Energy Singularity Challenge 2020: Testing Novel Grid Fee Models and Intelligent Peer-to-Peer Trading Strategies

The Context

The Energy Singularity Challenge at Odyssey Momentum, led by Grid Singularity and Engie, and further supported by grid operators Stedin, Alliander, Enpuls and E.ON, as well as the Florence School of Regulation, engaged diverse stakeholders to simulate, experiment and advance the deployment of agent-based bottom-up energy markets. The Energy Singularity Challenge took place virtually on 13–15 November 2020, with the first stream exploring the technical deployment of peer-to-peer energy markets, and the second stream soliciting new ways of engaging with pro/sumers to optimise and rethink an individual’s relationship with energy. This article analyses the outcomes of the Stream 1 experiments.

The Process

Three teams with experience in intelligent agent design and energy market expertise (Inavitas, OLI, and Rebase) were selected in March 2020 to compete in the Stream 1 of the Energy Singularity Challenge. In preparation for Odyssey Momentum, teams, challenge partners, and Grid Singularity worked closely together in:

• Workshops led by Grid Singularity to understand (i) the technical implementation of Grid Singularity Asset application programming interface (API), which allows intelligent trading agents representing energy assets to trade energy in Grid Singularity energy market simulations, and (ii) the broader vision and the ecosystem of the Energy Singularity Challenge including how each team’s unique expertise could contribute to the future of the energy grid;
• A virtual ‘Chaos Experiment’ hosted by Grid Singularity (GSy) in July 2020 to test the hypothesis of an agent-based bottom up energy market, where teams designed their first iterations of intelligent agents in a living, growing simulated energy community, and
• Webinars organised by Odyssey Momentum to prepare teams for the Energy Singularity Challenge at Odyssey, explore their motivation, and simplify their pitches in structured ecosystem interface slides.

To prepare and effectively conduct experiments for the Energy Singularity Challenge, the participating grid operators took part in a Grid Singularity-led innovation cycle to design and implement experimental grid fee models. Grid fees cover the infrastructure cost of the grid and can serve as a valuable tool to manage congestion by increasing the cost of using the grid at times of peak transmission to encourage self-sufficiency of energy communities during these times, and encourage prosumers to use flexible energy assets such as batteries to provide demand side management services.

The innovation cycle was organised in the following phases:

1. Plan — GSy and grid operators discussed a range of grid tariff models in a design sprint workshop, based on an adapted Google design sprint methodology with the objective to sketch solutions, including structural requirements and user interface design. The outcome was to select grid fee models that would be evaluated in the Energy Singularity Challenge, and define technical requirements to build required features in the GSy canary network to facilitate testing. A dataset suitable for running the desired experiments was identified and the baseline values were determined during the planning.
2. Build — Grid operators analysed data from a case study of a 25 member community whose dataset was used in the Stream 1 experiments to determine the specific parameters for the selected grid fee models (further explained later). Grid Singularity translated the technical requirements into specific features, which were developed and integrated into GSy’s Asset API, Grid Operator API and the GSy user interface (d3a.io), adding key performance metrics, such as the energy peak percentage, net energy, and total grid fees paid.
3. Test — The next phase in the innovation cycle was to test the selected grid fee models in the Energy Singularity Challenge, with competing teams reacting to implemented models and energy experts from corporations and leading researcher organisations, as well as civic activists and regulators participating in the discussion and analysis.
4. Deploy — Energy Singularity Challenge results will be integrated in future development, enabling researchers and grid operators to design and implement different grid fee models and trading agents in GSy simulations, and via long term testing in real-time using live data in the GSy Canary Test Network.
5. The final stages in the innovation cycle included Sharing the results in experiment debriefs and the current publication, with continued Learning to inform future development of tariff models and agent strategies.

