Haut PDF An adaptive multi-agent system for self-organizing continuous optimization

An adaptive multi-agent system for self-organizing continuous optimization

An adaptive multi-agent system for self-organizing continuous optimization

would take its decision. However, it is clear that one of the message will take more time to reach a 1 than the other, as it must transit by two intermediates. Why would the agent a 1 wait for this late message, without even knowing its existence, before acting? Many of the previous mechanisms we proposed rely on a timely information delivery, for example for the agents to be able to detect some NCSs. Based on this, it results that, without an adequate “synchronization” mechanism, many of them would fail to work correctly. On the other hand, having a fully synchronized system, where all the agents would make lengthy checks to ensure the complete propagation of all messages to all recipients, would be undesirable. Indeed, complex continuous optimization problems create big but loosely connected graphs, for which such full synchronized mechanism would be both costly and inefficient. Moreover, if we take in account the dynamic aspects of the problem (having a designer which can add or remove agent at any moment during solving), obtaining such synchronization guarantee can be nearly impossible.
En savoir plus

207 En savoir plus

Principle and Evaluation of a Self-Adaptive Multi-Agent System for State Estimation of Electrical Distribution Network

Principle and Evaluation of a Self-Adaptive Multi-Agent System for State Estimation of Electrical Distribution Network

II. S TATE E STIMATION IN D ISTRIBUTION S YSTEM A distribution system is the part of an electrical network which distributes electricity from transmission systems to consumers. These systems are made of nodes, also called buses, which are linked with other nodes through lines which may have various admittances. In order to have a power !ow in these systems, there must be at least one producer (generator) and at least one consumer (load) each one connected to a bus. For each node to which a producer or a consumer is connected, there is a power sensor (or at least a load-pattern). In addition, some other voltage sensors can be associated with other buses. These sensors provide noisy data about the state of the network.
En savoir plus

7 En savoir plus

A Self-adaptive Agent-based System for Cloud Platforms

A Self-adaptive Agent-based System for Cloud Platforms

Agents are networked software entities that can do specific tasks on behalf of a user and have a degree of intelligence that allows them to perform parts of their tasks autonomously and to inter- act with their environment in a successfully way. Agents are characterized by important features such as autonomy, sociality, rationality, responsiveness, proactiveness, and mobility [ 12 ]. Some studies have been proposed to design Cloud platforms based on multi-agent systems [ 1 ], [ 6 ], [ 14 ]. In [ 6 ], it is proposed the integration of agent-based system and Cloud computing for smart objects. This approach is suitable to effectively model Cooperative Smart Objects (CSO). In particular, a CSO is a smart object able to sense, store, and interpret infor- mation. In [ 14 ], it is presented a mechanism to provide dynamic load balancing for Clouds based on autonomous agents. The proposed mechanism is based on Ant mobile agents and whenever the load of a Virtual Machine (VM) reaches a threshold value, it initiates a search for a candidate VM from other datacenters, reducing in this way the alloca- tion time. This work only considers the workload of VMs and does not consider the physical machines. In [ 1 ], a multi-agent system is proposed to manage the Cloud resources, while taking into account the customers QoS requirements. In this work a VM migration occurs when its hosting physical machine is facing an overloading or under loading problem. This approach is the most similar to ours, however, our approach is autonomic self-adaptive, allowing controlling the user requirements depending on the system and agents states.
En savoir plus

8 En savoir plus

Self-adaptive multi-agent systems for aided decision-making : an application to maritime surveillance

Self-adaptive multi-agent systems for aided decision-making : an application to maritime surveillance

to associate a suspicious level to each ship and the component is thus able to detect the anomalies in the ship behaviours. Finally alerts are sent to the operators when it is found to be necessary. Ultimately, the operators can compare the alerts triggered by BEAN with what they analyse according to their knowledge and experience. If they agree with the automatic alert, they transmit it to the concerned authorities in order to identify and answer the situation (thread to tackle, need for assistance,. . . ). On the contrary, if the operators do not agree with the alert from BEAN, they can send a feedback aiming at the correction of the system. A feedback is also sent when the operators spot an alert that has not been triggered by BEAN. I2C is a European Project aiming at the surveillance of maritime areas to identify and answer threats and issues. This project integrates acquisition tools as well as anomaly detection mechanisms. In the BEAN component, we now focus on the role of the multi- agent system and how the different parts of MAS4AT can be used for it.
En savoir plus

177 En savoir plus

S-DLCAM: A Self-Design and Learning Cooperative Agent Model for Adaptive Multi-Agent Systems

S-DLCAM: A Self-Design and Learning Cooperative Agent Model for Adaptive Multi-Agent Systems

