of the entire command system into decision-making and action advantage.2) The cloud architecture is transformed into cloud-edge-end integration. Expand the original cloud resource sharing capabilities
[ 2 ] and extend them to the platforms, teams, and individual soldiers at the tactical edge, realize the integrated hybrid service capabilities of the battlefield center cloud, mobile cloud, and edge micro-cloud in a mobile environment, and enhance the tactical frontier resource service capabilities.3) Transformation from scheduled integration to resilient adaptability. Currently, the system is deployed and operated according to preset rules. When the mission changes, it must be regulated according to the pre-planned plan. In the future, battlefield systems are vulnerable to attacks and paralysis, requiring the system to have the ability to self-reconstruct, resilient and adaptable when disturbances occur to ensure that the core mission is uninterrupted
[ 13-14 ] .4) Transformation from computational intelligence to cognitive intelligence. Intelligence is manifested in computational intelligence, perceptual intelligence, and cognitive intelligence. Currently, computational intelligence provides a tactical deterministic solution method. In the future battlefield, intelligent technology must be used to improve the accuracy and real-time degree of cognition in terms of massive intelligence processing, situational awareness, and decision-making reasoning.5) Performance changes from fixed fixed to autonomous learning evolution. The system’s algorithm and performance are generally determined and fixed during the design period, and performance improvement is achieved through upgrading and transformation. Intelligent systems have the ability of self-learning and self-evolution, and can learn algorithms for situational awareness and intelligent decision-making online to improve system performance.6) Construction shifts from capability-based to knowledge-based. Command information systems are generally constructed based on capability elements, and system integration is integrated based on capability elements. Intelligent systems pay more attention to the intellectual construction of the system, focusing on the construction of system knowledge, rules, algorithms, and data.7) The interaction mode will shift to human-machine fusion intelligent interaction. Human-machine fusion intelligent perception, anthropomorphic interaction, intention-oriented intelligent human-machine interface interaction, wearable human-machine fusion computing, and fusion and linkage interaction will become the main interaction mode of future systems, and the human-machine control system will progress towards human-machine fusion.8) The separation of combat and training has shifted to the integration of combat, training, exercise and research. The fifth-generation command information system tightly couples combat command and tactical training, and has parallel simulation and reasoning capabilities. It can not only update intelligent algorithms, but also conduct combat and tactics confrontation research, obtain tactical data, and promote algorithm learning. Exercise training has developed from war game simulation to battlefield virtual game.
3 Overall Architecture Concept
The overall architecture of the future fifth-generation command information system should be a command information system that is knowledge-centric, human-machine integrated, intelligently empowered, cloud-edge integrated, autonomously evolving, and resilient and adaptable. The following article mainly describes the overall system from the perspectives of system architecture, service architecture, and technical architecture
[ 15 ] . The system architecture mainly refers to the composition of the system’s logical elements and their relationships, the service architecture describes the integration model of information and computing resources between systems, and the technical architecture describes the system’s technical reference model.
3.1 System Architecture Concept
The system is changing from “information-based, network-centric” to “intelligent, knowledge-centric”, while extending to the tactical edge. Based on the original system integration, the system integrates knowledge and algorithms, applies intelligent technology in functional domains such as situation, command, control, and support, and improves the cognition and decision-making efficiency of combat command. The system architecture is envisioned as follows:
Figure 1 Conceptualization of the fifth-generation command information system architecture
第五代指揮資訊系統架構概念
The fifth-generation system expands the functional domain of parallel deduction and learning training on the basis of functional elements such as situational awareness, command decision-making, action control, support and guarantee, and information services to meet the needs of combat branch evaluation and algorithm learning. In terms of situational awareness, it covers computational intelligence, perceptual intelligence, and cognitive intelligence, mainly completing battlefield intelligence processing and target identification, understanding and predicting the situation, having state and momentum, and improving information advantage; in terms of command decision-making, it is mainly based on cognitive intelligence, which can machine tactical reasoning, generate plans and plans, and improve decision-making level; in terms of action control, it is mainly based on computational intelligence and cognitive intelligence, completing task monitoring and temporary tactical control, and providing action optimization strategies based on knowledge reasoning, such as command guidance, firepower coordination, and unmanned cluster intelligent control; in terms of comprehensive guarantee, it is mainly based on computational intelligence, completing the optimal allocation of battlefield resources under prior knowledge and rules; in terms of parallel deduction and learning training, it combines command and control with simulation training, trains personnel and algorithms in peacetime, and conducts parallel plan deduction in wartime.In addition, the fifth-generation system has an autonomous evolving learning mechanism: first, autonomous learning within the node to optimize the algorithm and knowledge base; second, the nodes share intelligent algorithms and knowledge through the command cloud to collaboratively complete the evolution. Each node can upload the learned algorithms and knowledge to the command cloud to update the algorithms and knowledge of the knowledge center; third, the system issues instructions to tactical nodes, weapon nodes, detection nodes, and combat support nodes, and collects execution feedback. These feedback results can be used to learn and evolve the algorithm.Between the fifth-generation systems, based on the original comprehensive integration based on the cloud/end architecture, an integrated sharing method for knowledge and intelligent algorithms has been added. Each command information system uploads intelligent algorithms and knowledge rules to the knowledge center for plug-and-play sharing by heterogeneous nodes such as battlefield detection, command, and weapons. The command information system can obtain existing intelligent knowledge from the knowledge center and conduct secondary learning and training in combination with its own battlefield data to improve algorithm capabilities. The command cloud will eventually form an intelligent knowledge center for the battlefield, and a battlefield knowledge network will be formed between the intelligent command information systems.
