En Route to Energy Singularity Odyssey Momentum: Learnings from the 2020 Chaos Experiment
On July 16–18, 2020, a virtual live experiment took place to test the hypothesis of an agent-based bottom up energy market by simulating the conditions of a future smart grid. Intelligent trading bots with competing incentives designed by three teams, including experts from Inavitas energy intelligence platform, Greenlytics weather analytics solutions for the energy industry, and OLI Systems collaborating with Siemens and insmart.ai, faced off in an interconnected local energy market simulation enabled by Grid Singularity’s energy platform. They predicted energy prices, placed bids and offers for energy and optimised placement of energy assets, reacting to changing market conditions such as flexible grid fees designed by a grid operator team including Stedin, Alliander, Enpuls, and E.ON. This experiment is part of the Odyssey Momentum Energy Singularity Challenge, led by Grid Singularity and Engie.
Experiment Insights
The bottom-up agent-based local energy market model prevails over traditional market design, demonstrating that distributed energy exchanges are both possible and more effective for consumers and grid operators.
- Agent-based bottom up energy markets can be functionally enabled by using technology solutions such as Grid Singularity’s open source energy exchange platform.
- Multiple intelligent agents with competing incentives can work in ‘coopetition’ to impact economic and social metrics in local energy markets.
- Communities without their own generation or storage can also benefit from local electricity markets by purchasing energy surplus from participating prosumer communities.
- Flexible grid fees can be effectively deployed as a grid management tool to proactively plan for predicted congestion and reactively adapt to grid conditions.
- Local electricity markets can scale by connecting additional energy resources, households, and even entire communities to existing local energy exchanges.
Experiment Setup
EWS Elektrizitätswerke Schönau eG Digital Twin
The simulated local energy community was a digital twin of the EWS Elektrizitätswerke Schönau eG community in the south of Germany, including 23 prosumer homes with diverse combinations of load profiles, generation capabilities, and flexibility-providing energy storage. A consumer-only version of the same community was also modelled in a connected market to evaluate the impact of the neighbouring prosumer-enabled community. The Energy Singularity Chaos Experiment builds upon Grid Singularity’s previous research in enabling local electricity markets based on the EWS Elektrizitätswerke Schönau eG dataset simulations, described in our recent report.
Market Design
Without a local electricity market, devices that consume energy, if not supplied by the house’s storage or generation capacity, would acquire electricity at the market rate of 30 cents per kWh from the EWS Schönau utility. Photovoltaics (PVs) would sell excess energy to the utility at a feed-in-tariff rate (FiT), set at 11 cents per kWh. With a local electricity market (using the two sided pay-as-bid clearing type), a PV can sell energy to a nearby load instead of selling it to the grid at a price that is higher than the feed-in tariff and lower than the utility rate, a benefit for both the PV and load. This market design assumes a regulatory framework permissive of peer-to-peer energy trading and local self-consumption.
Intelligent Trading Strategies
Each energy asset such as a load, PV, or storage is represented by a trading agent which makes decisions to bid or offer energy at specific prices. The energy trade price is optimally a function of energy supply, demand and grid fees. For instance, it is likely to decrease during a day with high solar production. Agents try to predict this optimal price and arbitrage the energy market with their trading strategies, leveraging past market data (minimum, maximum, average and median prices), amount of energy traded, trading events, and energy requirements. The information available through Grid Singularity’s trading API is detailed in our documentation.
Template agents with deterministic trading strategies that do not take into account current market information are the current default in Grid Singularity’s platform, bidding or offering linearly between a starting and ending value. In the Chaos Experiment, each team managed the trading strategies for the energy assets of four homes in a prosumer community intelligently, overriding the template strategies to optimise a chosen metric, such as to minimise energy bill, maximise self-sufficiency or some combination of these. In the week prior to the Chaos Experiment, teams trained their agents using two months of historical energy data and the trading API.
Experiment Rounds
The Energy Singularity Chaos Experiment has been conducted in the framework of wider Odyssey Momentum, led by Rutger van Zuidam, who describes its ambition: “to feel the freedom to experiment together with new technologies, looking at a challenge from different sides of the market, and see what can happen in a space of sheer possibility.”
This notion is echoed by Kerstin Eichmann, Vice President for Partnership and Acceleration at E.ON Group Innovation GmbH:
“We decided to participate to learn and experiment with new tech concepts like open source, IoT, and AI to make the grid more sophisticated. I would like to see a world where all participants can access and benefit from an open and inclusive market design. While there is the risk that we disrupt ourselves, we prefer to learn and be an active participant in the innovation journey.”
