Agent-based Optimization Approach Research Articles

Decentralized heating grid operation: A comparison of centralized and agent-based optimization

Abstract Moving towards a sustainable heat supply calls for decentralized and smart heating grid solutions. One promising concept is the decentralized feed-in by consumers equipped with their own small production units (prosumers). Prosumers can provide an added value regarding security of supply, emission reduction and economic welfare, but in order to achieve this, in addition to advanced hydraulic control strategies also superordinate control strategies and appropriate market models become crucial. In this article we study methods to find a global optimum for the local energy community or at least an acceptable approximation to it. In contrast to standard centralized control approaches, based either on expert rules or mixed integer linear optimization, we adopt an agent-based, decentralized approach that allows for incorporation of nonlinear phenomena. While studied here in small-scale systems, this approach is particularly attractive for larger systems, since with an increasing number of interacting units, the optimization problem becomes more complex and the computational effort for centralized approaches increases dramatically. The agent-based optimization approach is compared to centralized optimization of the same prosumer-based setting as well as to a purely central setup. The comparison is based on the quality of the optimization solution, the computational effort and the scalability. For the comparison of these three approaches, three different scenarios have been set up and analysed for four seasons. In this analysis, no approach has emerged as clearly superior to the others; thus each of them is justified in certain situations.

An Agent Based Optimization Approach to Manufacturing Process Planning

Many optimization problems, such as manufacturing process planning optimization, are difficult problems due to the large number of potential configurations (process sequences) and associated (process) parameters. In addition, the search space is highly discontinuous and multi-modal. This paper introduces an agent based optimization algorithm that combines stochastic optimization techniques with knowledge based search. The motivation is that such a merging takes advantage of the benefits of stochastic optimization and accelerates the search process using domain knowledge. The result of applying this algorithm to computerized manufacturing process models is presented.