CYBERNETICS AND PHYSICS, VOL. 8, NO. 4, 2019, 257–261 THE MULTI-AGENT APPROACH FOR DEVELOPING A CYBER-PHYSICAL SYSTEM FOR MANAGING PRECISE FARMS WITH DIGITAL TWINS OF PLANTS Vladimir Laryukhin, Petr Skobelev Samara State Technical University Russia vladimir.larukhin@live.ru, petr.skobelev@gmail.com Oleg Lakhin, Sergey Grachev Knowledge Genesis Group Russia lakhin@kg.ru, grachev@kg.ru Vladimir Yalovenko, Olga Yalovenko, Peschanokopskaya Agrarian Laboratory Russia vladimiryalovenko@mail.ru, olyayalovenko@gmail.com Article history: Received 10.12.2019, Accepted 26.12.2019 Abstract The paper presents the multi-agent approach for devel- oping cyber-physical system for managing precise farms with digital twins of plants. It discusses complexity of the problem caused by a priori incompleteness of knowl- edge about factors of plant growth and development, high uncertainty of crops cultivation, variety of weather, business and technical requirements, etc. The approach proposes knowledge bases and multi-agent technology in combination with machine learning methods for de- signing considered systems. Digital twin of plant is specified as an agent based on ontology model of ob- jects relevant for plant cultivation (specific sort of plant, soil, etc) associated with history of operations and en- vironment conditions. The architecture and functions of system components are designed. The expected results of system implementation and the benefits for farmers are discussed. Key words Cyber-physical system, multi-agent system, knowl- edge base, digital twin, plant cultivation, precise farm- ing. 1 Introduction Cyber-physical systems (CPS) is a new type of sys- tems integrating computation, communication and con- trol components, including sensors, actuators and net- work connectors [Rajkumar et al., 2010]. Starting from embedded systems, CPS now are evolving as “smart systems” with fusion and interaction of real and virtual worlds, which can provide growing level of autonomous behavior with recognition of patterns, decision making and collaboration with users. Multi-agent systems (MAS) are considered as a power- ful tool for bringing a number of smart-* features to CPS [Leitao et al., 2016]: “MAS technologies share common ground with CPS and can empower them with multitude capabilities in their efforts to achieve complexity man- agement, decentralization, intelligence, modularity, flex- ibility, robustness, adaptation, and responsiveness”. The concept of digital twin was initially associated with product life-cycle in manufacturing [Tao et al., 2018]. For example, for aerospace vehicle “digital twin is a kind of ultra-high fidelity simulation integrating with an on-board health management system, maintenance history, and historical vehicle and fleet data. It can mir- ror the whole life of a specific flying physical twin (or tail number), which enables significant gains in safety and reliability”. Big data analytics and digital twins are considered as a basis for smart manufacturing [Qi and Tao, 2018]. Research subject of this paper is the key requirements, functionality and architecture of cyber-physical multi- agent system (CP MAS) for precise farm management. As a part of the designed system digital twins of plant will be introduced. Digital twin will be designed as computer model of plant growth and development based on available domain knowledge and various data ac- quired during the process of plant cultivation. Ontology- driven knowledge base, multi-agent technology and ma- chine learning methods will be proposed for creating and maintaining digital twin of plants for mirroring state and