The Odyssey Momentum Energy Singularity Challenge was structured such that each of the activities fulfilled a role towards future deployment of local energy communities:

1. Grid Singularity Canary Test Network — a prosumer community’s energy assets were connected to send live data and simulate trading in GSy’s peer-to-peer energy exchange.
2. Grid Fee Experiments — Three grid fee models generated during the plan and build phases of the innovation cycle described above were implemented using the GSy Grid Operator API, and tested against intelligent trading agents designed by competing teams.
3. Energy Community Optimisation Challenge — Teams optimised trading for a week of energy use of a simulated community with full foresight of energy supply and demand, using the GSy Asset API.
4. Challenge Questions — Questions aimed to ideate how the unique specialties of each team could fit into the growing Grid Singularity ecosystem were released to teams to answer and present during challenge debriefs.

At the Energy Singularity Challenge onset, Dr Etienne Gehain, Digital Innovation Officer at Engie focused on the value of a structured process of challenge design and the significance of created solutions for Engie as one of leading global utilities:

“I’m very impressed by what has been set up and quite grateful to be able to address a very important issue for Engie which is how to organise value sharing within an energy community.”

The Teams

Inavitas Team

The Inavitas team composed of Emre Ahmetbeyoglu, Şeyma Başaran, Erman Terciyanli, Rabia Şeyma Filmaz and captain Can Arslan is a group of professionals working in the digitalisation of energy distribution systems. The team believes that local energy markets will bring efficiency to distributed energy resource (DER) management, changing the way we produce and consume energy sustainably in our local communities.

OLI Team

With three companies in one team, the OLI team included Godwin Okwuibe, Mukund Wadhwa and captain Felix Förster from OLI Systems, Jan van Dither and Maarten Sloot from Siemens and Tako Tabak from InSignal.ai. They bring together energy market expertise, artificial intelligence and machine learning prowess, and knowledge of the technical implementations of energy communities to tackle the challenge of creating an intelligent energy trading agent.

Rebase Energy Team

Rebase Energy is a Stockholm based energy startup that aims to fight climate change by providing data and digital tools to empower the energy transformation. Rebase has developed two products, a Datahub that provides energy related data such as weather, asset and market data in a unified API, and a Toolkit that gives access to open source machine learning and optimisation tools to optimise the use of distributed energy resources. Rebase is made up by Mihai Chiru, Ilias Dimoulkas, Henrik Kӓlvegren, and captain Sebastian Haglund El Gaidi.

Priceless DSO Team

In addition to the three competing teams, Adam Alverbӓck, Hans Beckers, Robben Riksen, and Paul Bierling, represented the Dutch Distribution System Operators (DSOs) Stedin, Alliander, and Enpuls, together with Jan Pellis, Job Stuurman, Frank van Rossum, Wouter van den Akker, and Arjen Zuijderduijn serving in an advisory role. The team formed to inform the development of the Grid Operator API that enables DSOs to operate the grid more effectively by using dynamic grid fees reflecting the grid load, such as by increasing grid fees during periods of projected high energy use. The team collaborated to design grid tariff models tested during the Grid Fee Experiments.

The Challenges

Grid Singularity Canary Test Network

The Energy Singularity Challenge kicked off with the launch of Grid Singularity’s Canary Test Network, a peer-to-peer energy exchange with live energy asset data and real-time trading. Ewald Hesse, Co-Founder & CEO of Grid Singularity (GSy) revealed the story behind the network’s name:

“It’s called a Canary Network because if the canary dies while you’re working in a coal mine, you have to run. It’s a proof of concept and we will experiment simulated scenarios on a live running exchange with real data.” No canaries were harmed in the development of Grid Singularity’s Canary Network.

During the Energy Singularity Challenge, an energy community managed by OLI Systems was connected to the GSy Canary Test Network to share live data, reflecting energy requirements stemming from solar panels and smart meters through the GSy Asset API. In the GSy energy exchanges, each energy asset such as a load, photovoltaic panel (PV), or storage is represented by a trading agent which places bids or offers at specific prices, as a function of energy supply, demand, and grid fees. For instance, price is likely to decrease during a day with high solar production.

