Publications

Smart buildings are increasingly equipped with a multitude of sensors and IoT devices, generating vast amounts of data. Generating a common global efficient model for predicting energy consumption accurately in these environments is crucial for optimizing energy usage, reducing costs, and achieving sustainability goals. However, there are limitations in terms of communication resources and data privacy. Hence, Federated Learning (FL) has emerged as a promising solution to address the challenges of privacy, data decentralization, and scalability in such complex systems. However, FL can be challenging to optimize, as the parameters of the global model need to be tuned to achieve the best performance on a diverse set of devices (called clients). In this paper, we propose to combine the Genetic Algorithm (GA) with the Particle Swarm Optimization (PSO) algorithm to optimize FL for smart building energy consumption prediction. Experiments show that our proposed method increases the energy consumption prediction performance and reduce the Root Mean Square Error RMSE by 7.7% in comparison with the local models.