Frequency security issues often occur in a small number of outlier extreme scenarios, such as when large-scale wind and solar power generation leads to fewer conventional synchronous units being online, while existing planning method focuses on typical scenarios and overlooks extreme ones. This paper realizes the identification of adaptive time-interval extreme scenarios based on the PELT algorithm, constructs frequency security constraints using a multi-machine system frequency response model that incorporates wind, solar, thermal, energy storage, UHVDC, and demand-side response. On this basis, a three-layer iterative source-grid-storage collaborative planning model is established, including an optimal configuration master problem model, a typical scenario operation sub-problem model, and an extreme scenario operation sub-problem model. Model solving is achieved through methods such as MATLAB Simulink simulation, cut function feedback, and L-shape decomposition coordination. The method optimizes the construction plans for generators, transmission lines, and energy storages at the same time. Finally, the effectiveness of the proposed method is verified through a case study of an actual regional power grid, and the impacts of extreme scenarios, load, and new energy factors on the planning results are analyzed. This method ensures the frequency security of power systems in extreme scenarios by increasing a small amount of investment and optimizing unit operation, and can provide suggestions for the development boundaries of new energy.