Figures 1a and 1b. Architecture of Grid Singularity energy exchanges

In the Energy Singularity Challenge, OLI Systems served the role of an aggregator for the connected community of prosumers, by sharing the data streams generated from connected household hardware through the Grid Singularity Asset API (see Figure 4). Each competing team then registered to manage a set of assets by deploying their trading agent to represent the relevant assets and submit bids and offers to fulfil the asset’s energy requirements. While GSy exchanges will in future deployment trade 15 minutes prior to production or consumption, the energy in the GSy Canary Test Network is purchased 15 minutes post-production or post-consumption, removing the need for trading agents to predict energy use and simplifying the experimental setting.

For demonstration purposes, OLI team captain Felix Förster connected his home’s smart meter to the GSy canary test network, revealing a spike in energy consumption when he made tea. The grid operator team connected via the Grid Operator API to the community market and grid market connecting the community to the wider distribution grid (see the drawing in Figure 4) and implemented a custom grid fee model to control the grid fees for the community. Whilst data in the current, first version of the GSy Canary Test Network can be viewed by asset owners and the exchange operator, the next step in development is to ensure compliance with the General Data Protection Regulation (GDPR), enabling privacy settings in the entire workflow. For the purpose of the Energy Singularity Challenge 2020 experiments, outside of this demonstration exercise, anonymized data from the EWS Schönau energy community was deployed.

Figure 2. OLI captain Felix Förster connecting a prosumer community in Germany to the GSy Canary Test Network

All energy trades in the GSy Canary Test Network are simulated, meaning that no financial transaction occurs and flexible assets such as batteries may physically charge or discharge at different times than their simulated behaviour in the Canary Network. To prove that batteries and other flexible assets can synchronise physical behaviour to a live running digital energy exchange, OLI configured a battery that physically charged or discharged when simulated trades occurred in the test network and were reported through the GSy Asset API. In addition to a small display with detailed trading information, a polar bear-shaped light indicated with green colour that the battery was charging, and with red colour that it was discharging, as shown in the figure below.

Figure 3. Energy storage demonstration to show charge and discharge via terminal and visual indicator (red polar bear for discharge, green for charge)

The performance of the running test network was monitored during the Energy Singularity Challenge, applying pressure tests such as adding additional connected assets, registering an asset to be represented in the market by an intelligent trading algorithm, and simulating a case of grid blackout. This led to valuable insights to inform future development, including facilitation of scaling and effective introduction of external asset data.

The launched GSy Canary Test Network continues to operate, allowing agents to trade over weeks and months of time, for long term study of energy asset data and market behaviour. Grid operators participating in this test network can actively read aggregated market information such as prices and amount of energy transacted to predict times of peak energy use, and adjust grid fees to manage congestion.

The GSy Canary Test Network runs at real time, meaning that assets send their actual energy usage through the Asset API once every 15 minutes. This mimics how deployed exchanges will operate, but provides only a few sets of data points every hour, requiring iterations of experiments to which doesn’t meet the needs of the ongoing research efforts to determine effective market, grid fee, and agent designs. Determining the mechanics for deployable markets require frequent iterations and experimentation. Grid Singularity simulations allow for energy exchanges to be run at warp speed, meaning one week of trading can be simulated and analysed in less than two hours. This functionality allows for rapid prototyping of grid models and experimental setups.

Grid Fee Experiments

The grid fee models tested in the grid fee experiments were designed and implemented in the structured innovation cycle described earlier, with the objective to better understand the impact of grid fees on communities with intelligent energy trading strategies. They were implemented as scripts that run through Grid Singularity’s Grid Operator API, automatically adjusting grid fees up or down either at predetermined times or in reaction to market conditions (e.g. the amount of energy traded in a market). Each team competing in the Energy Singularity Challenge managed the trading behaviour of an entire energy community with 25 members and 69 managed energy assets, aiming to improve the performance matrix for the managed energy community such as the cost of energy or self-sufficiency while adapting trading strategies to changing grid fees.

Paul Bierling from Stedin explained the underlying interest of grid operators: “Our goal is to try to manage the chaos by putting incentives in the system for the teams to use their flexibility, not only in a way that’s good for them, but for the system as a whole, so that we can keep the system cost low. One of the ways to do this is to apply a grid fee. The GSy platform makes it easy to conduct numerous experiments in just a few hours, and iterate towards a new tariff system that allows for effective grid management.”