Keywords-Adaptive Multi-Agent Systems; Cooperative Agent; Self-Design and Learning Cooperative Agent Model. I. I NTRODUCTION For Adaptive Multi-Agent System (AMAS) [1], the de- velopment of adaptation implies the need to focus on the agent level. This is to give the agent the means to decide autonomously to change its relationships with other agents in order to move toward a cooperative organization. Thus, depending on the interactions that the AMAS has with its environment, the organization between its agents emerges. Building such self-organized systems is not a trivial task. In this paper, we propose a new cooperative agent model based on Self-Design and learning mechanisms developed from the agent model associated with the AMAS theory [1], [2], [3]. We take in account the following important works: [4] (in which Capera et al. present a model based upon a sort of extended automata product, dedicated to multi-agent systems) and [5] (in which Russel and Norvig present how an agent can find a sequence of actions that achieves its goals, when no single action will do). Indeed, we consider that the Self-Design and Learning Cooperative Agent (S-DLCA) life cycle goes through two levels: the preliminary level (PL) (nominal and cooperative behaviour) given by the designer and the heigh level (HL) which is responsible of the detection and correction of the Non Cooperative Situations (NCS) that the agent may encounter during its life. This model was developed under SeSAm (http://www.simsesam.de/) and it can be used by any AMAS designer in order to help him in the detection and correction of the NCS using the new ADELFE methodology extensions [6].
En savoir plus

4 En savoir plus

Rapid and adaptative mission planner for multi-satellite missions using a self-adaptative multi-agent system

Rapid and adaptative mission planner for multi-satellite missions using a self-adaptative multi-agent system

This paper contributes to this challenge with a new way to plan on-ground the mission of satellites: the ATLAS planning system (Adaptive saTellites pLanning for dynAmic earth obServation). ATLAS is an Adaptive Multi-Agent System, designed to plan missions of constellations of Earth observation satellites. The proposed system brings a major contribution: it is an open and continuous planning system. It has the capability to handle in real-time changes of constraints and/or new request arrivals. ATLAS possesses self-adaptation mechanism in order to locally self-adapt itself according to the dynamic arrival of requests to plan. Thus, ATLAS can dynamically reorganize the mission plan in order to propose a better one (integrating the changes). Because changes are made locally, the whole plan is not challenged and the new plan is provided in a reasonable time. ATLAS can also be stopped at any time and provides a good mission plan. Indeed, the system globally makes the mission plan by local interactions. To enable this capability for real-time adaptation, we use the Adaptive Multi-Agent Systems theory (AMAS). Such systems naturally provide self-adaptation capabilities required to solve this kind of problem. To design our system, we rely on the Adaptive Multi-Agent System For Optimization agent model, providing some design patterns to solve optimization problems using AMAS. In this model, agents are designed as close as possible to the natural description of the problem entities.
En savoir plus

10 En savoir plus

Principle and Evaluation of a Self-Adaptive Multi-Agent System for State Estimation of Electrical Distribution Network

Principle and Evaluation of a Self-Adaptive Multi-Agent System for State Estimation of Electrical Distribution Network

II. STATE ESTIMATION IN DISTRIBUTION SYSTEM A distribution system is the part of an electrical network which distributes electricity from transmission systems to consumers. These systems are made of nodes, also called buses, which are linked with other nodes through lines which may have various admittances. In order to have a power !ow in these systems, there must be at least one producer (generator) and at least one consumer (load) each one connected to a bus. For each node to which a producer or a consumer is connected, there is a power sensor (or at least a load-pattern). In addition, some other voltage sensors can be associated with other buses. These sensors provide noisy data about the state of the network.
En savoir plus

8 En savoir plus

Model-free Optimization of an Engine Control Unit thanks to Self-Adaptive Multi-Agent Systems

Model-free Optimization of an Engine Control Unit thanks to Self-Adaptive Multi-Agent Systems

{boes, migeon, glize}@irit.fr, erwan.salvy@aboard-eng.com Keywords: Control ; Multi-Agent Systems ; Self-Organization ; Auto-Calibration ; Automotive Abstract: Controlling complex systems, such as combustion engines, imposes to deal with high dynamics, non-linearity and multiple interdependencies. To handle these difficulties we can either build analytic models of the process to control, or enable the controller to learn how the process behaves. Tuning an engine control unit (ECU) is a complex task that demands several months of work. It requires a lot of tests, as the optimization problem is non-linear. Efforts are made by researchers and engineers to improve the development methods, and find quicker ways to perform the calibration. Adaptive Multi-Agent Systems (AMAS) are able to learn and adapt themselves to their environment thanks to the cooperative self-organization of their agents. A change in the organization of the agents results in a change of the emergent function. Thus we assume that AMAS are a good alternative for complex systems control. In this paper, we describe a multi-agent control system that was used to perform the automatic calibration of an ECU. Indeed, the problem of calibration is very similar to the one of control: finding the adequate values for a system to perform optimally.
En savoir plus

