Control Of Multi-agent Systems Research Articles

Spiking neural controllers in multi-agent competitive systems for adaptive targeted motor learning

The proposed work introduces a neural control strategy for guiding adaptation in spiking neural structures acting as nonlinear controllers in a group of bio-inspired robots which compete in reaching targets in a virtual environment. The neural structures embedded into each agent are inspired by a specific part of the insect brain, namely Central Complex, devoted to detect, learn and memorize visual features for targeted motor control. A reduced-order model of a spiking neuron is used as the basic building block for the neural controller. The control methodology employs bio-inspired, correlation based learning mechanisms like Spike timing dependent plasticity with the addition of a reward/punishment-based method experimentally found in insects. The reference signal for the overall multi-agent control system is imposed by a global reward, which guides motor learning to direct each agent towards specific visual targets. The neural controllers within the agents start from identical conditions: the learning strategy induces each robot to show anticipated targeting actions upon specific visual stimuli. The whole control structure also contributes to make the robots refractory or more sensitive to specific visual stimuli, showing distinct preferences in future choices. This leads to an environmentally induced, targeted motor control, even without a direct communication among the agents, giving robots, while running, the ability to perform adaptation in real-time. Experiments, carried out in a dynamic simulation environment, show the suitability of the proposed approach. Specific performance indexes, like Shannon׳s Entropy, are adopted to quantitatively analyze diversity and specialization within the group.

Demand side integration in LV smart grids with multi-agent control system

The low-carbon society requires significant changes in the operation and planning of distribution systems. In order to exploit existing low voltage systems, Demand Side Integration is essential for offering services that increase security and quality of supply, improve energy efficiency and reduce the energy cost. In this paper, a decentralised multi-agent system (MAS) for the coordination of active demand and plug-in electric vehicles is proposed. The load Aggregator aggregates the demand to sellenergy and services by operating a Master Agent and independent Agents. The Master Agent coordinates independent Agents for moving demand to the most convenient hours without degrading power quality. All Agents use a combination of local and global data to perform the optimisation following the Nash's theory on games. The MAS does not require huge bidirectional information flows and is better suited than centralised systems to LV systems.

Modeling and Control of a Hybrid Microgrid by Multi-Agent System

For economic and ecological considerations and following reasons for the depletion of certain oil resources, the coupling of renewable energy sources with the production of hydrogen by electrolysis of water contributes to the development of intelligent networks called Smart Grid (SG). In this context, this paper aims to develop an algorithm for the management and control of a hybrid microgrid by multi-agent system (MAS). This hybrid Microgrid is composed of a photovoltaic source (PVECS), a fuel cell (PEMFC), an electrolyzer (ELZ), and a hydrogen storage system (H2TANK). Oversight of this microgrid is under the control of MAS and ensures optimum performance, taking into account constraints such as intermittent renewable energy, supply / demand balance and the price per kWh.

Multicast dataset synchronization and agent negotiation in distributed manufacturing control systems

Multi agent systems represent an elegant approach for the control architecture of manufacturing systems. Distributed control architectures have the potential to achieve greater flexibility by being capable of local decision making based on real time reasoning. One of the main challenges of these distributed architectures is represented by the capability to synchronize the production data across all execution points in a reliable and consistent fashion. In this context, this paper aims to resolve the problems associated with real time production data synchronization in distributed multi-agent control systems by proposing a common dataset synchronized across all agent entities using multicast network communication. On top of this common dataset approach, an agent negotiation mechanism is proposed that addresses the operation sequencing and resource allocation in decentralized operation model. The pilot implementation is using JADE multi agent platform and JGroups for real time data synchronization and NetLogo for abstract representation of the simulation system. Experimental results gathered from the pilot implementation are discussed.

Multiagent voltage and reactive power control system

Multiagent voltage and reactive power control system

Distributed algebraic connectivity estimation for undirected graphs with upper and lower bounds

The algebraic connectivity of the graph Laplacian plays an essential role in various multi-agent control systems. In many cases a lower bound of this algebraic connectivity is necessary in order to achieve a certain performance. Lately, several methods based on distributed Power Iteration have been proposed for computing the algebraic connectivity of a symmetric Laplacian matrix. However, these methods cannot give any lower bound of the algebraic connectivity and their convergence rates are often unclear. In this paper, we present a distributed algorithm for estimating the algebraic connectivity for undirected graphs with symmetric Laplacian matrices. Our method relies on the distributed computation of the powers of the adjacency matrix and its main interest is that, at each iteration, agents obtain both upper and lower bounds for the true algebraic connectivity. Both bounds successively approach the true algebraic connectivity with the convergence speed no slower than O(1/k).

