Power grid in a changing world

Part 3: Transforming Energy Dynamics — Insights from the UK’s Demand Flexibility Service

Photo by Andrey Metelev on Unsplash

Introduction

With the risk of repeating myself in each article, balancing the supply and demand of electricity is a fundamental requirement for a reliable power grid. While previous discussions have primarily focused on the supply side (see parts 1 and 2), there’s a captivating story waiting to be told about the potential contributions from the demand side, offering exciting opportunities for tech and data companies.

This article delves into the Demand Flexibility Service (DFS) in the UK, a novel household demand-management system. Last winter, as concerns about electricity supply security loomed, the UK National Grid Electricity System Operator (NGESO) introduced the innovative DFS. This system hinges on consumers actively reducing their demand in response to operator notifications — a large-scale social experiment exploring the impact of financial incentives on energy consumption.

Join us as we explore the DFS, unraveling a story that goes beyond electricity grids — it’s a narrative of innovation, policy, and the evolving energy landscape.

Overview

UK electricity market

The prices for electricity in Great Britain are determined every half-hour through a nationwide system called the merit order. This system organises generation units by cost, from the cheapest to the most expensive. The most expensive power plant needed to meet demand sets the electricity price. Power generators, aware of market conditions, submit bids close to the marginal cost linked to the most expensive plant.

Because coal and gas plants have higher operational and resource costs, they often become the marginal fuel in the merit order. Therefore, even though gas generators contribute only about 40% to Great Britain’s total energy, gas prices influence electricity prices approximately 84% of the time. Simply put, if coal and gas are less necessary in the merit order, wholesale electricity prices drop as other generators would lower their bid prices.

The Demand Flexibility Service

The Demand Flexibility Service (DFS), the newest addition to the NGESO toolkit, aims to decrease demand during peak consumption, addressing two key concerns:

1. Avoiding Blackouts: DFS helps mitigate the risk of a blackout by managing peak load periods effectively.
2. Reducing Carbon-Intensive Sources: By encouraging households and businesses to use less electricity during peak times, DFS aims to minimise reliance on expensive and carbon-intensive sources like coal and gas, ultimately preventing spikes in energy prices.

Participation in the DFS was open to all UK households and businesses equipped with smart meters and served by a DFS provider, such as energy retailers or aggregators.

Last winter, the DFS initiative comprised 22 events (also called Saving Sessions), including both test and live sessions, held from November 2022 to March 2023. During these peak energy demand periods, participants were asked to reduce electricity usage for durations ranging from one to two hours. Incentives in the form of price reductions were offered for consumer demand reduction.

NGESO provided DFS providers with a financial incentive of at least £3 for every kWh of reduced energy. The providers had flexibility in deciding how to utilise this payment, including how much to allocate to incentivise consumers to reduce their energy demand.

DFS events data of Octopus Energy customers. Source: [1]

Estimating impact on demand reduction

Demand reduction is essentially the variance between a customer’s real consumption during a Saving Session and the consumption that would have happened if the customer hadn’t enrolled and opted in — the counterfactual scenario.

The tricky part in estimating demand reduction is that we can only directly observe a customer’s actual consumption. Determining the counterfactual consumption, or what would have happened without participation, poses a challenge in this estimation process.

NGESO’s Prescribed Approach

NGESO employs a specific methodology to calculate each customer’s counterfactual consumption, referred to as the customer’s “baseline.” The calculation for baseline consumption and demand reduction during each half-hour of a DFS event involves the following steps:

1. Calculate “Unadjusted” Baseline Consumption: Take the unweighted average of consumption during the same half-hour of the day for the ten most recent weekdays.
2. Compute Day-of Adjustment: Determine the day-of adjustment as the difference between consumption earlier on the same day and the historical baseline consumption during the same time windows.
3. Calculate “Adjusted” Baseline Consumption: Add the day-of adjustment to the “unadjusted” baseline consumption.
4. Clip Negative Demand Reduction: Ensure that negative demand reduction is clipped to 0 kWh.

The rationale behind the day-of adjustment is to acknowledge days where energy consumption is structurally higher or lower than the unadjusted baseline would suggest, accounting for factors like temperature variations. However, during the trial last year, NGESO identified instances where customers increased their consumption during the adjustment period to inflate their adjusted baseline and receive a larger demand reduction award.

