Abstract: This paper proposes a sensor management system framework and presents a multi-agent-based solution. This structure achieves sensor task allocation through mutual negotiation among multiple agents, effectively overcoming the shortcomings of the fusion center. Furthermore, it focuses on the coordination and cooperation among agents, implementing a coordination mechanism based on the KQML language. Keywords: multi-agent system; sensor management; data fusion; KQML **Research and Design of Sensor Management System Based on Multi-Agent** LI Wei, ZHANG Feng-ming ** Abstract:** This paper presents a framework for a sensor management system, providing a solution using multi-agent technology. This architecture utilizes agent negotiation to achieve sensor task allocation and overcomes the disadvantages of hybrid architectures. The agent coordination problem and the cooperation mechanism between agents in this architecture are implemented using KQML. **Keywords:** multi-agent system; sensor management; data fusion; KQML 1 Introduction Multi-sensor systems can acquire more global information about their observation environment from different angles and perspectives, and are receiving increasing attention in both military and civilian fields. To adapt to changes in the modern battlefield and achieve optimal combat results, many researchers have begun to focus on the automatic or semi-automatic management of sensor resources to fully utilize the capabilities of each sensor. Meanwhile, with the continuous advancement and upgrading of command automation, more and more legacy systems are facing researchers. How to rationally utilize these legacy systems is also a problem that researchers need to solve. One solution is to package legacy components and provide them with "Agent-level" functionality, enabling them to communicate and cooperate with other software components. 2 Current Status of Sensor Management The core issue of sensor management is to determine which sensor to select for a target and the sensor's operating mode and parameters based on certain optimal criteria. The task of the Sensor Management System (SMGS) is to utilize limited sensor resources to meet the requirements of multiple targets and scanning spaces, in order to obtain the optimal metric values ​​for each specific characteristic (detection probability, interception probability, sensor's own transmission capability, track accuracy or loss probability, etc.), and to rationally allocate sensor resources and select sensor operating modes and parameters according to optimal criteria. The current sensor management structure is divided into centralized and distributed. 2.1 Centralized Management Structure In a centralized management structure, the fusion center sends the tasks that need to be performed and the parameter set or operating mode for completing the task to all sensors. Its disadvantages are that the fusion center has difficulty in making real-time assessments of the load status of each sensor, which can lead to load imbalance during multi-tasking and even cause individual sensors to be severely overloaded and unable to complete the task. In addition, as the number of sensors increases, the computational load of the fusion center will increase sharply, and the communication load will also increase significantly. 2.2 Distributed Management Structure In a distributed management structure, management functions are distributed in different locations of the system or in different sensors. Its disadvantages are that information redundancy can cause serious problems. In addition, without a common communication facility, data exchange between nodes in the network can only be carried out strictly in an end-to-end manner. The delay between the sending end and the receiving end will cause momentary inconsistencies in the global state of different components in the network, resulting in a decrease in the overall performance of the system. Task conflicts and competition make task coordination more complex. 3 Sensor Management System Based on Multi-Agent Agents Agents are a conceptual model from distributed artificial intelligence. They usually refer to entities with goals, behaviors, and knowledge that can solve problems and perform corresponding activities autonomously in uncertain environments by planning, reasoning, and making decisions based on their own capabilities, states, resources, relevant knowledge, and external environmental information, in order to achieve a certain goal. A multi-agent system (MAS) is a group of interconnected systems linked together through communication or computer networks to achieve a common global goal. The agents within the system must negotiate, coordinate, and cooperate to complete shared tasks and achieve the global objective. MAS is valued in many fields due to its dynamic self-organizing capabilities and open infrastructure. 3.1 System Architecture and Features [align=center] Figure 1 Task Flow and External Interface Relationship of Sensor Management System (SMGS) [/align] ⑴ The decision-making process is distributed among the nodes on the network. This strategy allows each sensor node to leverage its own management potential (local managers make decisions about their own sensor tasks); ⑵ The global system task manager only needs to issue task requirements to its assigned sensors and then monitor the completion status (performance indicators) of the system tasks without needing to manage the sensors specifically; it is particularly noteworthy that the sensor agent settings give the sensors a high degree of autonomy and device independence (information exchange between sensors and fusion nodes, as well as between sensors, can be achieved through a standardized request/response query language and intelligence data chain), which creates favorable conditions for the mobile networking of sensors; ⑶ Each node in the system corresponds to a physical entity or decision-making entity, and they coordinate with each other through messages, making the system easy to construct; ⑷ By adopting a multi-agent-based system, the decisions made by the decision-maker in the model system can be quickly implemented in actual operation, making it easy to implement; (5) General simulation methods can only evaluate the effects of pre-set decision-making schemes, but cannot suggest new schemes. That is, they can only answer the "what-if" question, but not the "what's best" question. Multi-Agent systems provide the possibility of answering the "what's best" question. 