Network decision support based on distributed coevolutionary algorithms

摘要

In this thesis, efficient and scalable algorithms for distributed, network-based decision-making are considered. In a network-distributed environment internode communications are a primary factor in system performance, and the decision-making algorithms that function in this environment must demonstrate superior performance while simultaneously economizing internode communications.;First, a class of distributed decision problems is selected. This class of decision problems is the integrated design, supplier, and manufacturing planning for modular products, where suppliers and manufacturing resources are distributed. A formal model for this class of distributed decision problems is developed as a set of coupled nonlinear assignment problems. This model, called the design-supplier---manufacturing planning decision problem , is an example of a complex discrete optimization problem. The nonlinearities and coupling inherent in the problem class complicate application of exact optimization algorithms. Evolutionary algorithms however, are highly adaptable, place minimal restrictions on problem structure, and provide efficient and reliable solutions. A novel coevolutionary algorithm, based on distributed evolutionary algorithm components and software agents, is proposed as a network-efficient strategy for concurrent, cooperative exploration of a highly coupled space of design, supplier, and manufacturing decisions.;Next, a theoretical foundation for this class of coevolutionary algorithms is developed using techniques from stochastic process theory and mathematical analysis. In this framework, the distributed computation is described in terms of construction and evolution of sampling distributions over the feasible space. Convergence and convergence rate analyses are pursued for certain basic classes of objective functions, and analytical and simulation techniques are used to evaluate the network-based performance of the algorithms in this class.;As a case study in distributed, network-based decision-making, we present an implementation and detailed evaluation of the coevolutionary decision-making framework suited to distributed network-enabled design and manufacturing organizations. This implementation, the Coevolutionary Virtual Design Environment (CVDE), utilizes distributed evolutionary agents and mobile agents as principal entities that generate and execute queries and transactions among distributed applications and databases to support a global optimization of design, supplier, and manufacturing planning decisions. In this framework, an electronic interchange of design, supplier, and manufacturing information facilitates a concurrent, cooperative, network-efficient evolutionary search for promising planning alternatives.