It’s essential to note that the clipping step, initially added to prevent penalising and discouraging customers, introduces significant changes in the final result. It has been observed that this step can lead to considerable inflation in demand reduction estimation when compared to “unclipped” measures:

Average per-customer “clipped” vs “unclipped” demand reduction, in each Saving Session. Source: [1]

The method of estimating policy impact based on before-and-after states, while simple to understand and implement, can lead to a biased estimate. This bias arises because the “before” version is not a valid counterfactual to the “after” version. Multiple factors may have changed simultaneously during the observation period, and attributing all differences solely to the DFS is not accurate. In more formal terms, the before-after method is susceptible to omitted variable bias, making it an invalid causal estimation technique.

In response to this limitation, [1] proposes a more robust causal approach using the Difference-in-Differences (DiD) method. This method allows for a more nuanced understanding of the causal impact of the DFS by comparing changes in outcomes over time between a treatment group (those affected by DFS) and a control group (those not affected). The DiD method helps control for potential confounding variables and provides a more reliable estimate of the true causal effect of the policy.

Diff-in-diff impact estimation approach

In [1], using Saving Session data from Octopus Energy, three control/treatment designs are implemented:

• Comparison 1: A treatment group comprised of Octopus Energy customers who signed up to take part in DFS events before the first Saving Session versus a counterfactual group comprised of Octopus Energy customers who never signed up to take part in DFS events.
• Comparison 2: A treatment group comprised of Octopus Energy customers who signed up to take part in DFS events before the first Saving Session versus a counterfactual group comprised of Octopus customers who signed up after the ninth Saving Session.
• Comparison 3: A treatment group comprised of Octopus Energy customers invited to sign up to take part in DFS events versus a counterfactual group comprised of newly-acquired smart-meter Bulb Energy customers who were unable to be invited to participate.

They obtained the Average Treatment Effect (ATE) for each of the DiD designs by:

1. Collapsing Data: Condensing each customer’s time series data into two observations, one for each of the two study periods (pre-treatment vs. post-treatment).
2. Regression Model Fitting: Employing a regression model for each design. For instance, for the third design (Octopus customers with DFS access vs. Bulb customers without access), the equation takes the following form:

with:

• y, the customer’s average half-hourly consumption.
• S_post_treat_period, a binary indicator for pre-treatment period or post-treatment period.
• T_octo_customer, a binary indicator for Octopus Energy or Bulb customers.

The main coefficients can be interpreted as following:

• β0 is the expected average consumption in the pre-treatment period for control-group customers.
• β1 is the expected difference in consumption between the post and pre-treatment period.
• β2 is the expected difference in consumption between the treatment and control groups owing to eligibility for the DFS.
• β3 is the expected difference in the slope coefficient for time period between treatment groups. In other words, β3 is the difference-in-differences estimate. It indicates the differential effect of time on consumption for the treatment group compared to the control group, providing an estimate of the causal impact of the DFS.

Dif-in-diff estimation principle. Source

Final important remark on the assumption behind this method: as the illustration above suggests, in the absence of treatment, it is assumed that the rate of change between pre-treatment and post-treatment period would be the same for the treatment and control groups (the parallel trends assumption).

Now that we have the tools to estimate the DFS impact, let’s see the results of this policy.

Impact of DFS on demand reduction

Demand reduction measured by the official NGESO methodology. Source: [1]

As expected, demand reduction was much higher for customers who had signed up and opted in (0.305 kWh) than for the three sets of customers who were not participating (0.005–0.031 kWh). However, non-participating also show a small demand reduction. This outcome validates our earlier comments on the potential bias associated with the simplistic approach used in the official methodology.

Using the more robust diff-in-diff approach, the treatment effect is more in the order of 0.2 kWh per half-hour. This corresponds to a demand reducing rate of 40% comparing to the control group.

Dif-in-dif results: LATE is for Local Average Treatment Effect. Source: [1]

According to NGESO, last winter, the 1.6 millions households participating in the program delivered a total of 3.3GWh of demand reduction. However, with the more robust calculation, the number should be closer to 2.9 GWh. Even so, this amount of reduction is substantial, equivalent to powering approximately 9 million homes for an hour during peak times.