3.2 Concept of cloud-edge-device service architecture
In the future, ubiquitous network connections will extend from command units to various squads, individual soldiers, and platforms at the tactical edge. The fifth-generation command information system will use fog computing and distributed computing technologies to build tactical mobile clouds, squad micro-clouds (Cloudlet), and individual task group pico-clouds (Pico-Cloud) based on cloud architecture technology
[ 9 , 16 ] , forming tactical frontier mobile cloud service capabilities, realizing the hybrid service capabilities of battlefield centralized combat clouds, mobile tactical clouds, and edge micro-clouds and pico-clouds, forming an integrated resource service structure of “cloud, edge, and end”, and quickly building command chains and strike chains.
Concept of cloud-edge-end service architecture of the fifth-generation command information system
第五代指揮資訊系統雲端端服務架構構想
The cloud-edge-end integrated service capability supports the fifth-generation system to achieve dynamic aggregation and release of combat resources through “cloud deployment, cloud aggregation, cloud attack, and cloud dissipation”, thereby improving the combat effectiveness of the entire system
[ 17 ] . The centralized combat cloud is deployed in the command center in a fixed cloud manner
[ 16 ] to provide services for various combat nodes; air, land, and sea tactical clouds provide information, algorithms, computing, and storage services under mobile conditions for aircraft, ships, armored forces, and other forces at the tactical frontier, thereby improving the resource sharing level at the tactical frontier
[ 9 , 16 , 18-19 ] ; in tactical edge military operations, micro-clouds and pico – clouds are constructed. Micro-clouds are deployed in fog computing on vehicles, aircraft, and boats within one hop of the communication distance of the frontier contact unit, expanding the tactical information processing and sharing capabilities of the frontier unit personnel. When individual soldiers and units cannot access micro-clouds, mobile ad hoc networks and distributed computing technologies can be used to construct pico-clouds to support dynamic information aggregation and resource sharing end-to-end under weak connections at the tactical edge, thereby extending the command chain.
3.3 Technical Architecture Concept
The fifth-generation command information system will extend the war from the physical domain and information domain to the cognitive domain, and will change the way of command and control. Its technical architecture is as follows:
Technical architecture of the fifth-generation command information system
第五代指揮資訊系統技術架構
The fifth generation command information system adds tactical edge services and intelligent computing environments based on the networked computing environment of the fourth generation command information system, which is compatible with the system architecture and meets the intelligent requirements of the system. The tactical edge service computing environment provides micro-cloud and pico-cloud basic computing, storage, and information service platforms for weakly connected terminals; the intelligent computing environment provides intelligent services for situation, decision-making, control, and human-computer interaction.The intelligent technology environment layer includes the following five parts. The intelligent computing hardware platform is equipped with AI acceleration processors such as GPU, FPGA, and TPU to adapt to the computing power required by deep learning. Some algorithms use brain-like chips with neuron processing mechanisms or solidified dedicated intelligent computing chips; the intelligent data management platform mainly manages data, samples, cases, models, and knowledge; the deep learning framework integrates the runtime library and basic algorithm library of deep learning and reinforcement learning; the traditional artificial intelligence computing framework includes traditional algorithm support libraries such as spark and bigflow for search and solution, data mining, and parallel processing; intelligent services include application-oriented intelligent algorithm service libraries, such as intelligent interactive recognition, valuation network calculation, and strategy network calculation services, which provide solution interfaces for application development.The intelligent application layer mainly provides functional elements such as intelligent situational awareness, planning and decision-making, action control and information services, human-computer interaction, learning and training. It is the system’s main functional interface for users and the core problem that intelligence needs to solve.The fifth-generation system technology architecture model mentioned above mainly uses cloud computing and intelligent technology support services to achieve the sharing of situations, instructions, algorithms and knowledge between systems, and supports system autonomous evolution, algorithm upgrades and knowledge updates. System intelligence can be divided into levels 0 to 4
[ 20 ] . Level 0: full manual control; Level 1: computing intelligence, deterministic complex tactical calculations and information automation processing; Level 2: having certain perceptual intelligence, able to understand, evaluate and predict battlefield situations; Level 3: having cognitive intelligence, able to provide machine decision-making and decision-making deduction capabilities; Level 4: having human-machine integration and symbiosis capabilities, and the core algorithm can self-learn and self-evolve. At present, the intelligence level of the fourth-generation system is generally at level 1, and situation understanding and command decisions are still controlled by humans. The intelligence of the fifth-generation system can reach the fourth level through three stages. The first stage is to realize the ability to perceive, understand and evaluate the battlefield situation; the second stage is to build a knowledge base of tactics and enable machine decision-making based on rules, knowledge and algorithms; the third stage is to realize machine self-learning and self-evolution of core tasks, and have the function of autonomous decision-making, reaching a highly intelligent level of human-machine integration
[ 20 ] .
4 Key technologies of the system and its intelligent concept
The key technologies of the fifth-generation command information system mainly solve the above – mentioned problems of intelligence, cloud – edge-end integration, and system resilience and adaptability. The key technologies of the system and its intelligent concept is the following
Key technologies of the system and its intelligent concept
系統關鍵技術及智慧化理念
The key technologies of the fifth-generation command information system cover all aspects of the command and control OODA loop, and can support the system’s intelligence, resilience, and edge command and control requirements in terms of detection, decision-making, control, and strike, thereby building a precise perception chain, rapid control chain, precise strike chain, and agile service chain, extending to the tactical edge and improving command effectiveness.