11 En savoir plus

Multi-criteria and multi-objective dynamic planning by self-adaptive multi-agent system, application to earth observation satellite constellations

Multi-criteria and multi-objective dynamic planning by self-adaptive multi-agent system, application to earth observation satellite constellations

exists) the optimal solution. Still, as they explore the whole search space, they are too slow and cannot be applied to solve complex problems. Approximate methods find a solution using a heuristic to cross the space search. An important drawback of all of those methods is that they cannot manage dynamic, what is one of the characteristics of complex problems. To overcome this limitation, new methods are currently being studied. Those methods are mainly decentralized. Thus, the exploration of the search space is parallelized and dynamics can be managed. Among those new resolution methods, the SMAC team (Systèmes Multi Agent Coopératifs for Cooperative Multi-Agent System) proposed the Adaptive Multi-Agent Systems (AMAS) [Gleizes, 2012]. In an AMAS, cooperative agents cooperate together to make emerge a solution. Thanks to self-organization mechanism, AMAS naturally provide the flexibility and the robustness necessary to solve complex problems. AMAS are applied with success in many fields, like complex problem solving [Capera et al., 2003]. A generic model of cooperative agents, AMAS4Opt [Kaddoum, 2008], has been proposed and validated on complex problems like manufacturing control. Even if first results are quite good, several lacks exist and must be implemented to apply AMAS4Opt model on new and various complex problems.
En savoir plus

169 En savoir plus

Study of conditions for the emergence of cellular communication using self-adaptive multi-agent systems

Study of conditions for the emergence of cellular communication using self-adaptive multi-agent systems

If genes are not the programming language of the cell, then is there any and what is it? A cell, in a multicellular organism, usually does not act on its own accord. To start a new process, produce a new metabolite, differentiate or divide, it needs cues from its environment or direct orders. What are these cues? They can be of several types: Physicochemical properties of the environment like pH, ions concentration, temperature, specific molecules recognized by cellular receptors, gradients of chemicals (protein or small molecules), or physical pressure. Direct orders can come from neighbor cells through signaling molecules or from far away organs through hormones (Figure 2-3). It is actually "quite easy" to make cells divide in a Petri dish: Meet all their resource requirements and then add a growth factor protein at a certain concentration, and they will soon multiply, obeying the request. When an entity responds to stimuli by changing its behavior without affecting its structure, can this be called programming? If the answer is yes, then molecules used in cell-cell communication would be the true programming language of cells. In this context, DNA would represent the blueprint of the machinery that can interpret and execute this programming language. But at this point the cell-computer metaphor is probably reaching its limits since a computer is not a complex system and each of its components has a clearly defined role. In a cell, function boundaries are sometimes blurred and roles are mixed or contextual. For example, DNA stores information but it can also have a catalytic activity and directly transform other molecules (like RNA). An enzyme has a catalytic role but when chemically modified becomes an element of information that can regulate other processes.
En savoir plus

172 En savoir plus

An adaptive multi-agent system for the distribution of intelligence in electrical distribution networks: state estimation

An adaptive multi-agent system for the distribution of intelligence in electrical distribution networks: state estimation

In the electrotechnical field, the contributions are more obvious. Initially, the problem addressed was to find a way to “distribute intelligence in medium voltage and low voltage networks”. This is part of the Smart Grid concept. The complexity of the problem and the amount of characteristics that should be considered make it impossible to be treated as one unique three-years study. The general project aims at providing an automatic system able to take into account the complexity and a lot of requirements of the Smart Grid such as: scalability, self-adaptation, ... My thesis of 3 years is a contribution to this general project. It focuses on the creation of a framework adapted to some advanced functions of the Smart Grid and allowing to develop the characteristics of the Smart Grid (see 2.3 page 19). Moreover, it has been showed that this fits with the resolution of two major problems in electricity: the Load Flow and the State Estimation.
En savoir plus

202 En savoir plus

Adaptive Autonomous Navigation using Reactive Multi-agent System for Control Law Merging

Adaptive Autonomous Navigation using Reactive Multi-agent System for Control Law Merging

