Assessing Probabilistic Load Forecasting Accuracy Against Customer Privacy Constraints At a Low Aggregate Level.

Facilitating data sharing while upholding privacy is essential for driving meaningful insights, promoting informed decision-making, and fostering responsible data-driven solutions in today’s energy management. This is especially important in the context of load forecasting, where accurate predictions of energy consumption patterns are needed for efficient energy management. The existing models guarantee the protection of the load profile data of all households for low-aggregate level load forecasting. However, in the real world, some customers may be willing to fully share their data and others may not. Moreover, persuading some customers to give consent for direct use and full disclosure of their load profile data is not always possible. Accordingly, this paper introduces the concepts of full and limited consent levels- the former refers to the willingness of an individual to fully disclose their ground truth, otherwise, their privacy must be protected under the limited consent level while releasing the aggregate forecast. Moreover, this study contributes to the research gap on the intricate balance between utility and privacy in the domain of probabilistic low-aggregate load forecasting particularly when a community comprises various proportions of customers with full/limited consent levels. In doing so, a differential privacy (DP) model for short-term probabilistic load forecasting is used so that the user, such as grid operators or retail providers, receives the forecast retaining the 95% confidence interval (CI) of the unperturbed/original forecast. Furthermore, a Bayesian neural network (BNN) is utilized as the forecasting engine.