1) Situational Awareness Machine Analysis TechnologyIntelligence compilation and analysis technology.
Use big data, deep learning, knowledge graphs and other technologies to perform intelligent information correlation matching, text semantics intelligent analysis, and public opinion intelligent search and extraction to obtain valuable intelligence from massive, multi-source, and heterogeneous battlefield information.
Multiple target rapid recognition technology. Using deep learning methods, a multi-layer CNN convolutional neural network is constructed, and sample feature parameter learning is used to complete feature extraction and rapid target recognition of optical, infrared, electromagnetic, and acoustic information.Situation recognition and understanding technology. Analyze the enemy’s combat intentions and combat capabilities, use the reinforcement learning valuation network technology to simulate the commander’s situation recognition process, and combine the CNN nonlinear battlefield situation fitting ability to establish a mapping from situation images to situation understanding
Situation machine prediction and assessment technology. Based on situation understanding, the enemy’s tactical behavior is estimated. First, the strategy network is used to obtain the enemy’s activity rules, and then the parallel deduction method is used to perform multi-branch situation deduction. Finally, a prediction network is constructed to predict the situation.
2) Operational planning machine decision-making technology.
Combat mission space and strategy modeling technology. Modeling the state and action strategy of the combat mission space and determining the description method of the mission state, strategy, and feedback are the basis for deep reinforcement learning to make decisions.Mission planning machine decision-making technology. Use operations optimization to complete target analysis and task allocation. Use deep reinforcement learning and swarm intelligence algorithms to machine plan force composition, firepower configuration, and collaborative paths. Tactical planning tends to be rule-based reasoning and easy to break through; campaign planning tends to be knowledge-based reasoning based on experience, involving the art of command, and is more difficult to break through.
Parallel simulation technology for combat plans. With reference to the parallel simulation technology of the “deep green” system the Monte Carlo search tree and game test method are used to simulate enemy combat behavior, rehearse and evaluate the action process, and accumulate feedback reward and punishment functions for learning, training, and decision optimization.
Intelligent generation technology of combat plans. Using intelligent perception algorithms such as natural language understanding, voice command recognition, and sketch recognition, combined with the extraction of elements from the task model, the knowledge graph is used to automatically extract the plan to generate combat plans and command sequences .
Rapid decision-making technology on the spot. Based on the current situation, using the learning data accumulated by the game platform, automatically matching the most appropriate plan adjustment, making dynamic decisions on the plan based on Monte Carlo tree search and transfer learning algorithms, reverse reinforcement learning, and enhancing the generalization ability of the plan.3) Intelligent motion control technologySituation-based improvisation control technology. According to the effects and deviations of combat operations, the resources, paths, and coordination modes of the mission are dynamically adjusted, and parallel simulation multi-branch deduction and reinforcement learning technology are used to correct the deviations, thus realizing tactical “feedforward” control .
Swarm intelligence collaborative control technology. Promote the maximization of the overall effectiveness of battlefield intelligent bodies in collaborative operations, use ant colony and bee colony control algorithms and deep reinforcement learning methods to build a global tactical value network, establish an effect feedback model, and perform strategic control based on the value network.Firepower collaborative control technology. Improve the speed and accuracy of friend-or-foe identification, firepower allocation, and collaborative dispatch, use swarm intelligence and deep reinforcement learning algorithms to automatically plan, coordinate and optimize the strike chain, and have a certain degree of autonomous decision-making ability.
4) Manned/unmanned collaborative command technology.
Multi-domain cluster system autonomous collaborative machine planning technology. Use branch search solution, knowledge reasoning, and deep reinforcement learning to plan and allocate collaborative tasks for manned/unmanned systems, and use swarm intelligence optimization algorithms to plan collaborative trajectories for unmanned and manned platforms.Multi-domain cluster system autonomous collaborative command and control technology. It monitors the missions of unmanned clusters and provides autonomous collaborative command and guidance. It uses swarm intelligence algorithms to detect conflicts and avoid collisions among multiple unmanned platforms, and coordinates grouping, routing, and load.
5) Intelligent information service technology.
Intelligent battlefield information sharing technology uses reinforcement learning and semantic association technology to analyze users’ information needs and preferences, generate information needs based on users’ differentiated characteristics, and intelligently push tactical information to users.
6) Human-machine fusion intelligent interaction technology.
Human-computer fusion intelligent perception interaction technology. Construct multi-channel human-computer interaction methods including sketches, spoken language, gestures, head postures, expressions, eye movements, etc., and provide natural, sensitive, accurate and anthropomorphic interaction strategies . Intention-oriented intelligent human-computer interface technology. Using FCM fuzzy cognitive interactive reasoning technology, infer the user’s interactive intention, and organize the interactive interface output by integrating different means such as spoken language, gestures, sketches, and natural language according to the user’s interface needs and interaction preferences.Smart wearable human-machine fusion technology. It uses edge computing technology and new human-machine interaction methods such as voice, gestures, eye movements, brain-computer interfaces, and augmented reality to provide soldiers with smart wearable devices that have a collaborative, integrated, and linked human-machine interaction mode.
