Agents In Multi-agent System Research Articles

Онтолого-орiєнтована мультиагентна система для децентралiзованого керування групою БПЛА

Introduction. Today, UAVs are becoming an increasingly important tool for performing complex tasks in various fields of application, both civil (economic) and military, as they are particularly effective in dynamically uncertain environments with hard-to-reach areas. In addition, technological advances such as blockchain, artificial intelligence (AI) and machine learning have enabled the development of updated and improved UAV systems. To create and deploy a swarm of UAVs, coordinate actions, manage, and exchange data, a model of a multi-agent system (MAC) based on an ontological representation of knowledge is proposed. This model enables a swarm of UAVs to effectively make decisions in various situations while performing assigned tasks. This approach enables the safety, reliability, and efficiency of the tasks of the UAV group. The purpose of the paper is to develop further the theoretical and practical foundations of the integration of the multi-agent system (MAS) based on the ontological representation of knowledge with the UAV network. This involves the development of a MAS architecture and a hierarchical set of ontologies of different levels. The goal is to create a common data description lan guage, define data semantics to ensure data uniqueness and consistency, provide support for decision-making during UAV swarm management, and swarm survivability in the event of aircraft failures or loss. It is necessary to develop algorithms and a method of dividing a complex task into sub-tasks in a swarm of UAVs among all MAS agents. This is to ensure reliable exchange of messages (data) between agents during the joint performance of the assigned task, and the possibility of dynamic redistribution of roles between UAV agents as needed. Methods. During the research, the general theory of intelligent information technologies was applied; agent theory methods in particular intelligent BDI agents; methods of analyzing the performance of wireless data exchange networks; theory of combinatorial optimization for dividing tasks into subtasks; methods of ontological analysis and descriptive logic to create an ontological hierarchical model of the subject area; methods of enriching ontological models from external semantically marked information resources. Results. As a result of the performed scientific research, the MAS architecture was proposed and its main functions were determined for the decentralized control of a swarm of UAVs. A set of agents with assigned roles was formed, who jointly (cooperatively) perform tasks, exchanging messages, and information with each other, which ensures the survivability of the system (in case of a failure or loss of the device, its task must be distributed among other drones). Plans and scenarios of MAS actions for various situations and means of coordinating actions between agents have been developed to perform the mission by a swarm of UAVs. A hierarchical ontological model of the subject area related to the work of the UAV swarm has been created. The algorithms and methods were based on the integration of semantic technologies that support the MAS during the execution of the UAV swarm mission, decision-making, assessment of the dynamic environment, and response to its changes. Conclusions. An original approach, algorithms, and method for improving the system of decentralized control of a group of UAVs were proposed. Expanding the functionality of the system for maintaining the interaction of a swarm of unmanned systems based on MAS artificial intelligence was suggested. This system was based on ontological models. The models describe knowledge of the subject area, processes of UAV swarm operation, scenarios of actions in difficult situations, distribution of roles to agents, principles of planning, and coordination. The proposed MAS is integrated with the UAV swarm software platform, which makes it possible to improve the efficiency of the decentralized control system and adapt UAVs to dynamic changes in the environment. The practical result of the work will be a prototype of a software agent system that interacts with ontologies while performing simple tasks. The economic significance of the work consists of focusing on the creation of new intelligent information technologies, which were based on AI and knowledge of the subject area, and this significantly increases the efficiency of the functioning of modern systems. Keywords: multi-agent system, ontology, formalization of knowledge, UAV, drone, decentralized control, task allocation.

Information manifold and fault detection of multi-agent systems

With the increase of the number of agents in multi-agent systems and the rapid increase of the complexity of the overall structure of the system, the fault detection and diagnosis work has brought great challenges. Researchers have carried out considerable research work on fault detection and diagnosis of multi-agent systems, but there is no research on fault state estimation and diagnosis based on the information and state of the whole multi-agent system. Based on the global perspective of information geometry theory, this paper presents two new physical quantities of the information manifold of multi-agent systems, as Lagrangian and energy–momentum tensor, to express the state of the overall information of multi-agent systems, and to characterize the energy state and development trend of faults. In this paper, two new physical parameters are introduced into the research of multi-agent fault detection and diagnosis, and the fault state and trend of multi-agent system are evaluated from the global perspective, which provides more comprehensive theoretical support for designing more scientific and reasonable fault diagnosis and fault recovery strategies. Simulation of the application example confirms the competitive performance of the proposed method.