Figure 4. Energy Singularity 2020 Experiments Setup: DSOs managed the Grid and Community market fees using the GSy Grid Operator API. The trading strategies of 68 assets in the community were managed by competing teams through the Asset API.

The three grid fee experiments included:

1. Baseline — a set grid fee of 4 cents per kWh in the Community Market and 4 cents per kWh in the grid market with the market maker served as a baseline scenario for comparison.
2. Time of Use Grid Fees (ToU) — a grid fee schedule, known to the teams and agents, is set and automatically enforced by a grid fee agent. The aim is to reduce peaks by encouraging local consumption of energy during times of predicted peaks during the day.
3. Aziiz Model Reactive Grid Fees — a grid fee agent is implemented that reacts to the total amount of energy imported / exported by the community, increasing or decreasing grid fees to moderate peak congestion loading of the grid (based on Smartgrid Greenparc Bleiswijk Layered Energy Flexibility Project no. TEUE518012 funded by the TKI Urban energy Netherlands).

Each grid fee model was implemented as a python script, available open source on the GSy GitHub, with Wiki documentation on Grid Operator API interaction.

The experiment results showed that grid tariffs have a notable effect on the energy import and export peaks of a community. In these initial tests, the dynamic grid fee implementations reduced import and export peaks by up to 10% by rewarding agents for intelligently using their battery flexibility. The difference in peak levels is explained by the change of behaviour of flexible assets that react to the grid fees adjustments. For instance, under the Aziiz tariff model, with lower fee values, the community imported and exported more energy during the late morning than under the ToU tariff. With the ToU model, the fees applied are higher during the same period and thus incentivise the community to store its excess generation and use it when grid fees increase. Another interesting quality of the Aziiz model vs. the Time of Use model is the oscillating spikes of energy usage due to the Aziiz model drastically changing grid fees in reaction to energy use conditions, as shown by the energy usage behaviour in Figure 5. Iteration and optimisation of the grid fee and trading algorithms is being carried out to further improve these metrics and understand the bounds of benefit.

Figure 5. Different grid tariff models applied in Energy Singularity Challenge 2020 (Time of use in red and Aziiz model in orange) induce various behaviours in the use of flexibility, impacting the import and export of the community (energy imported and exported every 15 minutes under the Time of Use in brown and Aziiz model in green).

The competing teams shared a similar view that the energy production and consumption of each community member should be individually balanced before trading for any excess generation or consumption with other members of the community. However, each team took a different approach to design their trading agent;

• Inavitas built a decision tree algorithm, which input the state of the market and their managed energy assets and took action according to the designed path,
• OLI implemented a reinforcement learning agent that learns to interact with the environment (here: Grid Singularity exchange) and improves based on past experiences. During the Energy Singularity Challenge, OLI also created a mechanism it named the Burrito Factor, which indicates when the agent should adjust its learning strategy based on recent performance, and
• Rebase designed an optimisation algorithm with the objective function of reducing the community’s aggregated energy bill.

The varied approaches provided experimental data to evaluate and further investigate and develop the selected grid fee models. Robben Riksen of Alliander observed:

“It was interesting to see how teams reacted with the different strategies and that in each of the cases, the ability of using smart agents to reduce peaks was shown, and the different approaches of the teams led to different outcomes. However, lower peak doesn’t necessarily imply a lower grid fee.”

The balance between market design, agent behaviour, and grid fee models is a complex control system that requires many iterations to fully understand and implement effectively. These experiments bring additional questions in the design of smart agents such as optimising single homes vs the entire community. Felix Förster of OLI Systems deduced:

“We have a right set of questions that we can take out of this hackathon and continue to answer in order to come up with specific requirements to build a real-life product to serve real-life people, with real-life devices and real-life apps.”

Energy Community Optimisation Challenge

To better understand the upper bounds of benefit for trading agents to optimise trading behaviour in an energy community, competing teams were provided with a digital twin of the 25-member EWS Schӧnau energy community dataset including one week of data, providing perfect predictions of energy use for the period used in the experiment. The task was to use the data to optimise the agent’s trading strategy to reduce the transmission peak in the grid market, with community energy bills and self-sufficiency of the community also monitored.

