Navigating the Accuracy-Privacy Dilemma in Load Forecasting with Noisy Data

Accurate load forecasting facilitates the system operators to manage the power grid effectively. However, including noise to safeguard user privacy can adversely affect forecast precision. This study investigates applying differential privacy (DP) mechanisms to balance data privacy with forecasting utility. We introduce a quantitative method to evaluate the impact of DP-induced noise at varying privacy levels on load forecasts. Through simulations using linear regression, support vector regression, and artificial neural network models, we showcase the effectiveness of our approach across diverse forecasting methodologies. Our results offer valuable insights into achieving an optimal equilibrium between privacy preservation and forecasting accuracy, enabling various stakeholders in the electricity sector to navigate the accuracy-privacy dilemma when making data-driven decisions.