Supervisor:
Zhe Chen

Co-Supervisor:
Zhou Liu

Assessment Committee:
Associate professor Zhenyu Yang (Chair)
Professor Frank Pillipson, Maastricht University, Netherlands
Professor Andreas Sumper, Universitat Politècnica de Catalunya, Barcelona Spain

Moderator:
Yanbo Wang

Abstract:

With the development of the global economy, the problems of energy shortage and environmental pollution are becoming more and more prominent, which promotes the development of new energy generation. Renewable energy generation through distributed access to the power grid is an important way for new energy generation to access the power system. However, renewable energy generation has strong randomness and volatility, and the access of a large number of renewable energy sources (RESs) brings profound changes to the operation of the power grid. In addition to this, the stochastic nature of energy prices and fluctuating behaviours of controllable loads including electric vehicles (EVs) bring risk to the operation of power system.
Another ongoing energy transition is the integration of different forms of energy. Multi-energy system (MES), have the potential to yield significant sustainable, efficient, economic and resiliency benefits. However, intermittent RE generation, uncertain and heterogeneous load demands, and balance-of-system costs render the traditional energy analysis methods obsolete.
Artificial Intelligence (AI) technology is an important tool to address the above challenges. As an important branch in the field of computer science, AI technology aims to realize the self-improvement of computers and the simulation of human intelligence by refining knowledge and experience from data. Since the knowledge extracted from data has a certain generalization ability, AI methods can cope with the uncertainty of the system's source load and enable online decision-making.
Therefore, the objective of this thesis is to apply AI methods to ensure the effective and reliable energy management strategy for the MESs. To this end, this project starts with a low-carbon economic dispatch strategy for the electricity-gas MES, in which the flexible coordination between the carbon capture system and power-to-gas units is considered. Relative contents are presented in Chapter 2. Second, to address the issues of centralized energy management, a decentralized energy management strategy is developed in a residential MES, where multi-agent deep reinforcement learning (MADRL) method is applied to regulate the internal energy conversion and external energy trading behaviours. Corresponding contents are presented in Chapter 3. In Chapter 4, to investigate a decentralized energy management strategy for multiple MESs, a bilevel energy management framework is proposed, where an MADRL-based control strategy is proposed for the bottom-layer multi-energy microgrid (MG) cluster, and the upper-layer energy routers determine the optimal energy trading based on the bottom-layer information feedback. Furthermore, energy hub (EH) is an effective solution to provide energy management for the MES. In recent years, electric vehicles have also been connected to the grid on a large scale. Therefore, considering EHs and EVs belong to different entities, an improved MADRL-based decentralized energy management strategy is proposed to maximize the profits of EV entity and minimize the energy costs of EH entity. Besides, a specific neural network is used to tackle the complex uncertainties so that the performance of the proposed method is improved. Relative
contents are presented in Chapter 5. Finally, the conclusions of the thesis are introduced in Chapter 6.
To conduct efficient simulation of the proposed AI-based energy management strategy, a series of case studies were performed on Python. Specifically, training datasets come from real-world historical datasets, and AI algorithms are programmed based on TensorFlow. Besides, algorithm comparison is also conducted to illustrate the superiority of the proposed method. Simulation results demonstrate that the proposed strategy can (i) reduce energy costs, (ii) deal with uncertainties, (iii) provide real-time energy dispatch, and (iV) realize decentralized energy management for different entities. Besides, the proposed method shifts the computation from online to offline, which greatly reduces the computation of online execution and facilitates later applications.

As the issue of climate change is being more and more serious, the action of carbon emission reduction is becoming increasingly important, and energy systems play a significant role in it. With the growing popularity of the integrated energy system (IES) concept, better performance can be achieved with coordination among different kinds of energy systems.

At present, the generation based on fossil fuel still has a relatively large proportion of energy supply, and the connections among different kinds of energy systems are not strong enough. Accordingly, it is worth investigating coordinated planning strategy of IES, which can help facilitate it to evolve the energy supply structure and reduce carbon emission.

In terms of operation investigation, coordinated operation of IES can also help reduce carbon emission. However, it is usually difficult for operators of different energy systems to share detailed operation parameters, due to the requirement of privacy protection. Hence, it is meaningful to explore the privacy-preserving operation strategy of IES.

Under these circumstances, the research has been conducted in this project. The main works are summarized as follows:

1. A multi-stage planning strategy of the integrated electricity-gas-heating system (IEGHS). This strategy can coordinate the installations of new facilities with the retirements of coal-fired generation plants. In which, the new facilities mainly include renewable and low emission generation plants as well as transmission lines and pipelines. It can smooth the transition of the energy supply structure of the IEGHS from the configuration of high carbon emission to that of low carbon emission. Besides, the flexibility of the gas and heating systems is used to help reduce the total economic cost of the planning.

2. A privacy-preserving operation strategy of the integrated electricity-heating system (IEHS) based on the distributed optimization algorithm. This strategy is proposed for the coordinated operation of the IEHS, which can help reduce the economic cost of the operation and reduce the carbon emission as well. It divides the optimization problem into the electricity subproblem and the heating subproblems. The iterative calculations are conducted among them to obtain the detailed operation plan for the whole IEHS. Since only limited operation parameters are exchanged in the iterative calculations, the requirement of privacy protection is satisfied.

3. A privacy-preserving operation strategy of the IEHS based on the equivalent model of the heating system. This strategy is proposed with the same purpose as that in the work 2. The equivalent model is formulated based on the neural network, and the comprehensive strategy is proposed to obtain the detailed operation plan of the whole IEHS. Since the equivalent model only contains limited operation parameters, the requirement of privacy protection is satisfied.

Based on the above works, planning and operation of integrated energy system in the context of carbon emission reduction are investigated to a certain extent. Besides, the related case studies verify the effectiveness of the proposed strategies.