Multi-Agent Behavioral Control System for the Ore Grinding Process

In this paper, a small multi-agent system (MAS) is proposed based on behavioral approach for the complex grinding processes. Causal association agents were established according to the material balance of grinding processes, and prediction agents and stability control agents were built by adding prediction and control algorithms. The simulation results prove that the system has good stability and anti-interference performance. Multi-agent behavioral control method can be considered as a potential solution to stabilize grinding processes.

Stochastic optimized intelligent controller for smart energy efficient buildings

Abstract Smart buildings are rising concept for energy resource management in the buildings to reduce energy consumption and wastage. Energy management and control system comprises of variety of technologies, which affect human working comfort. The challenging task of the building control is to achieve interior building environment comfort with high-energy efficiency. In this study, multi-agent control system has been developed in combination with stochastic optimization using genetic algorithm (GA). The corresponding simulations of effective management of energy and consumer comfort are presented. The developed control system provides significant improvement in energy consumption and interior environmental comfort in smart buildings.

Application of Q Learning-Based Self-Tuning PID with DRNN in the Strip Flatness and Gauge System

In view of the process of automatic flatness control and automatic gauge control that is a nonlinear system with multi-dimensions, multi-variables, strong coupling and time variation, a novel control method called self-tuning PID with diagonal recurrent neural network (DRNN-PID) based on Q learning is proposed. It is able to coordinate the coupling of flatness control and gauge control agents to get the satisfactory control requirements without decoupling directly and amend output control laws by DRNN-PID adaptively. Decomposition-coordination is utilized to establish a novel multi-agent system for coordination control including flatness agent, gauge agent and Q learning agent. Simulation result demonstrates the validity of our proposed method.

Distributed Key Figure Optimization Approaches for Global Goal Coordination in Multi-agent Systems for Production Control

Performance measurement by means of key performance indicators (KPIs) is a widespread device for communicating qualitative and quantitative business objectives throughout an organization and monitoring their achievement. It seems desirable to also apply this concept to dependably influencing the operational behaviour of distributed agent controlled processes, which have progressively gained in importance in production logistics and control over recent years as compared to centralized control methods. However, planning and scheduling with numeric goal systems at present still poses a challenging task within the field of computational intelligence.In this paper, a framework is presented that enables global coordination of agents in a multi-agent system through user-configurable numeric key performance indicators and associated objectives. A practical approach to distributed control based on periodical mathematical optimization of the defined KPI goal system is described that focuses on how automatic key figure aggregation over changing groups and hierarchies of agents and other business objects can be appropriately managed when proactive modification of these organizational structures, e.g., agents deliberately forming and disbanding groups, is an integral part of the control problem. Five variants of this approach are compared in a case study with a simulated shop floor where incoming manufacturing orders need to be assigned to different machine tools.

406 組み立て工程におけるエージェントベース制御システム(OS5 : 製造の見える化(2))

406 組み立て工程におけるエージェントベース制御システム(OS5 : 製造の見える化(2))

Research on Cooperation Mechanism of Multi-Agent Grinding Control System Based on Blackboard Model

In order to solve the complex industrial control problems in grinding control,we construct multi-agent grinding control system based on the blackboard model in this paper. Through the study on the structure and control mechanism of the blackboard model, we design a framework for blackboard model in multi-agent grinding control system, divide and introduce the function model of the blackboard in detail, establish blackboard control agent to control and coordinate how the individual agent takes use of the blackboard resources centralizedly.Aiming at actual situations of the grinding control,we put forward cooperation mechanism multi-agent based on a combination of tasks sharing and results sharing, establish control decision agent to solve the conflict caused by multi-agent cooperation. The cooperation mechanism multi-agent grinding control system improves the execution ability of the individual agent, avoids the conflict caused by different goals and the competition of resources, ensures the coordinated executive of multiple tasks.

Synchronizing data flows in smart multiagent control systems for the iron and steel industry

The effectiveness of diagnostic and control models and algorithms will depend on the quality of the initial data, which, in turn, is determined by the delay in generating the data and the influence of perturbations. The data must also be synchronized. A method of data synchronization is considered for smart control, diagnostic, and optimization systems, in the case of iron-ore sintering.

