AI for flexible electricity system

May 28th, 2023

This project aims to develop AI (Artificial Intelligence) technologies to manage Distributed Energy Resources (DER, i.e. solar, wind, energy storage, and electric vehicles) to reduce greenhouse gas emissions and increase the reliability of electricity network operation with high penetration of renewables. This project builds on the existing digital energy infrastructure, Data Clearing House (DCH) and advances state of the art by incorporating new AI-based tools and methods that will enable automated participation of DER resources in the electricity market in a reliable and secure way.

Background

High penetration of Distributed Energy Resources (DER) (up to 45% by 2050) is likely to play a major role in transitioning the Australian electricity sector to net zero emissions by 2050. In addition to their emission reduction benefits, individual and aggregated DER systems such as rooftop PV (Photovoltaics), storage and Electric Vehicles (EV) can participate in the future electricity market providing flexibility services, a critical need for maintaining stability in an electricity system powered by variable renewable energy resources.

However, management of DER resources has many challenges:

DER assets are distributed and are often not visible to the electricity market operators. Low cost – interoperable ways are required to facilitate their participation in the market.
Automated control and management of these assets for participation in the market requires optimal control that uses real time data, forecast information about their availability and requires control instructions to be dispatched for their management.
High uptake of DER in the network places the consumers (prosumers) at the centre of this transition. Perceived and real data and information (metadata) management risks need to be addressed to support larger scale consumer participation.

Figure 1. Aggregator platform for supporting behind the meter DER participation in the electricity market

This project seeks to develop a digital infrastructure (figure 1) that can be used to control two-way energy flows in a secure, reliable, and efficient manner while allowing consumers to derive economic benefits.

Approach

The realisation of a ‘Digital grid’ requires the integration of many enabling technologies such as real-time sensing, forecasting, controls, and optimisation. Further participation of DER in the electricity market requires secure and reliable communication and hardware interfaces. The project takes an AI-centric approach to address the three main challenges listed as work packages.

AI-assisted metadata/context modelling: Information interoperability underpins scalable deployment of data driven analytics via the flexibility platform. Information interoperability in the platform is achieved through representing all metadata in DER systems and buildings in a standardised form that can be queried. However, due to (a) heterogenous nature of sensors, (b) different naming convention from providers, and (c) how they are integrated into a building, this process is laborious and error prone. This work package will investigate AI methods at the intersection of deep learning-based time-series analytics, and semantic graphs to automate the process of smart building modelling/building digital twin for DER systems.
Improving the robustness and resilience of AIML models for electricity networks : Operating an automated, AI driven flexibility platform needs to consider many operational challenges, risks and constraints. For example, each geographically dispersed device and communication link in DER systems are susceptible to faults, manipulation, malicious and adversarial attacks. Additionally, preserving privacy of sensitive data, fairness in model deployment considerations as well. We are developing protocols and algorithms for enhancing resilience of AI models to generic and new attacks and tolerance to faults in deployment environment.
AI for DER control and dispatch: The ability to model and control complex electricity systems of the future requires a paradigm shift in how these systems are modelled now. The complexity associated with non-linear models can be addressed through the availability of data and the development of data-driven methods. The dynamic nature of variable generation from DER and consumption can be accurately predicted using novel AI methods. Such methods will need to operate over large geographic areas in near real-time, requiring careful data/compute prioritisation and the ability to shift intelligence closer to the data sources. Although model free Reinforcement Learning (RL) methods are showing promise for energy domain problems, there are various research gaps to understand the role of these methods in DER control and optimisation. Another novelty of this work package is integration of grid related constraints in building, aggregated DER control objectives, a topic that is currently overlooked in most of the studies.