Dissertation
ADVANCED DISTRIBUTION SYSTEM ALGORITHMS: PLANNING CONSIDERING MICROGRID CLUSTERING, LOCALIZED VOLT-VAR OPTIMIZATION INCLUDING DISTRIBUTED ENERGY RESOURCES, AND RESTORATION USING REINFORCEMENT LEARNING
Washington State University
Doctor of Philosophy (PhD), Washington State University
01/2021
DOI:
https://doi.org/10.7273/000003140
Handle:
https://hdl.handle.net/2376/122285
Abstract
With more and more applications of distributed energy resources powered by renewable energy in distribution power systems, there is an increasing need to improve power system’s resiliency for reliable and continuous power supply and efficiency while reducing carbon emissions. To reach this target, we investigate the path to study distributed energy resources (DERs) in distribution systems from planning, real-time operation and control, restoration and reconfiguration after faults. First, we introduce metrics of microgrid performance and use these metrics in DERs planning and storage analysis with testing of battery model to evaluate transient stability and mode transition. The metrics are used in the following chapters to evaluate power loss, voltage deviation in Volt-VAR control, and operational complexity in restoration. The combination of distribution system and DERs modeling are used to validate clustering method for size and site solutions for distributed generation using time series simulation. Furthermore, sizing and siting of DERs in a distribution system has been studied to intentionally form multiple self-sufficient microgrids to increase resiliency. However, DER planning is not adequate for real-time operations and avoiding voltage violations caused by intermittent power generation from photovoltaic sources. Thus, our second contribution is a localized model-free Volt-VAR optimization (VVO) approach for network power loss minimization. The proposed controller is a combination of the extremum seeking algorithm to achieve the network-level objective without communications with other decision-making agents and an adaptive droop controller to achieve a stable response under fast varying phenomena. We also evaluate the theoretical conditions for localized results to approach to centralized optimal solutions. Finally, we implement deep Q learning (DQN) in load restoration using switch operations for random fault scenarios. To evaluate the capability of the DQN model, we build OpenAI gym compatible environment integrating OpenDSS, tune learning rate and hidden layers configuration of neural network, and make predictions using trained model. These results show that the DQN provides decisions perform close to human experience results with only partial measurements and status of the system. These three contributions include DERs planning, volt-VAR optimization and control, and load restoration in distribution power systems.
Metrics
52 File views/ downloads
71 Record Views
Details
- Title
- ADVANCED DISTRIBUTION SYSTEM ALGORITHMS: PLANNING CONSIDERING MICROGRID CLUSTERING, LOCALIZED VOLT-VAR OPTIMIZATION INCLUDING DISTRIBUTED ENERGY RESOURCES, AND RESTORATION USING REINFORCEMENT LEARNING
- Creators
- Hongda Ren
- Contributors
- Noel N Schulz (Advisor)Anurag K Srivastava (Committee Member)Anamika Dubey (Committee Member)Josue Campos do Prado (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Doctor of Philosophy (PhD), Washington State University
- Publisher
- Washington State University
- Number of pages
- 199
- Identifiers
- 99900651900801842
- Language
- English
- Resource Type
- Dissertation