Leveraging automated demand response to accelerate the clean energy transition

Achieving global net zero emissions targets necessitates the rapid deployment and integration of variable renewable energy onto the grid. This makes the task of balancing the energy system increasingly challenging as patterns of generation and demand are not aligned over the hours of the day. Demand-side management – a broad class of programmes designed to encourage consumers to modify patterns of their energy use – has the potential to offer at least a partial solution to the intermittency challenge, as consumers often use energy services that are time-flexible. 

The usage of power-intensive household appliances such as air conditioners, washing machines, dryers and immersion heaters could potentially be delayed by short time intervals if users are adequately incentivised to do so. Simple and innovative IoT technologies such as smart thermostats and smart switches are now available to automate demand-side response. However, our understanding of the latent flexibility that exists across appliances and consumers remains limited.

Understanding the role of automation and incentives in household electricity demand

In collaboration with Tata Power, one of India’s largest integrated electric power companies, we conducted a randomised controlled trial with residential electricity consumers in Delhi and Mumbai to study the role of automation and incentives in leveraging flexibility in household power demand. If demand can be made more flexible, we can avoid bringing on polluting power backups at times when there is not enough clean energy available on the grid, which would also lead to cost savings. In addition, demand flexibility could also help electricity suppliers avoid emergency response at times when margins are tight, lowering the likelihood of forced outages.

As part of the ongoing study, we offer customers simple Wi-Fi enabled smart switches, which control the operation of an appliance such as a room air conditioner (AC) or a refrigerator. Using the POWBAL web platform, which was developed at Imperial College London, we trigger brief automated switch-off events through the smart switch and reward participants per unit of electricity consumption avoided during those events. We randomise within users the timing of the events, the level of the financial incentive offered, and the amount of notice time. Users can opt out of individual events via a smartphone app linked to the smart switch. 

Automated switching off of electronic devices reduced household electricity consumption

As of August 2024, 1,050 devices have been installed, approximately 80% of which were connected to AC units, while the rest were mostly connected to refrigerators and electric geysers. A handful of customers also connected the smart switch to other types of appliances, such as air coolers, light bulbs, microwave ovens, washing machines, water pump motors, and electric car chargers. 

Our data shows that among the switch-off events conducted by us, the device was consuming electricity 38% of the times and was users then opted out of 16% of those events. We find that automated switch-off events reduced appliance usage by 60% during the event and by 69% within an hour of the event. We found 8.5% reductions in household power demand on average, which increases to a 15% in the evenings when aggregate power demand is the highest as shown in Figure 1. The absolute load reductions at the device level closely match those at the meter level, indicating that device readings may be enough to implement automated demand response programmes even among consumers that do not have smart meters without compromising the benefits to the grid.

Figure 1: Hourly impact of switch-off events on electricity consumption

Need for creating awareness about the benefits of automated load control programmes 

We find that providing a short two-hour notice period creates a perverse anticipation effect, where users manually turn off their device before the event begins expecting to maximise their rewards. However, given that households are opted into the event (and can opt-out if they wish), there is no need to take manual action to participate in switch-off events, and this behaviour reflects a misunderstanding of how rewards are earned. 

Giving a longer eight-hour notice period does not create a perverse anticipatory effect, suggesting that text alerts should be sent several hours in advance of the event or should not be sent at all. Given the novelty of automated load control programmes in this context, creating awareness about the uses and potential benefits of smart technologies for load balancing is important. 

Households are indifferent to price incentives and more efficient with automation

Households are rewarded in proportion to the power consumption that is avoided or the power their appliance would most likely have consumed if there was no switch-off event. We vary these reward rates throughout the trial, with the lowest offered rate equal to INR 6 per kWh, approximately the average wholesale price of electricity during this period, and the highest offered reward rate is five times that level. While a higher reward rate modestly reduces the probability of overriding the event entirely, we do not find any evidence that households respond more strongly to higher reward rates, indicating that offering volumetric price incentives higher than the average wholesale price of electricity does not deliver additional benefit when demand response is automated.

Finally, we do not find any evidence of compensating effects. The appliances connected to the smart switch consume no more electricity after the event than before the event started. This finding suggests that electricity may be used inefficiently by these households and highlights the potential of automated demand response to not only deliver benefits to the energy system but also to make household power usage more efficient.

Assessing the broader potential of an electricity demand management approach through a counterfactual analysis

While our study consists of an urban population, the analysis suggests that findings could apply to residential consumers broadly for a few reasons. 

1. Underlying differences in sensitivity to price incentives among small versus large electricity consumers matter less when the smart device turns off an appliance automatically and the user does not have to take any manual action to reduce consumption unless they want to override via the app (which also involves some degree of effort and attention). 
2. Even among the relatively more affluent urban consumers in our study, we do not see evidence of substitution to other household appliances due to switch-off events. One would expect that to also be true for poorer consumers who may not have other appliances they can use instead. 
3. We do not see any difference in the percentage reduction in electricity use during switch-off events across consumers in different quartiles of annual electricity use at baseline. 

To assess the aggregate potential of a POWBAL-like demand management approach, we conduct a counterfactual exercise. Using new estimates of the marginal emission factors of the Indian power grid and granular estimates of household-level responses, we examine how much carbon emissions of households could be reduced with the current set of generation assets. We find an average CO₂ emission reduction effect of around 2.3%, which increases to more than 3% in some households, and an average cost saving of 2.5%. Taking the device and installation costs of the smart switches (USD 24) into account, the average net CO₂ mitigation cost was -USD 23 per ton of CO_2, with a negative net carbon mitigation cost computed for nearly three-quarters of the households (see Figure 2).

Figure 2: Counterfactual analysis of net CO₂ mitigation cost

Way forward

Taken in the context of existing literature, our findings suggest that automation coupled with price incentives can achieve far more than what price incentives can achieve on their own, largely because automation reduces the cognitive burden on consumers to provide flexibility to the system. 

We conclude that household-level demand management can play an effective and economically meaningful role in the Indian electricity system. In future work, we will examine how a POWBAL-like demand management approach could interact with potential future investment pathways for generation assets. We will consider how demand management could reduce the system cost of adding intermittent renewable generation assets.