3 Institut Pascal, Blaise Pascal University, Clermont-Ferrand, France. name.surname@univ-bpclermont.fr Abstract This paper deals with intelligent autonomous navigation of a vehicle in cluttered environment. We present a control architecture for safe and smooth navigation of a Unmanned Ground Vehicles (UGV). This control architecture is designed to allow the use of a single control law for different vehicle contexts (attraction to the target, obstacle avoidance, etc.) [4]. The reactive obstacle avoidance strategy is based on the limit-cycle approach [2]. To manage the interaction between the controllers according to the context, the multi-agent system is proposed. Multi-agent systems are an efficient approach for problem solving and decision making. They can be applied to a wide range of applications thanks to their intrinsic properties such as self-organization/emergent phenomena. Merging approach between control laws is based on their properties to adapt the control to the environment. Different simulations on cluttered environments show the performance and the efficiency of our proposal, to obtain fully reactive and safe control strategy, for the navigation of a UGV.
En savoir plus

11 En savoir plus

Model-free Optimization of an Engine Control Unit thanks to Self-Adaptive Multi-Agent Systems

Model-free Optimization of an Engine Control Unit thanks to Self-Adaptive Multi-Agent Systems

{boes, migeon, glize}@irit.fr, erwan.salvy@aboard-eng.com Keywords: Control ; Multi-Agent Systems ; Self-Organization ; Auto-Calibration ; Automotive Abstract: Controlling complex systems, such as combustion engines, imposes to deal with high dynamics, non-linearity and multiple interdependencies. To handle these difficulties we can either build analytic models of the process to control, or enable the controller to learn how the process behaves. Tuning an engine control unit (ECU) is a complex task that demands several months of work. It requires a lot of tests, as the optimization problem is non-linear. Efforts are made by researchers and engineers to improve the development methods, and find quicker ways to perform the calibration. Adaptive Multi-Agent Systems (AMAS) are able to learn and adapt themselves to their environment thanks to the cooperative self-organization of their agents. A change in the organization of the agents results in a change of the emergent function. Thus we assume that AMAS are a good alternative for complex systems control. In this paper, we describe a multi-agent control system that was used to perform the automatic calibration of an ECU. Indeed, the problem of calibration is very similar to the one of control: finding the adequate values for a system to perform optimally.
En savoir plus

12 En savoir plus

Self Adaptive Support Vector Machine:  A Multi-Agent Optimization Perspective

Self Adaptive Support Vector Machine: A Multi-Agent Optimization Perspective

efficient on very complex learning tasks as it is known that MKL can outperform single kernel techniques. Clearly, AMAS provides a natural formalism to express learning problems. Further research and implementations should confirm that this framework could be an alternative solution for building learning systems that require high level of adaptability especially for complex learning tasks where distribution of problem solving among agents is necessary. Furthermore, high degree of parallelization is possible since most of the time all elementary agent decisions are independent and asynchronized. AMAS are therefore suited to high dimensional problems arising in many applications such as in genomic data analysis in biology or in webdata analysis from the internet environment. Even though we did not detail the idea of online learning with data being fed to the system one at a time, it is also important to mention that all concepts that were discussed can be extended to such situations as long as the AMAS system implementation allows online gener- ation and integration of new agents.
En savoir plus

16 En savoir plus

Multi-satellite mission planning using a self-adaptive multi-agent system

Multi-satellite mission planning using a self-adaptive multi-agent system

Exact Methods guarantee to find the optimal solution, if it exists. Dynamic Programming is used to decompose the problem into simpler sub-problems. The difficulty of such algorithms comes from the ability to recursively cut the initial problem. As shown in [16], when applying the algorithm linearly, the execution will quickly use most of the memory resources making the approach inapplicable in case of complex linked requests, as it is the case of stereoscopic imaging. It requires several acquisitions of the same mesh with an angular distance which creates links between sub-problems making the decomposition impossible. [18] emit the same criticism of algorithms built on Branch and Bound. Indeed, even if all the solutions are not explored as the algorithm uses the properties of the problem to avoid some branches, the use of such methods is extremely expensive in exploration time and required memory space. These methods are therefore not suitable for problems with very large search spaces, like the problem treated in this paper.
En savoir plus

11 En savoir plus

An introduction to continuous optimization for imaging

An introduction to continuous optimization for imaging

(a) subset of the well-classified digits (b) the 221 (out of 10 000) wrongly classified digits Figure 7.21. MNIST classification results. negative). If everything were perfect, a digit would be classified nine times as its true label, hence it is natural to consider that the label that gets the maximal number of ‘votes’ is the expected one. 24 This very elementary ap- proach leads to an error of 2.21% on the 10 000 test digits of the database, which is much worse than the best results reached so far but quite reasonable for such a simple approach. Let us observe that for each failed classifica- tion, the second vote was correct except for 70 digits, that is, 0.7% of the base: hence it is not surprising that more elaborate representations or clas- sification techniques can reach classification error rates of this order: 0.49% for SVM methods based on ‘scattering’ networks (Bruna and Mallat 2013) and 0.23% for the most recent CNN-based methods (Ciresan, Meier and Schmidhuber 2012). The failed classifications are shown in Figure 7.21, as well as some of the well-recognized digits.
En savoir plus

