Most of the current collaborative network risk studies focus on modelling risks in the project implementation phase, rarely using data analysis methods, and the identification of risks inevitably lags behind. This study addresses the problem of risk element identification in complex network scenarios, based on the grid feasibility study, and constructs a multi-layer network of risk elements based on unstructured data analysis. The network contains risk element layer, risk structure layer and collaborative network layer, which realises potential risk identification and prevention in the project reserve stage. The main research contents are as follows: 1) constructing the risk element layer using TF-IDF algorithm and co-occurrence analysis; 2) building the risk structure layer and completing the cross-layer association through text similarity and keyword matching; 3) establishing the collaborative network layer and completing the cross-layer association by leveraging semantic matching methods. Finally, a case study of a provincial grid company’s production technical reform project confirms the model’s validity and practical utility, which provides a scientific basis for risk management and responsibility allocation of power grid projects.