Industrial Automation Component Configuration via Reinforcement Learning

The patent applies preliminary action by pre-collecting runtime property data from multiple hardware and software combinations and storing it in a database before actual prediction is needed. This pre-prepared data foundation enables the reinforcement learning model to make accurate predictions without requiring time-consuming iterative testing at the time of configuration, thus resolving the contradiction between prediction accuracy and time consumption

The patent uses copying by creating a reinforcement learning model that learns from copied data patterns in the database rather than physically testing each configuration. The model captures the essential relationships between hardware/software combinations and their runtime properties, enabling accurate predictions without replicating the actual testing process, thereby reducing time consumption while maintaining prediction accuracy

The patent applies feedback by using the reinforcement learning model to predict runtime properties for specific hardware/software combinations before deployment. This feedback mechanism provides confidence levels and predicted performance metrics that allow engineers to make informed decisions about hardware sizing, avoiding both over-provisioning and under-provisioning, thus resolving the contradiction between reliability and resource utilization efficiency

The patent uses parameter changes by adjusting hardware dimensions and configurations based on predicted runtime properties from the reinforcement learning model. Instead of using fixed generous dimensions for all cases, the system optimizes hardware parameters according to the specific requirements and predictions for each configuration, achieving reliable real-time behavior while improving resource utilization efficiency