Exploiting distributed energy resources with a virtual power plant : Intelligent market participation based on forecasts

Rakshith Subramanya

Research output: ThesisDoctoral ThesisCollection of Articles

Abstract

Virtual power plants (VPPs) are a promising solution for integrating renewable energy sources, battery energy storage, and smart loads into the modern power grid. They offer an alternative to traditional centralized power generation, which is often based on fossil fuel or nuclear power, and a key characteristic of a VPP is the profitable exploitation of the distributed energy resources that it manages. This is done by trading the capacity provided by these renewable energy resources on various electricity markets. To ensure the stability of the power grid, Frequency reserve markets are used, and VPPs, especially in Northern Europe, aggregate and trade DERs on such frequency reserve markets. The industrial informatics aspects of VPPs involve coordinating a pool of intelligent Distributed Energy Resources (DERs), predicting market prices using Artificial Intelligence (AI), and developing industrial informatics architectures for VPPs in the AI era. AI is utilized to analyze extensive datasets of historical data like electricity markets or DER capacity to discern patterns and trends. This information is then leveraged to forecast future demand and supply, aiding VPPs in optimizing their operations. Similarly, with the frequency reserve market forecasts, a VPP can make better decisions about allocating resources and participating in energy markets. This dissertation explores the integration of VPPs with DERs using various industry standards. For the optimal operation and profitability of the VPPs, DER capacity and reserve market forecasting are performed and integrated into VPPs. Also, reinforcement Learning is employed for the reserve market bidding. All the proposed architectural components, such as VPP, forecasting, and DER integration, are implemented on the cloud for seamless operation. Also, a multi-tenant architecture is proposed to implement the scalability of DER integration and various Software as a Service (SaaS) integrations like forecasting to a VPP. Building continuous software engineering practices is one of the main challenges in machine learning (ML) applications. For this purpose, this work also introduces Machine Learning and Operation (MLOps) and Cloud Design Patterns (CDPs) in the context of VPP. This research contributes to realizing a more efficient, resilient, and environmentally friendly energy system by addressing the challenges of DER integration with VPP, market participation, forecasting, and cloudification of a VPP with all the sub-systems. The dissertation begins by presenting the related work in the field, establishing the context for the proposed system. Four use cases define and explain the functional and non-functional system requirements and their implementation in detail. At last, the results are presented with conclusions.
Translated title of the contributionExploiting distributed energy resources with a virtual power plant : Intelligent market participation based on forecasts
Original languageEnglish
QualificationDoctor's degree
Awarding Institution
  • Aalto University
Supervisors/Advisors
  • Vyatkin, Valeriy, Supervising Professor
  • Sierla, Seppo, Thesis Advisor
Publisher
Print ISBNs978-952-64-1910-7
Electronic ISBNs978-952-64-1911-4
Publication statusPublished - 2024
MoE publication typeG5 Doctoral dissertation (article)

Keywords

  • virtual power plant
  • electricity market
  • battery energy storage systems
  • frequency containment reserve
  • artificial intelligence
  • reinforcement learning
  • cloud computing
  • SaaS
  • MLOps

Fingerprint

Dive into the research topics of 'Exploiting distributed energy resources with a virtual power plant : Intelligent market participation based on forecasts'. Together they form a unique fingerprint.

Cite this