7) Virtual gaming and training evaluation technology.
The combat virtual game technology builds a game confrontation test platform, conducts combat knowledge modeling, and uses parallel simulation, branch decision, differential confrontation and other technologies to conduct red-blue confrontation, which not only trains tactics and methods, but also collects tactical data.Machine training and evaluation technology uses the data accumulated by the game platform and the experience of personnel to model, adopts small sample transfer learning technology to train and optimize the algorithm, replays the real data afterwards, performs transfer learning optimization on the decision model, and updates the decision plan.
8) System resilience adaptive reconstruction technology.
Environmental perception and autonomous fault detection technology. Under soft and hard damage, it can detect the main faults and analyze abnormal correlations, predict the occurrence of faults that affect task execution, evaluate the impact of faults on tasks, and realize active perception and rapid location of system resources and faults.System self-healing and reconstruction intelligent technology. When key nodes of the system fail, an adaptive mechanism is used to reallocate resources, achieve capacity regeneration, and continuously ensure the completion of core tasks. The system changes from a fault repair method with preset rules and manual participation to an intelligent system reconstruction method.
9) Tactical edge computing technology.
Mobile micro-cloud service platform technology. Deployed in fog computing mode on vehicles, aircraft, and boats within one hop of the enemy, it provides shared processing capabilities for combat teams and expands the tactical information processing capabilities of team members.Pi-cloud resource sharing technology under weak connection ad hoc network. Based on the individual soldier ad hoc network, the Pi-cloud is constructed using distributed computing technology to support end-to-end autonomous collaborative information sharing and resource sharing between individual soldier mobile devices under weak connection to meet tactical edge needs.
5 Development ideas and ideas
1) Gradually progress in stages, starting with the easy and then moving on to the difficult. In the first stage, image, voice, gesture, face recognition, and natural language understanding are applied to intelligence analysis; in the second stage, deep learning and reinforcement learning are applied to situational awareness and command decision-making; in the third stage, cloud computing is used to realize a knowledge-centered, intelligently empowered system.
