The efficiency improvement advisor can improve the quality of the emergent solutions created by self-organizing emergent multi-agent systems by identifying recurring tasks. In particular, those recurring tasks that the agents in the self-organizing system do not solve well become valuable knowledge because this knowledge is used to create exception rules for the appropriate agents that improve their task-fulfilling behavior. In this paper, we present an extension to the advisor that allows it to use certain knowledge about future tasks in addition to the (somewhat uncertain) knowledge gained from the system history. By now creating groups of exception rules for each expected task, the self-organizing emergent system can achieve near optimal solutions for static problem instances and good solutions for a range of expected tasks, while still being able to deal with dynamic (and unpredicted) tasks, as shown by experiments in a pickup and delivery transportation scenario.
Generated: 9/12/18