Research on Operation Optimization of Building Energy Systems Based on Machine Learning

学生名:徐 阳

研究テーマ:Research on Operation Optimization of Building Energy Systems Based on Machine Learning

(機械学習に基づく建物エネルギーシステムの運用最適化に関する研究)

入学年月:2020.10

修了年月:2023.09

取得学位:博士（工学）

論文概要: Renewable energy has developed steadily in recent years in the context of energy shortages and safe supply requirements. Since over 40% of total energy consumption comes from buildings, increasing the self-sufficiency rate of renewable energy in buildings is critical. While Japan's implementation of the feed-in tariff in 2011 led to explosive growth in household renewable energy equipment, the trend slowed as the feed-in tariff price decreased. Therefore, it is urgent to reduce further the cost of running household renewable energy equipment. This research focuses on applying machine learning in optimizing building energy system operations further to reduce the operation cost of building energy systems and increase the self-sufficiency rate of renewable energy.
In Chapter 1, Introduction and Purpose of the research.
In Chapter 2, Methodology. Chapter 2 focuses on the key concepts and methods used in the study.
In Chapter 3, Materials and Data Preprocessing. Chapter 3 provides an in-depth analysis of the data resources and this study's preprocessing steps.
In Chapter 4, Potential Analysis of the Attention-based LSTM Model in Building Energy System. Chapter 4 aimed to evaluate the potential of using an attentional-based LSTM network (A-LSTM) to predict HVAC energy consumption in practical applications.
In Chapter 5, Operational Optimization for Building Energy Systems Using Value-based Reinforcement Learning. Chapter 5 presented the proposed model-based deep reinforcement learning algorithm called Model based Double-Dueling Deep Q-Networks (MB-D3QN). This algorithm optimizes the cost-effective operation of a residential house equipped with a grid-connected PV-battery system in Japan.
In Chapter 6, Operational Optimization for Building Energy Systems Using Actor-Critic based Reinforcement Learning Considering Real-time Energy Prediction. Chapter 6 proposed a model-based RL control method considering real-time prediction values for operation optimization of the residential PV-battery system. The optimization goals aim at reducing the energy cost of the microgrid and ensuring that the PV self-consumption ratio is not lower than the baseline model.
In Chapter 7, Conclusion and Outlook. A summary of each Chapter is concluded.