Tjorben Bogon

An Agent Based Parallel Particle Swarm Optimization - APPSO

As the complexity of optimization problems increases, new scalable architectures for variable problem complexitys are needed. In this paper we introduce an agent based framework for distributing and managing a particle swarm on several interconnected computers. Agent Based Parallel Particle Swarm Optimization (APPSO) accelerates the optimization through parallelization and strategical niching, offers dynamic scalability at runtime, and fault tolerance. Due to its load balancing feature APPSO runs efficient on heterogeneuos system. Two experiment series on a prototype implementation demonstrate the performance gain achieved by APPSO.

Towards assisted input and output data analysis in manufacturing simulation: the EDASim approach

Discrete-event simulation has been established as an important methodology in various domains. In particular in the automotive industry, simulation is used to plan, control, and monitor processes including the flow of material and information. Procedure models help to perform simulation studies in a structured way and tools for data preparation or statistical analysis provide assistance in some phases of simulation studies. However, there is no comprehensive data assistance following all phases of such procedure models. In this article, a new approach combining assistance functionalities for input and output data analysis is presented. The developed tool - EDASim - focuses on supporting the user in selection, validation, and preparation of input data as well as to assist the analysis of output data. The proposed methods have been implemented and initial evaluations of the concepts have led to promising feedback from practitioners.

A knowledge-based approach to automated simulation model adaptation

Simulation has become a widely accepted technology for analyzing or planning systems in various domains. In production logistics, for instance, many companies use simulation to evaluate scenarios before actually the construction or modifications of the production hall or processes are performed in order to get insights about the performance of planned configurations. In this paper, we propose an approach to knowledge-based adaptation of simulation models. The vision of this work is to go one step beyond parameter optimization, namely to provide means for automated structural changes in simulation models, and thus for the generation of simulation model variants. For a first evaluation of our approach, we introduce a system consisting of a simulation control as well as a model adaptation module with a set of adaptation operations. Our implementation is coupled to the simulation system Plant Simulation in order to perform simulation runs. For illustration we apply our system to a test scenario and present first results.

Automatic parameter configuration of particle swarm optimization by classification of function features

Metaheuristics in stochastic local search are used in numerical optimization problems in high-dimensional spaces. A characteristic of these metaheuristics is the configuration of the parameters. These parameters are essential for the optimization behavior but depend on the objective function. In this paper we introduce a new approach to automatic parameter configuration of Particle Swarm Optimization (PSO) by classifying features of the objective function. This classification utilizes a decision tree that is trained by 32 different function features. These features result from the characteristics of the underlying function landscape and of the PSO behavior. An efficient set of parameters influences the optimization in speed and performance. In literature standard configurations are introduced for different types of metaheuristics which perform a not optimal but an adequate optimization behavior for most objective functions. PSO is an example for the parameter configuration problem [2]. The swarm behavior depends mainly on the chosen parameter and leads to solutions of different quality, i.e. bad parameter sets can lead to a disadvantageous balance between exploitation and exploration. One problem by choosing the right parameter without knowledge about the objective function is to describe the characteristics of the function which are comparable to another function.

Teaching Distributed Artificial Intelligence with RoboRally

Teaching Artificial Intelligence (AI) or multi-agent systems is a challenging task as algorithms are in question which are advantageous in highly complex and dynamic environments. Explaining multi-agent systems (MAS) in lectures requires interactive approaches accompanied by exercises. The key challenge in using practical exercises within lectures on MAS is to establish an environment for testing which is extremely time consuming. It is not reasonable that students do this work as they do have not enough time focussing on the important aspects. In this paper, we introduce a system which supports experimenting with AI and Distributed Artificial Intelligence (DAI) algorithms concurrently to the lecture. Our system is based on a board game called RoboRally. Different issues from the field of AI and DAI can be implemented and tested in a kind of challenge.

Convergence Classification and Replication Prediction for Simulation Studies

Providing assistance systems for simulation studies can support the user by performing monotonous tasks and keeping track of relevant results. In this paper we present approaches to estimate, if – and when – statistically significant results are expected for certain investigations. This information can be used to control simulation runs or to provide information to the user for interaction. The first approach is used to classify if significance is expected to occur for given samples and the second approach estimates the needed replications until significance is expected be reached. For an initial evaluation of the approaches, experiments are performed on samples drawn from normal distributions.

Visualizing the Impact of Probability Distributions on Particle Swarm Optimization

In this paper we present a simulation tool for the visualization of the impact of different probability distributions on Particle Swarm Optimization (PSO). PSO is influenced by a high number of random values in order to simulate a more nature like behaviour. Based on these random numbers the optimization process may vary. Usually the uniform distribution is chosen but regarding certain underlying fitness functions this may not the best choice. To test the influence of different probability distributions on PSO and to compare the different approaches, the presented simulation system consist of a simple user interface and allows the integration of own distribution formulas in order to test their impact on PSO.

Setting Up Particle Swarm Optimization by Decision Tree Learning Out of Function Features

This work describes an approach for the computation of function features out of optimization functions to train a decision tree. This decision tree is used to identify adequate parameter settings for Particle Swarm Optimization (PSO). The function features describe different characteristics of the fitness landscape of the underlying function. We distinguish between three types of features: The first type provides a short overview of the whole search space, the second describes a more detailed view on a specific range of the search space and the remaining features test an artificial PSO behavior on the function. With these features it is possible to classify fitness functions and to identify a parameter set which leads to an equal or better optimization process compared to the standard parameter set for Particle Swarm Optimization.

A Decentralized Architecture for Distributed Neighborhood Based Search

We present a decentralized self-X architecture for distributed neighborhood based search problems using an overlay network based on random graphs. This approach provides a scalable and robust architecture with low requirements for bandwidth and computational power as well as an adequate neighborhood topology, e.g. for several instances of parallel local search and distributed learning. Together with an adapted load balancing schema our architecture is self-organizing, self-healing and self-optimizing.