Physics-Informed Graph Neural Operator for Mean Field Games on Graph: A Scalable Learning Approach

Mean-field games (MFGs) are developed to model the decision-making processes of a large number of interacting agents in multi-agent systems. This paper studies mean-field games on graphs (G-MFGs). The equilibria of G-MFGs, namely, mean-field equilibria (MFE), are challenging to solve for their high-dimensional action space because each agent has to make decisions when they are at junction nodes or on edges. Furthermore, when the initial population state varies on graphs, we have to recompute MFE, which could be computationally challenging and memory-demanding. To improve the scalability and avoid repeatedly solving G-MFGs every time their initial state changes, this paper proposes physics-informed graph neural operators (PIGNO). The PIGNO utilizes a graph neural operator to generate population dynamics, given initial population distributions. To better train the neural operator, it leverages physics knowledge to propagate population state transitions on graphs. A learning algorithm is developed, and its performance is evaluated on autonomous driving games on road networks. Our results demonstrate that the PIGNO is scalable and generalizable when tested under unseen initial conditions.

Identifying Disconnected Agents in Multiagent Systems via External Estimators.

This article addresses the problem of identifying disconnected agents in multiagent systems via external estimators. Specifically, we employ external estimators with an appropriately designed decision rule to identify the disconnectedness (i.e., the status of being disconnected) between two arbitrarily chosen agents in formation-control multiagent systems. The design of the decision rule is inspired by the unit-root testing problem of autoregressive time series. To make the best possible decision, a best-effort procedure is also proposed. Then, by introducing the concept of connected components (or just components) in graph theory, and using the methods of consensus analysis and time-series analysis, we develop an analytical framework to show the theoretical performance of the designed decision rule. A particularly important result shown by our analysis is that the miss probability of the decision rule can converge to 0 as the number of data samples increases. Finally, simulation results validate the performance of the decision rule and the best-effort procedure, showing that they can perform well even in small samples.

The use of multi-agent systems in an automated restoration process for power distribution networks

The capacity of future smart grids to automatically detect problems, isolate them, and restore service is a key feature. Traditional mathematical programming techniques and heuristic approaches become computationally highly expensive when used for the restoration issue, which is a comprehensive nonlinear optimization challenge. Distributed control must offer comfortable and safe power system operation and improve consumer supply quality. In this article, we suggest the multi-agent system (MAS) as the creative process that applies to the existing power distribution network. The MAS agents are installed in "Intelligent electronic devices (IEDs)" and interact with one another using open protocols from the "Foundation for Intelligent Physical Agents (FIAP)." A Java-based distribution system simulator was built to test how well MAS works. In-lab experiments were conducted with the simulator connected to the MAS. The test findings demonstrated that the MAS effectively restored healthy divides by changing them to their original configuration following a predetermined sequence, notwithstanding the system's operating limits. The suggested MAS has shown the efficacy of a distributed control method that uses in the absence of the technologies currently employed in IEDs for control and protection.

Application of Multilevel Thresholding and CNN for the Diagnosis of Skin Cancer Utilizing a Multi-Agent Fuzzy Buzzard Algorithm

This paper presents a novel methodology for skin cancer diagnosis based on deep learning and metaheuristics. The method combines a multi-level optimal thresholding segmentation technique based on the histogram of the input images with a new improved metaheuristic algorithm, based on Multi-agent Fuzzy Buzzard Algorithm (MAFBUZO) for this purpose. The MAFBUZO combines local search agents in multi-agent systems with the global search ability of the BUZO algorithm, enabling an appropriate balance of exploitation and exploration steps during optimization. To show the algorithm efficiency, it has been validated with four benchmark functions and its comparison with some other bio-inspired methods shows that using the proposed MAFBUZO, with 16.3459–16, 9.6538, 2.7412, and7.3258e-5 error provides the minimum error values during the minimization which shows its effectiveness to use in our application. After the region of interest segmentation, the images have been injected into an optimum convolutional neural network based on MAFBUZO for feature extraction and the final diagnosis. The technique is evaluated against Dermquest and DermIS datasets, and the obtained results are compared against those provided by other published techniques. The results show that using the proposed method with 0.95, 0.88, 0.94, and 0.93 values for NPV, PPV, Accuracy, and Specificity, provides the best outcomes against the other comparative approaches. The final results show that this method improves the results of the existing techniques in terms of overcoming weaknesses such as being trapped in local optimal points or having raw convergence issues.

