Development of smart cities is now in a stage of deep digitalisation and all-encompassing intelligence. The multiple types of data produced during urban operation have large scales, are multi-modal, are changing rapidly, and come from various sources. The traditional data governance framework has been facing several problems, such as dispersed data collection, isolated storage systems, inefficient processing capabilities and shallow application implementation; it is thus unable to meet the high standards of timely, smart and intelligent governance in a smart city. Large-scale models have shown good performance in multimodal understanding, autonomous learning, deep reasoning and generative optimisation, and can provide strong technical support for addressing data governance bottlenecks and reconstructing end-to-end governance processes in smart cities. Assess the present state of multi-dimensional data governance in smart cities, identify the principal shortcomings of the traditional governance model, and present numerous merits that large-scale models offer for supporting data governance. Six representative links in this study are selected: data collection and aggregation; cleaning and preprocessing; integration and sharing; analysis and mining; security management; and application implementation. It seeks to build a closed-loop, intelligent and integrated data governance system, address the implementation problems and optimisation paths of this system, and thus provide theoretical support and technical reference for high-quality smart city construction, efficient data element circulation, and modernisation of urban governance.
