Artificial dataset generation to enhance the design exploration of residential buildings through data-informed energy load forecasting models

This study aims to assist urban planners and building designers in taking informed decisions based on energy performance – simulating a real-world urban development scenario – using limited computational resources. In particular, this paper proposes a new approach that integrates existing studies on building loads forecasting by using a Generative Adversarial Network (GAN) generated dataset based on significant geometrical parameters. This overcomes the needs for large datasets – often difficult to access.The results demonstrate that the data-driven approaches have addressed the buildings' load predictions with a reasonable accuracy while significantly reducing the calculation time required.