Special Issue “Advances in Multi-Agent Systems”: Editorial

Multi-agent systems (MAS) are collections of autonomous computational entities (the agents) capable of pro-actively pursuing goals and re-actively adapting to environment change. Agents in MAS exploit their social abilities, interacting with peers, and their situated capabilities as well, by perceiving and acting on the world around them. From distributed computing to intelligent systems, the relevance of agents and MAS as software abstractions is steadily growing as they are extensively and increasingly used to model, simulate, and build heterogeneous systems across a huge variety of diverse application scenarios and business domains, ranging from industrial manufacturing to robotics, from social simulation to applications, and more. The recent, renewed popularity of AI techniques has further spread the adoption of MAS, focusing in particular on the cognitive capabilities of agents, so that intelligent systems can be modelled and built as MAS. Along those lines, this Special Issue gathers five contributions that well represent the many diverse advancements that are currently ongoing in the MAS field.

A novel hybrid consensus-based variational Bayesian filter with position uncertainty in multi-agent system

Most existing distributed state estimation approaches assume that the position of agent in multi-agent system (MAS) is precisely known. However, in target tracking applications, the position of agent is noisy, which makes traditional distributed state estimation methods suffer from accuracy reduction. This paper focuses on developing a novel hybrid consensus-based variational Bayesian filter (HCVB) for target tracking with position uncertainty in nonlinear MAS. To improve the nonlinear approximation accuracy, high-degree cubature information filter framework is used. To cope with agent position uncertainty, the measurement update is derived by using variational Bayesian (VB) method to approximate the joint posterior distribution of the target state and agent position, which can estimate the state and agent position jointly. To decrease the required consensus iterations, a hybrid consensus method is proposed, which can achieve global consistency of all agents with a few consensus iterations. Simulations on UAV tracking in MAS verify the superiority of the proposed method.

Towards Robustness Analysis for Adaptive Artificial Intelligence in Multi-Autonomous agent systems

Individual autonomous agents can employ adaptive Artificial intelligence. The individual agent can respond fast, near real-time, to the environment and avoid hurdles and collisions while keeping a certain distance. Although the systems learn to handle the environment with new conditions and situations, it does not, per see, guarantee that the system is robust meaning the system performs predictably while its variables and assumptions are altered. Nor does it mean that adaptive individual autonomous agents in multi-agent systems perform well in dynamic, distributed and partially observable environments where unexpected things happen.This paper presents an approach to applying robustness analysis on adaptive Artificial intelligence in multi-autonomous agent systems. Adaptiveness modifies the agents’ learning by generating appropriate responses to new situations to achieve resilience to perturbations. The robustness analysis explores the quality of these responses so the agent can continuously operate despite abnormalities in input and consequently safely tolerate perturbations. A multi-agent prototype simulating the integration of adaptive agents with robustness analysis shows that it is possible to apply robustness analysis to the responses of adaptive Artificial intelligence. In the system, each autonomous agent collects data about the surrounding environment and applies adaptation to provide options, i.e., possible actions and decisions. The robustness analysis examines the options in order to validate and adjust them as needed. With these updated options, the agents retrain to resist malfunctioning and achieve resilience and robustness for given situations.

Supervisor Synthesis for Multi-Agent Systems With Mutually Exclusive Resources

In multi-agent systems (MAS), it is a basic mission to coordinate and control each agent to complete tasks on mutually exclusive resources due to the high cost of some hardware and software in reality. In supervisory control theory (SCT) of discrete-event systems (DES), a supremal supervisor is synthesized with respect to the system and all the specifications to make the system act as desired. However, the size of the plant model increases exponentially with the number of agents in MAS. To alleviate the state explosion problem when applying SCT to MAS, a synchronous product with constraints using bread-first search strategy is proposed to synthesize a supervisor for the system with respect to the mutex resources. And a framework for computing the supremal supervisor for the entire MAS with respect to all the specifications is presented. Experiments are studied to estimate the effectiveness of the proposed approach. Contributions of the research are two-fold. Firstly, the state explosion problem associated with supervisory control of MAS is alleviated in a certain extent. Secondly, compared with decentralized or distributed control, it is not necessary to design a coordinator additionally.

