Resource Allocation in Decentralized, Self-organized, Multi-agent Industrial Systems Using Deep Deterministic Policy Gradient

The fourth industrial revolution has changed how the manufacturing process is organized in the industrial setting. A flexible and dynamic manufacturing environment is vital to this transformation. This includes self-organized, decentralized, multi-agent industrial systems. This work introduces a novel approach for resource allocation in such systems based on current deep reinforcement learning algorithms. A virtual environment is developed in which algorithms can learn and be tested according to global performance measurements. Agents learn their true cost function by interacting with the environment, thus improving their cost estimation accuracy and precision. This, in turn, leads to an increase in the system's global performance by allowing the resources to be distributed more efficiently. When compared to existing distance-based estimation algorithms, it is shown that this approach leads to improved cost estimation accuracy.