2) Select intelligent algorithms for application. Focusing on the application of deep learning in situation and deep reinforcement learning in planning and decision-making, select appropriate tactical backgrounds to verify intelligent algorithms. Tactical-level planning of paths, firepower, tasks, etc. can be used as breakthroughs.
3) Strengthen the construction of knowledge engineering in the field of combat command. Expert rules, military regulations, and actual combat data are the basis of intelligent command. The existing combat rules should be modeled and represented in a knowledge-based manner, and the input and output mapping relationship between knowledge representation and deep learning should be established. The research on knowledge learning and knowledge reasoning methods should be strengthened .
4) Establish a virtual confrontation game platform to accumulate data. Intelligent algorithms require a large number of learning samples. The ways to accumulate samples are: Establish a confrontation game platform to conduct war games, human-machine confrontation, and red-blue confrontation to accumulate data; Collect tactical data from actual combat exercises and build models as training samples.
6 Conclusion
This paper proposes the overall and intelligent concept of the fifth-generation command information system, constructs a new generation of command information system architecture with “intelligent empowerment, human-machine integration, cloud-edge integration, autonomous evolution, cloud-intelligence sharing, and resilience and adaptability”, analyzes its key technologies and capability characteristics, and attempts to achieve cognitive advantages, decision-making advantages, and action advantages based on the fourth-generation system in the world .
There are not many technical verifications for the fifth-generation system in the world, so we should not rush for quick success and still need to conduct sufficient research.
現代國語：
現代戰爭呈現戰場資訊爆炸性成長與新型作戰形態。隨著人工智慧、邊緣運算等新技術的不斷湧現，新一代指揮資訊系統呼之欲出。本文在國際第四代指揮資訊系統的基礎上，構想了第五代指揮資訊系統的整體架構，闡述了其知識中心化、智慧賦能、雲邊融合、自主演進和彈性適配的技術特徵，分析了其關鍵技術，不斷提升戰場資訊優勢，並向戰場認知優勢、決策優勢和行動轉化。
習主席在中國共產黨十九大報告中明確指出,「加速軍事智能化發展,提高基於網路資訊體系的聯合作戰能力、全域作戰能力」[1]。這個論述指明了未來戰爭將是基於網路化、智慧化的體係作戰,第五代指揮資訊系統將以智慧化為核心,強化戰場資訊優勢,爭取戰場認知優勢、決策優勢與行動優勢。據相關通報,國際上指揮資訊系統經歷了四個階段的發展過程[2],正在向第五代指揮資訊系統演化,系統體系結構向智慧化、知識化、雲端端、服務化發展。國際上第四代系統主要以網路化、服務化、雲端化等手段建構了整體協同的指揮資訊體系[2],一定程度上滿足協同作戰需求,實現了資訊優勢。但隨著戰場資訊的爆發式增長,系統資訊優勢很難轉化為指揮的認知與決策優勢,隨著無人作戰、賽博作戰等新型作戰樣式的出現,為了適應作戰指揮的複雜性、非線性特徵,指揮資訊系統須突破認知技術,提供準確的戰場態勢認知與籌劃決策能力。第五代指揮資訊系統設想以人工智慧、邊緣運算、雲腦技術為核心,提升戰場認知優勢、決策優勢與行動優勢,支援作戰指揮由資訊域邁向認知領域、實現資訊知識化、決策智慧化、指控敏捷化、協同多域化、服務邊緣化等能力。
