Refinement of Approximate Reasoning-based Controllers by Reinforcement Learning

摘要: Abstract Previous reinforcement learning models for control do not use existing knowledge of a physical system's behavior, but rather train the network from scratch. The process is usually long, and even after completed, resulting can be easily explained. On other hand, approximate reasoning-based controllers provide clear understanding strategy learn experience. In this paper, we introduce new method to refine rules controllers. A technique used in conjunction with multi-layer neural model an controller. learns by updating its prediction behavior. Unlike previous models, our experienced operator fine-tune it through learning. We demonstrate application approach small challenging real-world problem.