Collaborative Research: Autonomous Hierarchical Adaptive Dynamic Programming for Decision Making in Complex Environment

点击翻译按钮获取中文摘要

The recent big wave of artificial intelligence (AI) not only provided tremendous advancements ranging from fundamental research to a wide range of exciting applications, but also presents enormous amounts of opportunities as well as challenges to the community. Among many of the AI techniques, adaptive dynamic programming and reinforcement learning (ADP/RL) is widely considered as one of the key methodologies for learning-based intelligent decision-making process. The objective of this project is to develop an innovative autonomous hierarchical ADP/RL approach for decision making in complex environments. By autonomously providing a hierarchical representation of sub-goals for improved learning and exploration capability, the proposed research provides a new approach to systematically and adaptively develop an optimal multi-step hierarchical temporal abstraction sequence, rather than the one-step primitive action in traditional methods. The research method advances the foundations, principles, architectures, and algorithms for autonomous learning and hierarchical control, which will facilitate the capability of learning and generalization for decision-making. This project provides unique opportunities to attract and educate future professionals by bridging the connections of ADP/RL and energy systems, and for students to work on cutting-edge problems. The team consists of two PIs with strong collaborations and complementary expertise in computational intelligence, machine learning, autonomous control, and the smart grid. This research advances the scientific foundations and methodologies of intelligent decision making in complex environments with high-dimensionality, big data, and uncertainty. The collaborations with industry integrates fundamental research into a microgrid application providing critical technical innovations to the energy sector. In addition, the developed ADP/RL based intelligent decision making method can benefit other types of complex engineering systems. Furthermore, the research results of this project are also expected to fulfill a critical need in the community by training and preparing future workforce in the cross-disciplinary areas of machine learning and energy systems. The integrative outreach and education activities will provide unique opportunities to attract women and minorities into the intelligent system and smart grid field.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.