This paper proposes a multimodal decoupling-based meta-learning prediction model for the cost of prefabricated buildings (MMDE-CMAF-MAML), aiming to address the issues of multi-source heterogeneous data fusion and insufficient generalization ability in the cost prediction of prefabricated buildings. The model extracts independent modality features through a multimodal decoupling encoder (MMDE) and dynamically allocates modality weights using a cross-modal attention fusion module (CMAF). It then achieves rapid adaptive prediction by combining the model-agnostic meta-learning (MAML) framework. Experimental results show that the model outperforms existing methods in terms of prediction accuracy, with a reduction of about 43% in MSE, as well as in generalization ability and robustness, providing an effective data-driven solution for the cost management of prefabricated buildings.
Houxiang Zha (Thu,) studied this question.