Active tracking and pursuit under different levels of occlusion: a two-layer approach

We present an algorithm for real-time, robust, vision-based active tracking and pursuit. The algorithm was designed to overcome problems arising from active vision-based pursuit, such as target occlusion. Our method employs two layers to deal with occlusions of different lengths. The first layer is for short- or medium-term occlusions: those where a known method--such as mean shift combined with a Kalman filter--fails. For this layer we designed the hybrid filter for active pursuit (HAP). HAP utilizes a Kalman filter modified to respond to two different modes of action: one in which the target is positively identified and one in which the target identification is uncertain. For long-term occlusions we use the second layer. This layer is a decision algorithm that follows a learning procedure and is based on game theory-related reinforcement (Cesa-Bianchi and Lugosi, Prediction Learning and Games, 2006). The learning process is based on trial and error and is designed to perform adequately with a small number of samples. The algorithm produces a data structure that can be shared among agents or sent to a central control of a multi-agent system. The learning process is designed so that agents perform tasks according to their skills: an efficient agent will pursue targets while an inefficient agent will search for entering targets. These capacities make this system well suited for embedding in a multi-agent control system.

Research on Grinding Control System Design Method Based on Multi-Agent System Model

This paper proposes a Multi-Agent grinding control system model based on the blackboard model and in accordance with the Multi-Agent Systems modeling method . We hierarchical decompose of the grinding control task, establish a grinding control subsystem model, we establish the specific function of each subsystem model and establish a complete multi-Agent grinding control system structure. We establish a Multi-Agent System based on the blackboard model and multi-Agent Communication and Coordination Mechanism. This paper describes the architecture of the control multi-Agent grinding system and its function results in details and designs a blackboard coordination model, the results shared collaborative mechanisms and Multi-Agent collaboration conflict resolution strategies. This system model reduces the difficulty of the operation of the grinding control, multi-Agent collaboration enhance the system's ability to execute, decision-making consultation mechanisms improves the control precision and accuracy of the system.

A Multi-agent Control System for Intelligent and Sustainable Buildings

A Multi-agent Control System for Intelligent and Sustainable Buildings

Holonic multi-agent system for traffic signals control

Agent-based technologies are rapidly growing as a powerful tool for modelling and developing large-scale distributed systems. Recently, multi-agent systems are largely used for intelligent transportation systems modelling. Traffic signals control is a challenging issue in this area, especially in a large-scale urban network. In a large traffic network, where each agent represents a traffic signals controller, there are many entities interacting with each other and hence it is a complex system. An approach to reduce the complexity of such systems is using organisation-based multi-agent system. In this paper, we use an organisation called holonic multi-agent system (HMAS) to model a large traffic network. A traffic network containing fifty intersections is partitioned into a number of regions and holons are assigned to control each region. The holons are hierarchically arranged in two levels, intersection controller holons in the first level and region controller holons in the second level. We introduce holonic Q-learning to control the signals in both levels. The inter-level interactions between the holons in the two levels contribute to the learning process. Experimental results show that the holonic Q-learning prevents the network to be over-saturated while it causes less average delay time and higher flow rate.

Automated Guided Vehicle System for Work-in-Process Movement

The paper presents an automated guided vehicle transportation subsystem used for work-in-process movement, built at the Production Engineering Institute of Cracow University of Technology. It describes design and operational parameters of built vehicles as well as the principles of integration of AGV control subsystem with the AIM multi-agent manufacturing control system. Furthermore, results of the verification of applied path-finding, anti-collision and anti-deadlock algorithms are included.

Enclosure Active Structure Acoustic Control Based on Multi-Agent System

For enclosure noise reduction, limited discrete distributed control is an ideal choice when it is difficult to understand sound-structure coupling characteristics accurately. An enclosure medel of two flexible plates is established in this paper. Two secondary control channels are constructed by applying distributed force on flexible plate. In the basis of control problem decomposition a multi-agent system (MAS) structure is established and the coordination relationship between agents is regulated. The simulaition resultes show the Active Structure Acoustic Control based on MAS is effective. The MAS control method improves the extension of noise control system and can be used in “plug-and–play” control field to realize noise reduction. MAS is a better effective tool to realize the divide-and-conquer approach in enclosure noise control field.

The learnable evolution model in agent-based delivery optimization

The learnable evolution model is a stochastic optimization method which employs machine learning to guide the optimization process. LEM3, its newest implementation, combines its machine learning mode with other search operators. The presented research concerns its application within a multi-agent system for autonomous control of container on-carriage operations. Specifically, LEM3 is used by transport management agents that act on behalf of the trucks of a forwarding agency for the planning of individual transport schedules.