Year: 2024
Type: conference-paper
Abstract: Part of Industry 4.0 is building computer systems by combining artificial intelligence with robotics. Such computer systems play an important role in the planning of cargo transportation in supply chain management. One of the approaches to building such computer systems is the use of multi-agent systems. The aim of the work is to create a methodology for constructing proactive agents based on reinforcement learning to solve the problem of optimal planning of cargo transportation. To solve the problem of insufficient efficiency of computer agents, the existing methods of statistical and machine learning were investigated. To date, the most efficient approaches to creating proactive agents are reinforcement learning approaches. The formalization of the functioning of proactive agents is performed. As a part of creating a model for the functioning of proactive agents based on reinforcement learning, a procedure for generating a quasi-optimal action plan is proposed that models the planning function of a proactive agent, which speeds up the decision-making process. Multi-agent reinforcement learning methods are proposed, which are close to random search at the initial iterations, and close to directed search at the final iterations. This is ensured by the use of dynamic parameters and allows the increase in the learning rate by approximately 10 times while maintaining the mean squared error of the method.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7166301227