Towards an Adaptive Approach for Distributed Resource Allocation in a Multi-agent System for Solving Dynamic Vehicle Routing Problems

Introduction. The existing problem of continuous planning in transportation logistics requires the solving of dynamic Vehicle Routing Problems (dynamic VRPs) which is an NP-complete optimization problem. The task of continuous planning assumes the existence of individual commit times for orders to be released for execution with specific commitment strategies and is similar to the problem of maintaining a guaranteed response time in real-time systems that, in a dynamic environment, applies additional restrictions on planning algorithms. This paper describes the developed multi-agent platform for solving the dynamic multi-vehicle pickup and delivery problem with soft time windows (dynamic m-PDPSTW) that supports goaldriven behavior of autonomous agents with a multi-objective decision-making model. Further research on the design of adaptive mechanisms for run-time feedback-directed adjustment of scheduling algorithms through learning and experience of applied decision options is outlined. The market-based method for distributed dynamic resource allocation. The authors’ multi-agent approach to continuous planning in complex uncertain environments implements the concept of the dynamic distributed resource allocation (Modi et al. 2002) whereby decision making is carried out by means of asynchronous quasi-parallel processes of negotiation between the agents of resources and the agents of orders which are working distributively in a virtual marketplace over all the agents of the complex shared ontology object. Goal-driven behavior of autonomous agents is supported by the developed multiobjective decision-making model based on individual goals, criteria, and preference functions defined in the virtual marketplace ontology and methods of agent self-appraisal and decision-making strategies (priority of criteria, value superposition, and homeostasis). In the implemented framework for agent knowledge representation, based on Prof. Marvin Minsky’s concept of frames, each autonomous agent in the virtual marketplace has its individual knowledge, agent memory with a dynamic decision-making table, experience of applied deci-