The increasing use of automated guided vehicle (AGV) systems in intralogistics places growing demands on on-time delivery, throughput, energy efficiency, and robust materialflow control. Particularly in large, dynamically operated spare-parts warehouses, the joint coordination of transport-order dispatching and vehicle charging constitutes a complex optimization problem that conventional heuristics and fixed threshold-based strategies cannot adequately address. Building on a simulation model of an industrial spare-parts warehouse that was developed in Siemens Plant Simulation and subjected to plausibility checks, this dissertation combines several artificial intelligence (AI)-based approaches in an integrated framework. Its main components are (i) a deep reinforcement learning dispatching agent that models the dispatching problem as a Markov decision process and learns state-dependent transport-order assignments using a deep Q-network (DQN), (ii) a supervised-learning charging agent based on multilayer perceptrons, support vector machines, and random forests that approximates charging decisions generated by a mixed-integer programming (MIP) model for charging and battery management, and (iii) a fusion approach in which dispatching and charging decisions are coupled within an integrated agent system. The approaches are evaluated in Siemens Plant Simulation using a realistic test scenario with varying demand profiles, including morning, midday, and end-of-day peak periods as well as high-load scenarios. The main key performance indicators are the number and distribution of delivery due-date violations. The results show that the DQN-based dispatching agent clearly outperforms conventional distance-based and time-window-based dispatching rules, particularly under high system utilization. The charging agent achieves decision quality close to the optimization-based reference while requiring substantially less computation time than directly solving the MIP model. Compared with a simple charging heuristic, it also improves system stability and vehicle availability. The fusion of the two agents combines the advantages of both component strategies. Depending on the fusion variant, both the total number of delivery due-date violations and the frequency of extreme delays can be reduced without compromising the robustness of the overall system. This dissertation contributes to the systematic integration of deep reinforcement learning and supervised learning methods into the operational control of AGV fleets. It demonstrates how a simulation-based testbed can be used to design, train, and evaluate AI-based dispatching, charging, and fusion strategies. The findings provide a basis for further research, including multi-agent control, integration with energy-sector demand-side management concepts, and the implementation of hardware-in-the-loop and real-time pilot applications in intralogistics.     «

The increasing use of automated guided vehicle (AGV) systems in intralogistics places growing demands on on-time delivery, throughput, energy efficiency, and robust materialflow control. Particularly in large, dynamically operated spare-parts warehouses, the joint coordination of transport-order dispatching and vehicle charging constitutes a complex optimization problem that conventional heuristics and fixed threshold-based strategies cannot adequately address. Building on a simulation model of...     »