Four experiment rounds were conducted, each simulating seven days of energy trading in the Schönau market in approximately two hours time (Table 1). Between simulations, teams adapted their agent strategies and discussed results with the participating experts from Grid Singularity and energy corporations. A virtual whiteboard tool was used to facilitate the collaboration process, developing and sharing ideas and obtaining feedback. Video highlights of the Energy Singularity: The Chaos Experiment are displayed on Grid Singularity’s Youtube Channel (Kickoff, Round 1: Sandbox, Round 2: Baseline Definition, Round 3: Grid Fee Mayhem and Round 4: Virtual Rollout) and key goals and outcomes presented in Table 1 below.
Discussion of Experiment Findings
- Agent-based bottom up energy markets can be functionally enabled by using technology solutions such as Grid Singularity’s open source energy exchange platform.
Round 1 of the Chaos Experiment was the first time that multiple intelligent trading agents with competing incentives interacted in Grid Singularity’s simulated energy market, acting as a proof point that this complex interaction is possible in conditions reflecting real-world local electricity markets. Each team registered to manage their set of energy assets through a provisional device registry, allowing their agents to receive market information, energy requirements, and other input data, then make bids and offers based on the outputs of their trading algorithms. The bids and offers were injected into the live-running simulation through the Grid Singularity API, which facilitated the interaction of asset trading strategies.
Etienne Gehain, Engie’s Digital Innovation Officer, described the Grid Singularity energy exchange platform deployed for the Energy Singularity Chaos Experiment as “an actual minimum viable product, a future service offering for current and newly established energy communities.” He emphasised the value of combining emerging technology solutions to move forward: “We strongly believe that this kind of technology will be part of our future.”
In addition to proving the impact of smart trading strategies, a wealth of feedback was received that has been integrated into Grid Singularity’s development plan, including the following;
- Additional data to be provided through the API to enhance prediction models and trading strategies, such as reporting grid fees;
- Simplification of agent decision making, such as automatically clearing the energy usage of a PV and load in the same house, to streamline the trading process;
- Additional concepts for charts, graphs, and metrics that supplement information on the user dashboard to clearly convey the state of the market and owned assets.
2. Multiple intelligent agents with competing incentives can work in ‘coopetition’ to impact economic and social metrics in local energy markets.
Three scenarios are compared in Table 2 below to illustrate how intelligent trading agents impact social and economic metrics:
- The baseline scenario represents the current state of the energy market. After clearing internally, households bought remaining energy demand at the market maker rate of 30 cents per kWh and sold any surplus generation at the feed in tariff rate of 11 cents per kWh.
- The template agent scenario introduces local electricity trading among households using the deterministic trading strategy default in Grid Singularity simulations.
- The smart agent scenario introduces team-managed intelligent trading strategies for approximately half of the prosumer community’s energy assets.
Both the template and smart agents show a dramatic improvement in self-sufficiency compared to the baseline, proving the benefit of local electricity exchanges to community social metrics. This improvement is likely due to the more effective use of flexible storage devices at the local level, leading to more energy used from local sources.
There is a significant improvement to energy bills in the community with the introduction of local electricity markets. However, the value and the recipient of the benefit can shift when smart trading strategies are introduced. Table 3 below shows the average prices and total revenue of a battery, improving three-fold when managed by an intelligent agent. This more intelligent battery strategy results in slightly higher total electricity bills of the prosumer community, as PVs and batteries value their generation more highly than with the template agents.
This result indicates that further research is required to learn about how to implement smart agents to positively impact community and individual economic and social metrics. One major learning is that teams need more time and scenarios to effectively train and test their trading agents. Training on a single community structure with limited time and data does not allow agent designs to optimally scale or adapt to unforeseen situations. The development of effective smart trading strategies will be an iterative process.
3. Communities without their own generation or storage can also benefit from local electricity markets by purchasing energy surplus from connected prosumer communities.
Consumer-only communities can connect to prosumer communities to purchase surplus energy generation at a price lower than the utility rate, reducing electricity bills and increasing local energy consumption during these periods.
A number of factors influence the benefit to connected consumer-only communities:
- The amount of generation surplus available from connected prosumer communities,
- The amount of competitive energy demand within the consumer-only community or from other, connected communities,
- The trading strategies of energy assets in the consumer communities, and
- The grid fee structure between the connected communities.
Communities can also benefit from being interconnected during extreme events, such as a grid blackout. When a blackout was simulated during round 3 of the Chaos Experiment, the consumer-only community was able to sustain grid independence and avoid a blackout for multiple hours by buying electricity from the prosumer community.
4. Flexible grid fees can be effectively deployed as a grid management tool to proactively plan for predicted congestion and reactively adapt to grid conditions.
Grid Singularity’s platform allows grid operators to flexibly set and adjust grid fees for each hierarchical market in order to account for cost consumers pay to use the grid infrastructure, such as transmission lines and transformers. Grid fees make it more expensive to trade energy over larger distances, encouraging simultaneous consumption of locally produced energy and thereby reducing the burden on the energy grid. Grid operators use historical data and simulations to predict that energy usage may peak at a certain time of day or under a certain combination of market conditions. A proactive grid fee schedule can then be set to balance predicted usage of a transmission line with its capacity.
