On the effectiveness of residential involuntary load curtailment programs

ABSTRACT

The critical peak hour problem is a fundamental issue for utility managers and policymakers. We address this issue by analyzing unique data from an involuntary load curtailment program, in which a group of households agreed to reduce the peak power consumption below the agreed load shedding limit. There are two types of results presented in the paper. Firstly, we show that residential customers reacted by lowering usage in critical peak hours by 6.30%, on average, and that 93.69% of the total consumption reduction is transferred to off-peak (earlier and later) hours on the same day. This decrease in usage is comparable with the effects of time of use (TOU) programs. Secondly, we identify that customers whose electricity usage was regularly below the load shedding limit also reduced their usage. We discuss such reactions to this non-binding restriction using a microeconomic approach and in the context of findings from behavioral science.

KEYWORDS:

• Load curtailment
• demand-side response
• energy consumption
• behavioral traits
• non-binding constraints
• utility function

Acknowledgments

The authors would like to thank Energa-Operator SA. for giving access to documentation of “Consumer Test in Kalisz” under agreement no. CJ00137/14 concluded with Energy Design Sp. z o.o. Sp. k. on May 9, 2014. We are also very thankful to Timothy Harrell for his advice concerning matters of English usage. Marcin Czupryna was supported by statutory means of the Cracow University of Economics.

Notes

¹ Märkle-Huß, Feuerriegel, and Neumann (2018) analyze the impact of demand response on electricity spot prices and load. D’hulst et al. (2015) list four reasons for more flexibility in the electricity grid. Firstly, the share of renewable resources in the energy mix increases. Secondly, these resources are quite decentralized geographically, leading to a local mismatch between demand and supply. Thirdly, the share of energy from traditional power plants is stagnating or even decreasing, resulting in greater volatility in energy production. Since energy storage facilities (batteries) are currently very costly, the possibilities of stabilizing the energy supply through some short-run supply-side interventions are very limited. Finally, anticipated changes in transport and heating will increase demand for electricity in the not-too-distant future.

² They wrote “ … residential demand management solutions – including direct load control technology that allows utilities to turn specific household appliances on and off during peak periods – are becoming increasingly important. While such technology has been available for decades, acceptance and adoption among residential consumers has not always kept pace … ” (p.76)

³ The data from the TOU component were previously analyzed in Morawski et al. (2018)

⁴ According to the literature, see e.g. Bertrand and Mullainathan (2001), the results obtained by observing actual behavior are more meaningful than those based solely on stated intentions. The number of households in a sample was limited due to the high cost of the experiment. Nonetheless, the number of households in a sample is sufficient to obtain statistically significant results.

⁵ As a reference level, the estimated Value of Lost Load (VoLL) in Poland is 12,15 EUR/kWh, see citet(shivakumar2017valuing)

⁶ Households equipped with 3-phase connections were offered PLN (Polish currency unit) 20 and PLN 30, respectively, in RED I and RED II. There were no statistically significant differences between participants’ behavior in these two groups. Therefore, both groups were combined in all the analysis with the exception of the regression analysis, which is presented in Eq. 10(10) $d{d}_{i,d}={\beta }_0+{\displaystyle \sum _{j=1}^{12}{\beta }_j{Q}_{i,j}+{\beta }_{13}{D}_i+{\beta }_{14}{\overline{e}}_i^c+{\beta }_{15}d+u_i+{\epsilon }_{i,d}}$(10) . However, there were very few households with such connections – 7 out of 125 in RED I and 8 out of 126 in RED II. This feature is ignored in the current reassessment for the sake of simplicity.

⁷ Recruiting of participants started in May 2013. We chose the months before recruitment began to exclude the impact of the recruitment process on the results obtained. Additionally, only working days were taken into consideration and reduction days were excluded from the calculation of averages.

⁸ Mahalanobis distance was applied

⁹ In the case of the difference-in-differences method, a PSM (propensity score matching) procedure will also be valid under a weaker assumption. Namely, if we assume that only the changes of observed variables (energy consumption) are conditionally independent given the propensity score.

¹⁰ We have also calculated the classical difference between the treatment and control groups on the reduction days and the Abadie-Imbens error, (Abadie and Imbens 2006). The Abadie-Imbens errors are used in the PSM procedure in a standard way. As these results are similar they are not reported here.

¹¹ In such a case, $u(s)$ is a log-normal distributed random variable.

¹² After we had removed the observations with missing values and the lowest values of observed mean electricity usage $m$ (to increase the robustness), 655 observations remained.

¹³ We used this method to take into account the repeated measurements for the same participant on different days in the data

¹⁴ p-value here refers to the factor: the type of premises in which the participant lives as a whole and not to a factor value