1 指揮控制新理念
1.1 智能化指揮控制
智慧化指揮控制就是利用人工智慧方法,實現從「資訊化、網路中心」轉變為「智慧化、知識中心」,輔助指揮者解決指揮領域的感知、理解、認知問題。指揮資訊系統的系統架構、技術架構都會改變。系統圍繞著態勢、指揮、控制、保障等功能域進行相應的智慧技術應用,提升作戰指揮的認知與決策效能。外軍十分關注作戰指揮智能化應用,美軍DARPA從2007年至今,發布了3份關於人工智能國家及軍事發展戰略白皮書,分別開展了“深綠”[3⇓-5]、“第3次抵消戰略的高科技聖杯”、“指揮官虛擬參謀”等計劃,在情報感知與戰術決策領域啟動了“Insight”、Xdata” 「分散式戰場管理」、「人機協作」、「Mind’sEye」、「Trace」、「人機協作」、「X-Plan」、「認知電子戰」、「AlphaAI空戰」等人工智慧專案,實現戰場情報深度理解、態勢認知預測及戰術方案自動生成與推演能力。此後,美軍也設置了「自主協商編隊」、「大狗」、「蜂鳥」等項目,提升有人與無人協同控制能力。整體而言,美軍目前具備全球領先的智慧化作戰指揮水準。此外,德、法、俄等國也紛紛在智慧化資訊感知與處理、智慧自主無人作戰平台等方面進行了大量研究,取得了豐碩的研究成果[6⇓-8]。
1.2 戰術邊緣指揮控制
隨著軍事科技的發展,傳統大規模集群作戰方式逐漸轉換為小範圍的非對稱作戰,戰術邊緣的作戰活動在戰爭中將扮演重要角色。戰術邊緣又稱為「第一戰術英里」[9],它遠離指揮中心,通信、計算、服務資源受限,通常由作戰平台、戰術分隊、特種單兵組成,為了獲得信息與決策優勢,各級指揮單元利用泛在網絡、微雲等技術,實現信息與資源共享。戰術邊緣的移動計算設備,採用霧計算方法,整合為更大的作戰單元,形成自組網下的微雲,戰術邊緣獲取的大量態勢信息,在戰術微雲進行計算、存儲、共享,簡化了與指揮中心的交互規模,提升了信息交互時效,解決以往戰術前沿服務能力不足的問題。
1.3 多域戰指揮控制
2016年美陸軍提出「多域戰」概念[10],將「同步跨域火力」與「全域機動」作為核心要素,推動作戰要素高度融合,增強全域打擊能力,試圖消除中俄等國的「反介入/區域拒止」能力,主要具備以下三個特徵[10]。一是作戰領域向多維擴展,使美陸軍能夠從地面向海、空、電、網等多個作戰域投送力量,具備與其他軍種融合能力。二是作戰要素高度融合,各軍兵種及作戰功能域之間能夠共享資訊、統籌戰術、同步行動,推動了軍種聯合向作戰能力要素融合轉變。三是指揮鏈向扁平方向發展,指揮機制高效靈活,既要集中計劃、分散執行,又要向各指揮節點和單兵共享信息和指令,延伸戰術指揮鏈,實現快速、多線、多域作戰指揮。
1.4 馬賽克作戰指揮控制
2017年,DARPA提出「馬賽克戰」的概念[11-12],兼顧「基於威脅」與「基於能力」的裝備建設方法,將多作戰域的感測器、指控節點、戰鬥平台以及相互協作的有人、無人系統進行按需靈活組合,形成任務系統。系統整合採用搭積木的方式,將分散的細粒度系統動態連結在一起,構成類似「馬賽克區塊」的作戰體系。 「馬賽克戰」,借助智慧化決策工具,提供分散式態勢感知與自適應規劃、控制,輔助進行作戰任務規劃,實施分散式作戰管理。 「馬賽克戰」要求以自適應體系重組取代固定式作戰力量編成,作戰指揮具有韌性適變的資訊體系,能面向任務、按需定制物理分散的混合編成的作戰單元,滿足各種動態、協同作戰需求[12⇓-14]。
2 第五代指揮資訊系統特徵轉變
1) 體係由網路中心轉變為知識中心。以網絡為中心帶來戰場資訊優勢,並向認知優勢、決策優勢轉變,系統間由資訊共享走向以知識為中心的智力共享,促進整個指揮體係向決策及行動優勢轉變。
2) 雲端架構轉向雲端端一體化。拓展原有的雲端資源共享能力[2],向戰術邊緣的平台、分隊、單兵延伸,實現移動環境下戰場中心雲、移動雲、邊緣微雲的一體化混合服務能力,提升戰術前沿資源服務能力。
3) 預定整合向韌性適變轉變。目前系統依預設規則部署運作,任務變更時,須依預先方案進行調控。未來戰場系統易受攻擊而癱瘓,要求系統在發生擾動時,具備自重構韌性適變能力,保證核心任務不間斷[13-14]。
4) 由計算智能轉變為認知智能。智能化表現在計算智能、感知智能、認知智能,目前計算智能提供了戰術確定性求解方法,未來戰場須在海量情報處理、態勢認知與決策推理等方面透過智能化技術提升認知的精準度、實時度。
5) 性能由固化既定轉變為自主學習演化。系統的演算法、性能一般在設計期就被決定與固化,性能的提升透過升級改造完成。智慧化系統具備自學習、自演化能力,可在線上進行態勢感知、智慧決策的演算法學習,提升系統效能。
6) 建設由基於能力轉變為基於知識。指揮資訊系統一般基於能力要素進行建構,系統整合以能力要素進行綜合整合,智慧化系統,更加關注系統的智力建構,聚焦系統的知識、規則、演算法、資料的建構。
7) 互動方式向人機融合智慧互動轉變。人機融合智能感知、擬人化交互、面向意圖的智能人機界面交互、可穿戴的人機融合計算協同於一體、融合聯動的交互模式,將成為未來系統主要交互模式,以人禦機的系統向人機融合進展。
8) 戰訓分離轉變為戰訓演研一體化。第五代指揮資訊系統將作戰指揮與戰術訓練緊密耦合,具有平行模擬、推理能力,既能更新智慧演算法,也可進行戰法對抗研究,取得戰術資料,促進演算法學習。演習訓練由兵棋推演向戰場虛擬賽局發展。
3 總體架構設想
未來第五代指揮資訊系統的整體架構應該是知識中心、人機融合、智慧賦能、雲邊一體、自主演化、韌性適變的指揮資訊系統。下文主要圍繞系統架構、服務架構、技術架構等主要視角對系統總體進行闡述[15],其中系統架構主要指系統邏輯要素組成及其關係,服務架構描述系統之間的資訊與計算資源的整合模式,技術架構描述了系統的技術參考模型。
3.1 系統架構設想