Game-based coordination control of multi-agent systems

In this paper, we are interested in how agents interact with each other in multi-agent systems. First, we model the interactions among agents in multi-agent systems as a multi-player game. The topology of the interactions among agents is a directed graph. We design cost functions for the game and assume that each agent in the systems tends to minimize its own cost. Then, the unique Nash equilibrium solution to the proposed multi-player game is obtained as the next state of the agent. A necessary and sufficient condition for achieving multi-agent consensus is established using the system transformation method and graph theory. Furthermore, we extend the result to the problem of containment control in the presence of leaders. The criterion for solving the containment control is also given in this paper. Finally, several simulation examples are given to verify the effectiveness of the theoretical results.

Method of Multi-Agent Reinforcement Learning in Systems with a Variable Number of Agents

Multi-agent reinforcement learning methods are one of the newest and actively developing areas of machine learning. Among the methods of multi-agent reinforcement learning, one of the most promising is the MADDPG method, the advantage of which is the high convergence of the learning process. The disadvantage of the MADDPG method is the need to ensure the equality of the number of agents N at the training stage and the number of agents K at the functioning stage. At the same time, target multi-agent systems (MAS), such as groups of UAVs or mobile ground robots, are systems with a variable number of agents, which does not allow the use of the MADDPG method in them. To solve this problem, the article proposes an improved MADDPG method for multi-agent reinforcement learning in systems with a variable number of agents. The improved MADDPG method is based on the hypothesis that to perform its functions, an agent needs information about the state of not all other MAS agents, but only a few nearest neighbors. Based on this hypothesis, a method of hybrid joint / independent learning of MAS with a variable number of agents is proposed, which involves training a small number of agents N to ensure the functioning of an arbitrary number of agents K, K> N. The experiments have shown that the improved MADDPG method provides an efficiency of MAS functioning com-parable to the original method with varying the number of K agents at the stage of functioning within wide limits.

Using Adaptive Logics for Expression of Context and Interoperability in DL Ontologies

Ontologies are logical theories that are used in computer science for describing different items such as web services, agents in multi-agent systems, or domain knowledge. Many ontologies exist, expressing various domains of knowledge with different abstraction levels (domain ontologies, top-level ontologies, and task ontologies are the usual categories). The conceptualization of the knowledge contained in an ontology is subject to change, whether because the context of its use changes, because the domain evolves, or because an ontology needs to interoperate with other elements using other ontologies. Change in logical theories is a form of defeasible reasoning, in which some formulas need to be added or removed from a knowledge base. Adaptive Logics (AL) is a logic managing defeasible reasoning that we investigate in this paper for managing change in ontologies expressed with Description Logics (DL). The adaptation of AL for DL will help express the context in which formulas remain valid or can be added to a DL knowledge base, and ease the interoperability between ontologies.

Consensus problems in multiagent systems: A vector based contraction approach

This paper addresses the exploitation of vector based contraction theory to synchronize the motion of agents in multiagent systems. The convergence analysis of the agents using this theory provides a flexible frame as the components of a vector distance norm follow less stringent conditions than classical contraction analysis. The key feature is that it encompasses the formulation of some comparison system, as a result, the problem of convergence analysis of a complex multiagent system reduces to the convergence analysis of a much simpler comparison system. Hence, despite the nonlinearities, the methods of designing consensus controls remain simple as compared to the existing literature. We address four different cases: 1) consensus of third-order dynamics multiagent systems subjected to acceleration and input constraints; 2) consensus of multiagent systems with heterogeneous nodes under communication imperfections; 3) synchronization of multiagent systems with connected and disconnected switching topologies; 4) synchronization in networked systems with time-varying couplings. Moreover, an effective synchronization control scheme for a class of multiagent systems with underactuated agent dynamics is developed using the method of aggregation of comparison systems. Finally, simulation results are illustrated that show the validity of the mathematical results.

Thou Shalt Love Thy Neighbor as Thyself When Thou Playest: Altruism in Game Theory

Game theory is typically used to model the interaction among (software) agents in multiagent systems and, therefore, is a key topic at leading AI conferences. Game-theoretic models, however, are often based on the assumption that agents are perfectly rational and narrowly selfish and are interested only in maximizing their own gains, no matter what the costs to the other agents are. This summary paper presents various ways of introducing certain notions of altruism into existing game-theoretic models in both noncooperative and cooperative games, in the hope that simulating altruistic behavior in AI systems will make AI better suit real-world applications—and thus may make the real world a better place.

