Privacy-Preserving User-Side Load Forecasting

Accurate user load forecasting can reduce power deviation, which directly impacts the revenue of retail electricity companies. Existing user-side load forecasting methods are predominantly centralized, overlooking the privacy of users' data. Moreover, current models lack the capability to effectively capture long-range dependencies in load sequences, thereby limiting forecasting accuracy improvements. To address these challenges, we propose a user cluster load forecasting method that combines federated clustering with an enhanced Informer model. The method first employs federated clustering to group users, preserving data privacy while capturing correlations in electricity consumption patterns. Subsequently, we construct load forecasting models for each user group using the enhanced Informer. This improved model incorporates dilated causal convolutions to capture long-range dependencies in load sequences and utilizes Wasserstein metric to measure the discrepancy between attention probability and uniform distributions. Case studies demonstrate that the proposed method outperforms baseline models in forecasting accuracy while simultaneously preserving user data privacy.