160 En savoir plus

An Energy-Aware Protocol for Self-Organizing Heterogeneous LTE Systems

An Energy-Aware Protocol for Self-Organizing Heterogeneous LTE Systems

On the other hand, techniques for improving cellular radio energy efficiency have recently attracted much at- tention. Auer et al [13] has investigated the amount of power consumptions for various types of base stations. Mclaughlin et al [14] has discussed various techniques for improving energy efficiency. Conte et al [15] has proposed to turn base stations to sleep mode when the network traffic is small to save energy. Son et al [16], Zhou et al [17], and Gong, Zhou, and Niu [18] have proposed various policies of allocating clients so that clients are mostly allocated to a few base stations. As a result, many base stations that do not have any clients can be turned to sleep mode to save energy. However, these studies require the knowledge of traffic of each client, and cannot be applied to scenarios where clients’ traffic is elastic. Chen et al [19] has studied the trade-off between spectrum efficiency and energy efficiency. Miao et al [20] and Li et al [21] have provided extensive surveys on energy- efficient wireless communications. However, they do not consider the interference and interactions between base
En savoir plus

12 En savoir plus

Cooperation in Adaptive Multi-Agent Systems through System of Systems modeling

Cooperation in Adaptive Multi-Agent Systems through System of Systems modeling

The AMAS approach - This approach is relevant to design adaptive multi-agent systems. It enables to solve complex problems that can be incompletely specified and for which an a priori known algorithmic solution does not exist. It considers the system as composed of parts (i.e. agents) and focuses on the local agent behavior for making them adaptive (to their local environment) while ensuring that the collective behavior emerging from interactions between agents is the one expected; in that case the system is said “functionally adequate”. Each agent must have a local cooperative behavior for this purpose [ 9 ] [ 12 ]. The AMAS approach incorporates the notion of criticality, defined as the “distance between the current situation and the local purpose of the agent” [ 15 ]. Thus, “the further the agent is from its goal, the more critical it considers its current situation”. Considering this notion, an agent is cooperative if it acts in order to help the most critical agent of its neighborhood. Thus, all agents within an AMAS try to continuously reduce the criticality of the most critical agent (possibly itself), while avoiding another agent becoming even more critical. If an agent is not able to help the most critical agent of its neighborhood, it may help other less critical agents. Thus, doing so, it hopes these agents will be able to help the most critical agent thanks to the reduction of their own criticality.
En savoir plus

14 En savoir plus

Towards a self-adaptive parameterization for aerodynamic shape optimization

Towards a self-adaptive parameterization for aerodynamic shape optimization

The shape optimization procedure including the adaption method is applied to this problem[3]. Here, α and ω(t) are chosen in such a way that the solution corresponds to a circular arc (α = 2 and ω(t) = 1), for which the theorical minimum value is 2Π. The arc is successively parameterized by B´ezier curves of increasing degrees. The cost function values obtained with respect to the degree are depicted in figure (3).

8 En savoir plus

Continuous Multi-eEchelon Inventory Optimization

Continuous Multi-eEchelon Inventory Optimization

Requirements for the Degree of Master of Applied Science in Supply Chain Management ABSTRACT Global supply chains are becoming increasingly complex systems that drive significant investments in inventory throughout the network. Our sponsor for this project uses a multi-echelon inventory optimization (MEIO) model to manage safety stock inventory across its network. The MEIO model helps them optimize inventory based on upstream and downstream supply chain performance but it does not guarantee year over year reductions in inventory levels that the company desires. To address this issue, we studied how the company can better utilize MEIO to systematically reduce its inventories over time and created a methodology that can be employed by other companies also. We applied the methodology on two products that are presented as case studies. For the chosen products, we found that variation in supply lead time is the primary reason for high MEIO safety stock values. We further identified the underlying cause of variation and provided recommendations to reduce variation in lead time in each case study. This research creates a framework that companies can use to systematically minimize MEIO safety stocks and presents case studies that apply this framework to minimize variation in supply lead time on two products and their corresponding MEIO safety stocks.
En savoir plus

33 En savoir plus

Show all 10000 documents...

Sujets connexes