The complexity of flexible job shop problems increases significantly when using autonomous guided vehicles (AGVs) for material handling. In this study, priority rules - commonly known for their simplicity and small computation time - for sequencing operations, routing jobs and dispatching vehicles are applied. Based on a discrete event simulation study with stochastic inter-arrival times, an artificial neural network is trained to learn interaction effects between the combination of different rules for sequencing, dispatching, routing, and the resulting system performance. Based on the trained network the combination of rules is optimized, reducing the mean tardiness of the jobs under varying system performance.