The competing teams ran the simulation locally on Grid Singularity’s open source energy exchange codebase to train and test their optimisation agents, and then tested their agents on Grid Singularity’s user interface to receive visual feedback and benchmark performance.

Each team took widely different approaches to solving this problem, adapted from their core strategies as described in the ‘Grid Fee Experiments’ section above. The team with the most significant outcome was Rebase, which was able to reduce export peaks by 31.4%. Their approach was to run an optimisation of energy trading over the week time period with the objective function set to minimise the export peak. With perfect foresight of energy usage patterns, they were able to see peaks coming and have their battery buy or sell accordingly. The other teams did not perform as well as expected, even increasing the baseline peak, which leads to questions about how the market mechanism, pay-as-bid in the experiments, may be designed such that agents in unfamiliar situations can be guided towards benefiting both the financial gain of a represented energy asset and the grid and social metrics of the community.

Figure 7. During the Energy Singularity Challenge 2020, the Rebase team was able to reduce export peaks to 68.6% of the baseline, although the import peak remained at 100% of the baseline.

Challenge Questions

A set of three challenge questions were revealed to teams periodically over the hackathon weekend to investigate while designing, implementing, and deploying their algorithms in the ongoing experiments. The objective was to understand how teams thought about the energy market problems they were solving and their expertise could support the future growth of local energy marketplaces

The teams were first asked to describe how they approached agent design, including the mathematical formulae, the flow of data and information used by their trading algorithms to make decisions in the energy market. Teams generated flow charts, equations, diagrams, and proposed novel KPIs (like the noted OLI’s ‘burrito factor’ to gauge if a household is performing at the same level as the neighbours) that will be used to inform the further development of the structure and function of the GSy Asset API.

Figure 8. The Energy Singularity Challenge 2020 Miro board where teams laid out how they approached designing their agents, market design, and future business models.

The second question addressed the market mechanism used in the Energy Singularity challenge experiments, set as the auction-based two-sided pay-as-bid. Teams compared auction vs. continuous market types, how different mechanisms could tie into country level wholesale markets, and how different market types affect the degrees of freedom in agent behaviour and potential benefit. Jedi, including government representatives, prosumer groups and energy researchers, debated market, regulatory, and technical impacts and shared relevant research articles.

For instance, Tim Schittekatte of Florence School of Regulation challenged self-consumption as a target:

“Why is self-consumption of local PV an objective? Is it monetisable? From a power system perspective, self-consumption isn’t that valuable except for avoiding some losses. The timing of in-feed and export is more important for improving grid flexibility, and this can be influenced by different grid fee models. Self-sufficiency is different and can be valuable in reaching higher grid resiliency.” He proceeded to suggest a research paper with further insight on the tradeoff of decentralised and centralised resources.

The discussed market mechanisms will be investigated for future technical integration into GSy energy exchanges. The grid operator team also described how different grid fee models may be implemented with different functionalities implemented in the Grid Operator API.

Jan Pellis, Strategist and Innovator at Stedin underscored:

“The grid fee models we tested in the framework of the Energy Singularity Challenge are valuable input for the future energy system design. The grid fee experiments showed both the workings and the value of the digital twin environment for innovating grid management tools. In the coming years we will see the first new grid tariff designs come to reality, and these may not be as advanced as those tested during this hackathon. We look forward to further development of the simulation model to enable realistic testing of current and future grid fee models, accelerating the implementation of innovative grid management. ”

Finally, the third question explored business model concepts that would enable data scientists to be rewarded for providing data or algorithms to members of energy communities, grid operators, or other decision makers in the broader energy market ecosystems. This is where each team’s unique specialisations outside of agent design came into play;

• Inavitas focused on addressing specific needs of grid operators, exchange operators, and utilities, by providing services for market and grid surveillance, demand forecasting and management, and algorithms for peak shaving and improved grid resilience,
• OLI laid out a profit sharing model for trading agent suppliers to build partnerships with consumers and allow consumer preference to be captured, and
• Rebase proposed a smart algorithm and data marketplace concept, where the Rebase toolkit would allow utilities, grid operators, and exchange operators to purposefully interface with consumer hardware and data streams.