Fully distributed dynamic event-triggered control for multi-agent systems with input saturation and actuator failure

This paper investigates the fully distributed dynamic event-triggered consensus problem for multi-agent systems with input saturation and actuator failure. Firstly, a distributed adaptive parameter is designed to estimate communication noise. Then, under the assumption that each agent is asymptotically null controllable with bounded controls, a low-gain state feedback control protocol is designed according to the algebraic Riccati equation. Furthermore, by introducing an internal dynamic variable, a novel fully distributed dynamic event-triggered control mechanism is proposed for multi-agent systems. Under this fully distributed dynamic event-triggered control mechanism, all agents of multi-agent systems can reach semi-global consensus and the Zeno behavior is excluded. Finally, a simulation example is presented to illustrate the effectiveness of the proposed algorithm.

Improving data collection and processing efficiency by using hierarchies of intelligent agents in multi-agent systems

The paper considers the challenge of improving data collection and processing efficiency by using various information and measurement systems. A structural scheme of a hierarchical multi-agent control system has been developed, incorporating horizontal (single-level) and vertical (multi-level) connections between intelligent agents. An algorithm is proposed for controlling a heterogeneous group of unmanned aerial vehicles participating in search and rescue operations in an emergency zone based on the hierarchical interaction of intelligent agents, cloning and coalition formation. KPI values have been calculated and the effectiveness of the proposed solution have been proved.

Multi-Agents Collaboration in Open System

Share constrained resources, accomplish complex tasks and achieve shared or individual goals are examples requiring collaboration between agents in multi-agent systems. The collaboration necessitates an effective team composed of a set of agents that do not have conflicting goals and express their willingness to cooperate. In such a team, the complex task is split into simple tasks, and each agent performs its assigned task to contribute to the fulfilment of the complex task. Nevertheless, team formation is challenging, especially in an open system that consists of self-interested agents performing tasks to achieve several simultaneous goals, usually clashing, by sharing constrained resources. The clashing goals obstruct the collaboration's success since the self-interested agent prefers its individual goals to the team’s shared goal. In open systems, the collaboration team construction process is impacted by the Multi-Agent System (MAS) model, the collaboration’s target, and dependencies between agents’ goals. This study investigates how to allow agents to build collaborative teams to realize a set of goals concurrently in open systems with constrained resources. This paper proposes a fully distributed approach to model the Collaborative Team Construction Model (CTCM). CTCM modifies the social reasoning model to allow agents to achieve their individual and shared goals concurrently by sharing resources in an open MAS by constructing collaborative teams. Each agent shares partial information (to preserve privacy) and models its goal relationships. The proposed team construction approach supports a distributed decision-making process. In CTCM, the agent adapts its self-interest level and adjusts its willingness to form an effective collaborative team.

Multi-Agent System-Based Hierarchical Protection Scheme for Distribution Networks With High Penetration of Electronically-Coupled DGs

With the increasing penetration of distributed generations, preserving the protection system coordination is of great importance for the distribution network operation. This paper presents a supplementary protection strategy for the multi-agent system (MAS)-based hierarchical protection scheme (HPS). The primary level of MAS-based HPS clears the fault and preserves the protection coordination while the higher levels update the relay settings. The key advantage of the proposed supplementary scheme is that the primary level can autonomously and quickly preserve the protection coordination when the main relay is during the time-consuming process of updating settings in the case of network operating condition change. By adopting converter-based DGs as the only MAS agents, the proposed supplementary scheme divides the distribution network into several zones that communicate together using point-to-point communication. Without communicating with higher protection levels, the supplementary protection limits the influence of DGs on the fault current by controlling the produced power of each zone. The effectiveness of the proposed strategy is validated through several simulation case studies on the Isfahan distribution network.

Consensus tracking control of discrete‐time second‐order agents over switching signed digraphs with arbitrary antagonistic relations

SummaryThis article studies the problem of consensus tracking control for second‐order agents in multiagent systems over switching signed digraphs. Compared with the existing consensus tracking works on the structurally balanced signed digraph where the antagonistic relations exist only between two independent subgroups, this article explores a more general case for the first time, in the sense that the antagonistic relation is allowed between any two agents. On the basis of the design of a cooperation‐antagonism environment‐based distributed algorithm, suitable model transformation vectors are utilized to convert the stability of original system into a product convergence problem of time‐varying superstochastic matrices. By analyzing the convergence, algebraic conditions between positive and negative weights are established to ensure that all followers can eventually reach the leader's state under switching signed digraphs with arbitrary antagonistic relations. Simulation examples are provided to demonstrate the effectiveness of our results.