Advanced Warehouse Energy Storage System Control Using Deep Supervised and Reinforcement Learning

Abstract

The world is undergoing a shift from fossil fuels to renewable energy sources due to the threat of global warming, which has led to a substantial increase in complex buildingintegrated energy systems. These systems increasingly feature local renewable energy production and energy storage systems that require intelligent control algorithms. Traditional approaches, such as rule-based algorithms, are dependent upon timeconsuming human expert design and maintenance to control the energy systems efficiently. Although machine learning has gained increasing amounts of research attention in recent years, its application to energy cost optimization in warehouses still remains in a relatively early stage. Suggested newer approaches are often too complex to implement efficiently, very computationally expensive, or lacking in performance. This Ph.D. thesis explores, designs, and verifies the use of deep learning and reinforcement learning approaches to solve the bottleneck of human expert resource dependency with respect to efficient control of complex building-integrated energy systems. A technologically advanced smart warehouse for food storage and distribution is utilized as acase study. The warehouse has a commercially available Intelligent Energy ManagementSystem (IEMS).