該系統從「資訊化、網路中心」轉變為「智慧化、知識中心」,同時向戰術邊緣延伸。系統綜合整合在原有基礎上,進行知識與演算法的共享整合,在態勢、指揮、控制、保障等功能域進行智慧化技術應用,提升作戰指揮的認知與決策效能。系統架構設想如圖1所示。
圖1 第五代指揮資訊系統架構設想
第五代系統在態勢感知、指揮決策、行動控制、支援保障、資訊服務等功能要素基礎上,擴展平行推演與學習訓練功能域,滿足作戰分支評估及演算法的學習需求。在態勢認知方面,涵蓋了計算智能、感知智能與認知智能,主要完成戰場情報處理及目標識別,對態勢進行理解、預測,有態有勢,提升信息優勢;指揮決策方面,以認知智能為主,能夠機器戰術推理、生成方案與計劃,提升決策水平;行動控制方面,以計算智能與認知智能為主,能夠完成任務監控及臨機戰術控制,提供知識推理的行動優化策略,例如指揮引導、火力協同、無人集群智能控制;綜合保障方面,以計算智能為主,在先驗知識與規則下,完成戰場資源的優化調配;平行推演與學習訓練方面,將指控與仿真訓練結合起來,平時訓練人員以及算法
此外,第五代系統具有自主演化的學習機制:一是節點內自主學習,優化演算法與知識庫;二是節點間透過指揮雲共享智慧演算法與知識,協同完成演化,各節點可將學習後的演算法與知識上傳至指揮雲,更新知識中心的演算法及知識;三是系統向戰術學習、武器節點、偵測節點運作保障
第五代系統之間,在原有基於雲/端架構的綜合集成基礎上,增加了面向知識與智能算法的集成共享方式,各指揮信息系統將智能算法與知識規則上傳到知識中心,供戰場探測、指揮、武器等異構節點進行即插即享,指揮信息系統可以從知識中心獲取已有的智能知識,結合其二次戰場數據提升自身的戰場數據進行學習能力。指揮雲最終形成戰場的智慧知識中心,各智慧化指揮資訊系統之間形成戰場知識網。
3.2 雲端邊端服務架構設想
未來泛在網路連結將從指揮單元向戰術邊緣的各類分隊、單兵、平台延伸。第五代指揮資訊系統將利用霧運算、分散運算技術,在雲端架構技術基礎上建構戰術移動雲、分隊微雲(Cloudlet)、單兵任務組皮雲(Pico-Cloud)[9,16],形成戰術前沿移動雲服務能力,實現戰場集中作戰雲、移動戰術雲、邊緣微皮鏈雲的混合服務能力,前沿移動雲服務能力,實現戰場集中作戰雲、移動式戰術雲、邊緣微皮鏈雲的混合服務能力,形成「雲、邊、指揮」結構的快速構建能力。如圖2所示。
圖2 第五代指揮資訊系統雲端端服務架構設想
雲端端一體化服務能力支援第五代系統以「雲端部署、雲端聚合、雲端攻擊、雲端消散」等方式,實現作戰資源動態聚、釋能,提升整個體係作戰效能[17]。集中式作戰雲採用固定雲的方式部署在指揮中心[16],為各類作戰節點提供服務;空中、陸上、海上戰術雲為戰術前沿的飛機、艦艇、裝甲等兵力提供移動條件下的信息、算法、計算、存貯服務,提升了戰術前沿的資源共享水平[9,16,18-19];微雲及皮雲,微雲以霧計算方式部署在距離前沿接敵分隊通信一跳距離的車、機、艇上,擴展前沿分隊人員的戰術信息處理與共享能力,當單兵及分隊無法訪問微雲時,可利用移動自組網與分散計算技術構建皮雲,支持戰術邊緣弱連接下,端到端的信息匯聚到端的信息匯聚到端。
3.3 技術架構設想
第五代指揮資訊系統將戰爭從物理域、資訊域延伸到認知域,將改變指控方式,其技術架構如圖3。
圖3 第五代指揮資訊系統技術架構設想
第五代指揮資訊系統在第四代指揮資訊系統的網路化運算環境基礎上,增加戰術邊緣服務、智慧運算環境,既相容系統的架構,又滿足系統的智慧化要求。戰術邊緣服務運算環境為弱連結終端提供微雲及皮雲的基礎運算、存貯、資訊服務平台;智慧化運算環境為態勢、決策、控制、人機互動提供智慧服務。
智慧科技環境層包括以下五部分內容。智慧型運算硬體平台,配置了GPU、FPGA、TPU等AI加速處理器,適應深度學習所要求的運算能力,個別演算法採用神經元處理機制的類腦晶片或固化的專用智慧運算晶片;智慧資料管理平台,主要進行資料、樣本、案例、模型、知識的管理;深度學習架構,整合了深度學習、強化學習的運行庫及基本演算法庫;傳統人工智慧計算框架,包括了spark、bigflow等用於搜尋求解、資料探勘、平行處理等方面的傳統演算法支援庫;智慧服務,包含了面向應用的智慧演算法服務庫,如智慧交互辨識、估值網計算、策略網計算等服務,為應用開發提供求解介面。
智慧應用層,主要提供智慧化態勢認知、規劃決策、行動控制及資訊服務、人機互動、學習與訓練等功能要素,是系統主要面向使用者的功能介面,是智慧化要解決的核心問題。
上述的第五代系統技術架構模型,主要利用雲端運算與智慧化技術的支援服務,實現系統間的態勢、指令及演算法與知識的共享,同時支援系統自主演化、演算法升級、知識更新。系統智能化可分為0~4級[20]。 0級,完全人工控制;1級,實現計算智能,實現確定性的複雜戰術計算與資訊自動化處理;2級,具有一定感知智能,能夠理解、評估、預測戰場態勢;第3級:具有認知智能,能提供機器決策及決策推演能力;4級,具有人機融合與共生能力,核心算法能夠自學習、自演化。目前第四代系統的智慧化水準一般處於1級,態勢理解、指揮決策仍由人把控。第五代系統的智能化可經過三個階段達到第4級,第一階段實現戰場態勢感知、理解與評估能力;第二階段構建戰法知識庫,能基於規則、知識、算法實現機器決策;第三階段實現核心任務的機器自學習、自演化,具備自主方案決策功能,達到人機融合的高度智能化水平[20]。
4 系統關鍵技術及其智慧化設想
第五代指揮資訊系統的關鍵技術主要解決上述智慧化、雲端端整合、系統韌性適變問題。系統關鍵技術及其智慧化設想如圖4所示。
圖4 系統關鍵技術及其智慧化設想
第五代指揮資訊系統的關鍵技術涵蓋指控OODA環的所有面向,能夠支撐系統從探測、決策、控制、打擊等方面的智能、韌性、邊緣指控要求,從而構建精準感知鏈、快速控制鏈、精確打擊鏈、敏捷服務鏈,向戰術邊緣延伸,提升指揮效能。
1) 態勢感知機器分析技術
情報整編分析技術。利用大數據及深度學習、知識圖譜等技術進行資訊智能關聯匹配、文本語義智能分析、輿情智能搜索與提取,從海量、多源、異構的戰場信息中獲取有價值情報。
多元目標快速辨識技術。利用深度學習方法,建構多層CNN卷積神經網路,採用樣本特徵參數學習完成光學、紅外線、電磁、聲學資訊進行特徵提取與目標快速辨識。
