Federated Learning Based Collaboration Framework of Data Sharing for Intelligent Design of Residential Buildings

The fast-developed artificial neural network (ANN) technique has been used in the architecture, engineering, and construction domain to facilitate the designing process, such as layout planning and structural designing. However, it is difficult to collect sufficient drawings from data owners for privacy reasons and due to the lack of proper incentive mechanism. To address the challenges, this paper proposes an incentive collaboration framework based on federated learning and game theory to enhance data sharing for intelligent design of building projects. Federated learning can train ANN models on data owners’ local devices and thus avoid directly publicising their drawings. A collaboration matching mechanism is designed based on game theory to encourage drawing owners to publicise their data. A case study is conducted using drawings of residential buildings to validate the feasibility of the proposed framework. Results show that (1) Data owners who provide data with higher quality and larger quantity can trade for more data in collaboration; and (2) The ANN model trained through federated learning on larger quantity of drawings performs better than the one trained on local data. Practically, the innovative framework should facilitate the development of intelligent design models with higher performance to assist building designers. Theoretically, the combination of federated learning and game theory could enhance the knowledge of addressing the data sharing dynamics and complexity in innovative construction such as modular building.