The Finale

The process used to design and implement the necessary features for the grid fee experiments conducted during the Energy Singularity Challenge brought together grid operators, utilities, academics, research groups, startups, and other stakeholders to discuss and test methods surrounding the latest in distribution grid tariffs, decentralised market mechanisms for congestion management, and user interface design for grid operators and regulators in Grid Singularity’s Energy Exchanges. The experiment outcomes, including initial findings that intelligent trading algorithms and carefully designed grid fee models can reduce peak congestion by 10% and the fruitful discussions, highlight the future research area needs in the design of a fair and inclusive market mechanism and further development of the business model concepts for data scientists and other technical solution providers to empower local electricity markets.

The team final presentations and the accompanying discussion may be viewed here:

As much as collaboration was central to the Energy Singularity Challenge, teams competed for a prize of 10,000 Euros, judged on the chances of implementing the agent in a pilot environment , solution prototype performance in experiments, and ecosystem alignment. The jury for the first stream of the Energy Singularity Challenge included diverse energy market stakeholders;

• Dr Etienne Gehain, Digital Innovation Office at Engie and sponsoring partner of the Energy Singularity Challenge
• Luis Hernández, Head of Innovation Energy Communities & Networks, E.ON Group Innovation
• Dr Sarah Hambridge, Grid Singularity Product Owner and local energy market expert
• Jan Pellis, Strategist and Innovator at Stedin and advising member of the Priceless DSO Team.

Following team final presentations, and additional jury questions and deliberation, the jury decided on the following:

1. First Place — Rebase, scoring the highest in pilot support and solution prototype
2. Second Place — OLI, strong in ecosystem alignment and meriting a special recognition
3. Third Place — Inavitas, with a well-rounded score across all categories.

Importantly, the competing teams have already entered post-hackathon discussions with the energy corporations that supported and observed the Energy Singularity Challenge, to commercially offer technical solutions. Rebase’s algorithm marketplace, OLI’s method to onboard communities and hardware assets to interface with the exchange, and Inavitas’ concepts for the introduction of local energy trading in the Turkish market will all play a vital role in the road to the deployment of local energy communities.

A major outcome of the preparation and execution of the Energy Singularity Challenge was the establishment of an innovation cycle to plan, build, test, and deploy incremental features of a grid-aware exchange, such as an enhanced Grid Operator API, and test grid management tools such as diverse grid fee models. This effective, accelerated prototyping with initial deployment in the live-running GSy Canary Test Network, will contribute to the continuous development of grid-aware local electricity markets, with real monetary and energy transactions, resulting in economic and social benefits for communities.

Luis Hernandez, Head of Energy Communities and Networks at E.ON Innovation, was enthused by novel business models centring on energy communities sparked by the Energy Singularity Challenge and enabled by Grid Singularity’s technology based on a bottom-up market design:

“It was inspiring to see how many ideas are already underway for the future of our energy industry that will be driven by energy communities. I’m looking forward to seeing the evolution of this interesting approach from Grid Singularity.”

The technical milestone of the launched GSy Canary Test Network will serve as a tool to continue the iterative innovation process towards an inclusive and functioning market design. Grid Singularity will continue collaborating with grid operators, utilities, participating teams, the Florence School of Regulation and other researchers and partners to expand the test network to include more energy assets and communities, monitoring its performance for potential issues such as data quality and data privacy and experimenting with different market types, grid fee models and intelligent trading agents.

Future experiments may be designed to:

• Optimally size or place new energy assets such as PV panels or battery storage in a community,
• Calculate the economic and social return on investment of energy assets,
• Create algorithmic approaches to managing the energy community in extreme scenarios such as a blackout, and
• Design new metrics to analyse market performance, such as fairness and satisfaction.

The initiated collaborative, peer-validated innovation process first implemented in the Energy Singularity Challenge will be continued, fostering knowledge sharing and advancing the technical development towards deployment of local energy markets that unlock new degrees of freedom for the individual user and create social energy networks, while optimising the grid infrastructure and energy resources.