Smart meters can help citizens in optimizing energy consumption patterns. However, mixed evidence exists on their effectiveness in reducing energy demand and especially in levelling off the daily peaks of electricity load curves. Here, we evaluate the impact of providing real-time feedback on electricity consumption from a field trial in Italy. We combine standard regressions with machine learning techniques on high-frequency data to quantify impacts on both levels and patterns of electricity use. Results indicate that real-time feedback can moderately decrease electricity consumption (between 0.5 and 1.9% depending on model specification), but that it does not promote load shifting throughout the day by itself. Machine learning reveals evidence of significant household heterogeneity in the behavioral response.

智能电表可以帮助市民优化能耗模式。但是，关于降低能源需求，尤其是降低电力负荷曲线的每日峰值的有效性，存在混合的证据。在这里，我们评估了提供实时反馈对意大利现场试验所消耗电力的影响。我们将标准回归与针对高频数据的机器学习技术结合在一起，以量化对用电量水平和模式的影响。结果表明，实时反馈可以适度降低用电量（取决于型号规格，在0.5％至1.9％之间），但是它本身并不能促进负载整天转移。机器学习揭示了行为反应中明显的家庭异质性的证据。