Toward Efficient Federated Load Forecasting: Personalization Mechanisms and Their Impact

This paper tackles the pressing challenge of load forecasting in smart buildings, aiming to enhance energy management in the light of the sector’s substantial energy consumption. Specifically, we focus on overcoming the issue of historical data being insufficient to train effective deep learning models. To this end, we introduce a federated load forecasting framework to exploit data from multiple buildings while maintaining the privacy of their load information, ensuring that data remains within each building’s local environment. Acknowledging the diverse load profiles shaped by factors like size, location, and user behavior, we investigate existing techniques to personalize the global model to accommodate each building’s specific characteristics. We perform a series of experiments across diverse datasets to compare and interpret the results of each personalization mechanism. Our findings indicate that federated learning significantly enhances forecasting accuracy across different buildings, with personalized federated learning providing even more substantial improvements.