態勢認知與理解技術。對敵進行作戰意圖、作戰能力分析,利用強化學習的估值網絡技術,模擬指揮員態勢認知的過程,結合CNN非線性戰場態勢擬合能力,建立態勢圖像到態勢理解的映射[22]。
態勢機器預測與評估技術。在態勢理解基礎上,對敵戰術行為進行預估,先利用策略網絡獲得敵方活動規律,再採用平行推演方法,進行多分支態勢推演,最後構建預測網絡進行態勢預測。
2) 作戰規劃機器決策技術
作戰任務空間及策略建模技術。對作戰任務空間的狀態及行動策略進行建模,確定任務狀態、策略、回饋的描述方法,是深度強化學習進行決策的基礎。
任務規劃機器決策技術。利用運籌優化完成目標分析、任務分配。利用深度強化學習、群體智慧演算法對兵力編成、火力配置、協同路徑進行機器規劃。戰術規劃偏向規則推理,易突破;戰役規劃偏向基於經驗的知識推理,涉及指揮藝術,較難突破。
作戰方案平行推演技術。參考「深綠」系統平行模擬技術[23],採用蒙特卡羅搜尋樹及博弈試驗方法,模擬敵作戰行為,對行動流程進行預演與評估,累積回饋賞罰函數,供學習訓練、最佳化決策。
作戰計劃智慧生成技術。利用自然語言理解、語音指令辨識、草圖辨識等智慧感知演算法,結合任務模型的要素提取,利用知識圖譜將方案進行自動提取生成作戰計畫與指令序列[24]。
臨機快速決策技術。基於當前態勢,利用博弈平台累積的學習資料,自動配對最適當的預案調整,基於蒙特卡羅樹搜尋及遷移學習演算法對計畫進行動態決策,反向強化學習,增強計畫泛化能力。
3) 行動控制智慧化技術
基於態勢的臨機行動控制技術。根據作戰行動的效果及偏差,對任務的資源、路徑、協同模式進行動態調整,利用平行模擬多支推演與強化學習技術進行糾偏,實現戰術「前饋式」的控制[4]。
群體智慧協同控制技術。促進戰場智能體協同作戰全局效能最大化,利用蟻群、蜂群控制演算法及深度強化學習方法,建構全局戰術價值網絡,建立效果回饋模型,根據價值網絡進行策略控制。
火力協同控制技術。提升敵我辨識、火力分配、協同調度的速度與精度,利用群智能及深度強化學習演算法自動規劃、協調優化打擊鏈,具備一定自主決策能力。
4) 有人/無人協同指揮技術
多域叢集系統自主協同機器規劃技術。利用分支搜尋求解、知識推理、深度強化學習進行有人/無人系統的協同任務規劃與分配,利用群智能最佳化演算法規劃無人、有人平台的協同軌跡。
多域集群系統自主協同指揮控制技術。對無人群集的巡航進行任務監控及自主協同指揮引導,利用群體智慧演算法進行多無人平台任務衝突偵測及避碰控制,進行編組、路徑、載重等調配。
5) 智慧化資訊服務技術
戰場資訊智慧共享技術,利用強化學習及語意關聯技術分析使用者的資訊需求及偏好,產生基於使用者差異化特徵的資訊需求,為使用者智慧推送戰術資訊。
6) 人機融合智慧化互動技術
人機融合智慧感知互動技術。建構多通道包含草圖、口語、手勢、頭勢、表情、眼動等多方式的人機互動手段,提供自然、靈敏、精準、擬人化的互動策略[5]。
面向意圖的智慧人機介面技術。利用FCM模糊認知互動推理技術,推理使用者的互動意圖,根據使用者的介面需求與互動喜好,整合不同的口語、手勢、草圖、自然語言等手段,組織互動介面輸出。
智慧穿戴式人機融合技術。採用邊緣運算技術,利用語音、手勢、眼動、腦機介面、擴增實境等新人機互動方式,為單兵提供智慧穿戴裝置,具備協同一體、融合連動的人機互動模式。
7) 虛擬博弈與訓練評估技術
作戰虛擬賽局技術建構賽局對抗試驗平台,進行作戰知識建模,利用平行模擬、分支決策、微分對抗等技術,進行紅藍對抗,既訓練戰術、戰法,又採集戰術資料。
機器訓練與評估技術,利用博弈平台累積的資料以及人員的經驗建模,採用小樣本遷移學習技術進行演算法的訓練與優化,對真實資料事後重播,對決策模型進行遷移學習優化,更新決策方案。
8) 系統韌性適變重建技術
環境感知與自主故障偵測技術。在軟硬毀傷下,進行主故障檢測、異常關聯分析,預測影響任務執行的故障發生,評估故障對任務的影響,實現對系統資源及故障的主動感知與快速定位。
系統自癒重構智慧技術。當系統關鍵節點失效時,採用適變機制,重新分配資源,實現能力再生,持續保障核心任務完成。由預置規則、人工參與的故障修復方式轉變為智慧化的系統重構方式。
9) 戰術邊緣運算技術
行動微雲服務平台技術。以霧運算方式部署在距離接敵一跳距離的車、機、艇上,為作戰分隊提供共享處理能力,擴展分隊人員的戰術資訊處理能力。
弱連接自組網下的皮雲資源共享技術。在單兵自組網基礎上,採用分散運算技術建構皮雲,支援弱連接下,端到端自主協同的資訊共享與單兵移動設備之間資源共用,滿足戰術邊緣需求。
5 發展思路設想
1) 分階段先易後難循序漸進。第一階段將圖像、語音、手勢、臉譜辨識及自然語言理解等應用到情報分析中;第二階段將深度學習、強化學習應用到態勢認知、指揮決策中;第三階段利用雲端運算實現知識中心,智慧賦能的系統[6]。
2) 選取智慧演算法進行應用。圍繞深度學習在態勢方面的應用、深度強化學習在規劃決策方面的應用,選取合適的戰術背景,對智能演算法進行驗證,可選用戰術層面的路徑、火力、任務等規劃作為突破口[25]。
3) 強化作戰指揮領域知識工程建設。專家規則、軍事條例、實戰資料是指揮智能化的基礎,對現有作戰規則進行知識化建模與表示,建立知識表示與深度學習的輸入、輸出映射關係,加強知識學習、知識推理的方法研究[4]。
4) 建立虛擬對抗博弈平台累積資料。智慧演算法需要大量學習樣本,樣本累積途徑有:①建立對抗賽局平台進行兵棋推演、人機對抗、紅藍對抗,累積資料;②收集實戰演習的戰術資料,進行建模作為訓練樣本[21]。
6 結束語
本文提出了第五代指揮資訊系統的總體及智慧化設想,建構了「智慧賦能、人機融合、雲邊一體、自主演化、雲智共享、韌性適變」的新一代指揮資訊系統架構,對其關鍵技術、能力特徵進行分析,試圖在國際上第四代系統的基礎上[2],實現認知優勢、決策優勢、行動優勢。國際上用於第五代系統的技術驗證不多,不可急功近利,仍需充分研究。
中國原創軍事資源：https://www.zhkzyfz.cn/EN/10.3969/j.issn.1673-3819.2021.05.00881