3.2 Operation mechanism of each agent in the multi-Agent SMGS model (1) Fusion Agent: Provides target status and attribute information for target and sensor pairing. Performs data fusion on all relevant information sent by each sensor; determines the system tasks and global performance indicators that each sensor agent needs to complete in the next sensor management cycle; monitors the performance indicators of the system tasks to confirm whether the required performance indicators have been met. (2) Sensor Agent: Acquires target and sensor data; manages the sensor tasks assigned after negotiation; controls data communication with other network nodes. [align=center] Figure 2 Structure of Sensor Agent[/align] The function of the task execution layer depends on the type of sensor and its ability to complete sensor tasks. The task planning layer is responsible for determining the set of tasks that sensors need to complete in the next management cycle. It processes information from the fusion center or other agents, as well as information provided by operators, to acquire environmental situational knowledge and determine the priority of each task. Information transmission between the planning and execution layers within an agent is handled by the communication layer. This layer also supports the negotiation process in distributed decision-making, receiving execution requests, notifications, or results from other agents. The negotiation process requires knowledge of the sensor's ability to complete a task. The communication layer, through the planning layer, understands the agent's task completion status. Once a task suggested by another agent is accepted, the communication layer transmits the negotiated sensor-level performance indicators to other layers. If a task cannot be executed, the communication layer compares the required sensor performance indicators with the information in the planning layer to determine a set of actions that should be performed by other agents, or notifies other agents of the results. Only tasks that can be completed by agents are transmitted to the planning layer. (3) Scheme Generation Agent: Generates optional pairing schemes between "sensors" or "sensor groups" and tasks based on the pre-set "specialties" of each sensor. For a specific target appearing in the current monitoring space, a specific sensor or sensor combination is paired for further detection to provide information about that target. (4) Scheme Optimization Agent: Optimizes the above schemes according to task requirements, forming a target priority ranking. (5) Sensor Control Agent: A conversion module responsible for converting sensor allocation schemes into commands executable by the sensors, and also performing specific operations such as sensor mode selection and parameter selection. (6) Task Planning Agent: Responsible for specific sensor task allocation. Based on the above aspects, a sensor management scheme is formed, directly adjusting the sensor configuration. Therefore, the agent-based sensor management system plays a feedback adjustment role in forming the closed-loop control mode of the data fusion system, increasing the system's robustness. 4. Coordination Mechanism between Multiple Agents SMGS emphasizes the collaborative work between various entities; therefore, whether the predetermined tasks can be completed in a coordinated manner is one of the key issues in the application of multiple systems. Coordination mechanisms between agents can be implemented in various ways; communication is an essential means for each to obtain information for negotiation and coordination. Our research focuses on the application of KQML (Knowledge Query and Manipulation Language)-based agent communication and coordination mechanisms in the SMGS system. 4.1 KQML is indispensable for any type of collaboration, method, and language among agents. Communication languages ​​have seen significant development. These languages ​​define syntax and semantics for inter-agent communication. KQML is a language and protocol for exchanging information and knowledge between software, providing a standard format for message expression and processing. Its main advantages include: (1) flexible structure and good extensibility; (2) independence from network transmission mechanisms; (3) independence from content layer expression languages; (4) ability to meet the basic requirements of inter-agent information transmission. For example: Agent A sends an action expression to Agent B: 4.2 Communication Server In KQML, not all dialogues are a simple question/answer processing model. Each agent can use ask(x) to log its knowledge to the communication server. An agent uses agent speech acts to find other agents to provide the information it needs. The communication server receives the query and searches for agents with the appropriate knowledge. [align=center] Figure 3 Communication Server[/align] When coordinating in the SMGS system, each Agent reviews the activity. If it finds that an Agent's activity cannot be satisfied under its current conditions, it sends a coordination request to the relevant Agent. The Agents participating in the coordination can negotiate through blackboard or other means. 5 Conclusion This paper proposes a new sensor management method, namely, a management method based on multi-Agent technology. The sensor management system constructed in the above way utilizes the Agent's knowledge representation and coordination solution capabilities, and has the characteristics of simple implementation, high system efficiency, and low environmental requirements. It fully leverages the sensor's own decision-making capabilities, reduces the dependence of the decision-making process on the fusion center, and improves the survivability of the entire sensor monitoring network. In the next step, extensive simulation experiments should be conducted for different needs to analyze the performance of various management strategies, and the rule base should be modified and adjusted. In practical applications, different situations should be treated differently to make full use of system resources and improve system operating efficiency. References [1] Zhou Jinling, Sun Yan. Research on Distributed Measurement and Control System Based on Multi-Agent Technology. Microcomputer Information, 2003, 19(9): 15-16. [2] Lin Yuefang, Ji Haiyan. Anti-theft alarm system based on multi-sensor information fusion technology [J]. Microcomputer Information, 2003, 19(10): 58-59. [3] Svensson P. Information fusion in the future. Journal of Military Technology, 2002, 69(3): 2-8. [4] Durrant Whyte HA Beginner's Guide to Decentralized Data Fusion. University of Sydney, Australia. Technical Document of Australian Centre for Field Robotics, 2000. [5] Tian Kangsheng, Zhu Guangxi, Xu Yu. Sensor management system based on multi-agent technology [J]. Modern Radar, 2004, 26(2): 9-13. [6] Yang Xiuzhen, et al. Quantitative management research of multi-sensor system [J]. Systems Engineering and Electronics Technology, 2001, 23(10): 69-72. [7] Yang Xiuzhen, et al. Research status and development trend of multi-sensor management system [J]. Sensor Technology, 2004, 23(1): 5-8. [8] Zheng Hongyuan, Li Haiyan. Supply chain management system based on multi-agent coordination [J]. Journal of Jilin University (Information Science Edition), 2005, 23(2): 172-178. [9] Shi Zhongzhi. Intelligent agents and their applications [M]. Beijing: Science Press, 2000. [10] Fan Yushun, Cao Junwei, et al. Theory, methods and applications of multi-agent systems. Beijing: Tsinghua University Press, 2002.