Jan Pellis of Stedin explained the strong interest of grid operators in experimenting with new grid management mechanisms:
“Chaos is the worst nightmare for grid operators. The current grid infrastructure and the grid tariffs were not designed or built for an open energy system. If there will be chaos, we want to understand how the grid can still work for everyone, and how grid tariffs and new means of engagement by grid operators can positively influence local market performance and grid utilisation.”
In round 3 of the Chaos Experiment, seven different grid fee scenarios were run consecutively in the simulation, by changing the grid fee once per day over seven identical days. The team trading strategies remained constant, but price matching was affected by the changing grid fees. Figure 3 compares two days of the same energy data with different grid fees (no fee on the left and 4 cent fee per kWh traded on the right). The peaks were reduced with the introduced grid fee, which can be seen by comparing the segments highlighted by orange arrows for each scenario. As the interaction between grid fees and energy usage has a complex relationship, more experimentation is required to fully understand and effectively implement this grid management tool.
Reactive grid fees allow grid operators to read in live data from the grid and market, and adjust grid fees accordingly. For example, if predicted usage of a household is 10 kWh for the 15 minute market slot, but the actual usage is 15 kWh, the grid operator could increase grid fees for the next relevant market slot to reduce energy demand along the affected distribution cable.
In round 4 of the Chaos Experiment, the grid operator team enacted a reactive grid fee schedule, increasing grid fees according to the grid utilisation, measured by the kilowatt hours of energy traded. The plan positively affected trading, but as other factors were not held constant in round 4, further experimentation is needed to more precisely assess its outcomes.
5. Local electricity markets can scale by connecting additional energy resources, households, and even entire communities to existing local energy exchanges.
Experiment rounds 1, 2 and 3 were run with a static number of energy communities and energy assets; two communities with 23 homes each, and a total of 80 energy assets were simulated. To meet the needs of the energy transition, such a system must be able to scale to integrate added assets, households, and communities to eventually incorporate the energy trading needs of the entire grid.
Round 4 of the Chaos Experiment simulated such a ‘Viral Rollout’, where energy assets, homes, large energy producing installations and entire communities were progressively added to the existing energy exchange over the course of the simulation. The timeline in Figure 4 below shows seven trading days in the simulation, with diamonds representing simulation events, which could be an added energy resource, home or community, or changes to the grid fee. The grid fees were adjusted frequently during each trading day based on a reactive grid fee schedule set by the grid operator team.
Multiple times in the simulation, teams were given the option to add a solar panel or a battery to one of the existing houses. They selected the asset type and placement based on the balance of supply and demand in their community and the performance of their respective agent strategies. Subsequently, more households were added to the existing communities, followed by adding an entire separate consumer-only community, which started competing to consume excess generation from the prosumers in the participating community. The teams also added solar farms at the community and next level distribution grid to balance energy supply and demand after discovering a significant mismatch between community demand and supply.
One team opted to install a solar panel in the hierarchy of one of the consumer-only communities, citing that they would expect to benefit more by purchasing electricity from a solar panel in their community rather than from the connected prosumer community. This caused prices in the formerly consumer-only community to drop below their previous minimum rate. Further analysis, including a calculation of the return of such an investment, presents a valuable use case of running these types of simulations, indicating the optimal sizing and time period of a potential investment versus reliance on trade with a prosumer community alone (or to a lesser extent).
Conclusion
The Energy Singularity Chaos Experiment acted as a proof point for agent-based bottom-up energy markets enabled by Grid Singularity’s energy exchange platform, and provided a means of feedback for its further development, as well as development of third-party trading strategies for energy assets participating in local electricity markets. The experiment validated new business models based on peer-to-peer energy trading, demonstrating the potential of hardware and software solutions for operating local exchanges, managing the smart grid and optimising connected energy assets with intelligent trading algorithms. The participating teams and the Energy Singularity partners exhibited an unexpectedly high level of engagement and openness to share knowledge, including analysis of results, ideas of how to compete better, how to prevent actions of a malicious actor, and even revealed the mathematical formulae behind their agent strategies.
Three Key Lessons
- Live user testing is invaluable in informing the software feature development to enable energy exchanges and thereby implement local electricity markets.
- A continuous research and development process is necessary. Scientifically sound analysis of results require additional iterations of both the trading agent design and the proactive and reactive grid fee design to estimate the extent of the smart trading potential revealed in the July 2020 Chaos Theory Experiment. Notably, a more elaborate public experiment will be conducted during the Odyssey Momentum Energy Singularity Challenge in November 2020, aiming towards deployment of a proven LEM structure in pilot energy communities.
- Multi-stakeholder collaboration benefits all market participants. Data scientists, grid operators, exchange operators, and consumers openly and effectively shared information and learned from each other, identifying complementary roles for a successful next stage deployment.
