Evaluating Probabilistic Load Forecasting in Predicting Long-Term Household Savings from Electricity Rate Plan Recommendations

Electricity rate plan recommender systems (EPRS) are tools that aim to predict and recommend which rate plans will be the least expensive for a household. In addition to its rate plan recommendation, EPRS could be greatly improved if they also generated probabilistic forecasts for the savings the household might incur if enrolled. Probabilistic load forecasting could be used to forecast cumulative usage costs and potential savings. However, load forecasts are generated from imperfect models, and so cumulative savings forecast distributions will also be imperfect. In this paper, we apply probabilistic load forecasting to generate cumulative savings forecast distributions for EPRS recommendations and show how load forecasting error affects savings forecasts. We show that these effects differ in severity over forecast horizons of a day, between months in the year, and over the full year. We show across 865 real households and 12 months that inaccuracies in forecasted load distributions can form bias in short-term savings forecasts but also compound, worsening the forecasted cumulative savings distributions. We conclude that forecasting cumulative savings from EPRS recommendations using probabilistic load forecasting has potential but requires careful consideration or mitigation of forecast bias induced by the nonstationarity of load.