Given a participation rate of roughly 30% (based on Octopus Energy’s client statistics) and considering the 30 million households in the UK, a nationwide strategy could potentially yield around 2 GWh of consumer flexibility per Saving Session. This figure is roughly equivalent to the entire capacity of the UK’s contingency coal power plants. The potential impact of such a strategy is significant and demonstrates the real possibility for leveraging consumer flexibility to manage and reduce peak energy demand across the nation.

Is Demand Destroyed or Displaced?

The question of whether the reduced consumption is “destroyed” or simply “displaced” is a crucial consideration for policymakers and grid operators. Destruction of demand implies a permanent reduction, such as when people choose alternative activities over energy consumption, while displacement suggests a temporary shift where the reduced consumption during the event is compensated for by increased demand before or after the event.

By employing the same modeling technique as described earlier but for multiple time steps before and after the event, a graph of counterfactual consumption can be constructed. This graph helps illustrate whether the reduced consumption during DFS events is sustained over time or if there’s evidence of demand displacement, where energy use is merely deferred to other periods.

Expected average consumption by hour from a series of regression models with different time steps as target.

There are small but meaningful reductions in neighbouring time windows, indicating evidence of demand destruction, is a positive finding. It suggests that the DFS is effective in achieving its goal of reducing peak consumption without creating perverse effects, such as merely shifting energy use to other periods.

The final question at hand is the economic viability of the DFS for the NGESO. Assessing the economic feasibility involves considering factors such as the cost-effectiveness of the program, the overall impact on grid operations, and the balance between the benefits of reduced peak consumption and the associated costs of implementing and maintaining the DFS.

Cost effectiveness

The cost-effectiveness analysis of the DFS program indicates promising results:

1. Cost per kWh of Reduced Electricity: NGESO paid households and businesses nearly £11 million to achieve a total reduction of 3.3 GWh during the 2022/23 winter period. The average cost per kWh of reduced electricity was approximately £3.3.
2. Comparison with Contingency Coal Units: In contrast, NGESO’s alternative measure for the winter of 2022/23 involved contracting five coal units at a cost of around £370 million. Despite comparable total utilisation, the budget of the DFS program accounts for only 3% of that allocated for the contingency coal plants.
3. Impact on Consumer Prices: With NGESO not contracting these coal plants for the upcoming winter, the absence of expensive coal in the merit order is expected to result in lower bid prices from other units. Consequently, consumer prices are likely to decrease.

Cautionary note: it is important to point out that the winter during the analysis period experienced milder temperatures, leading to less severe peak electricity consumption. In a more resource-constrained situation or during a colder winter, the bid price for the DFS could potentially increase. Scaling up the program to a national level could yield a more significant budget, especially in situations with higher demand and bid prices.

Conclusion

The Demand Flexibility Service implemented by NGESO has demonstrated its effectiveness in reducing peak consumption, showcasing a 40% demand-reducing rate. Economically, it has also proven to be a cost-effective solution, with a budget only a fraction of what allocated for contingency coal plants.

Despite the promising results, certain factors could impact the treatment effect that are not discussed here. Factors such as the notice period (whether notice is given one day ahead or just a few hours ahead), the incentive level (£ per kWh demand reduction), and the “fatigue” factor (potential decline in treatment effect over the course of the DFS season) warrant further exploration.

Looking ahead, the DFS, currently employed to reduce demand, presents the potential for broader applications in the future. One notable avenue is its use to manage periods of high generation on the system. For instance, during low-demand periods with high wind generation, DFS could be leveraged to increase demand, optimising the utilisation of renewable energy sources and reducing the need for curtailment.

In summary, the DFS emerges as a successful, cost-effective tool with possibilities for shaping energy dynamics in a sustainable manner.

Reference

[1] Jacob, Jenkinson et al. (2023). “The Impact of Demand Response on Energy Consumption and Economic Welfare.”

[2] Elia Group (2023). “The Power of Flex.”

[3] LCP Delta (2023). “Winter Balancing Costs Review.”

[4] National Grid ESo (2023). “Demand Flexibility Service.”