Abstract: With the restructuring of the energy structure and the proposal of the dual carbon goals carbon peaking and carbon neutrality, short-term power prediction for distributed photovoltaic (PV) systems has garnered extensive attention. High-precision power prediction can provide critical support for the optimal dispatch of distributed power stations. Considering that errors are unavoidable in the prediction process, and the preliminary prediction error sequence is generally characterized by coupled multi-scale features and high non-stationary randomness, which further impair prediction accuracy, this paper proposes an error correction method for short-term power prediction of distributed PV systems based on Sequential Variational Mode Decomposition (SVMD) and the Mamba model. The proposed method first employs SVMD to adaptively decompose the preliminary prediction error sequence into several modal components with distinct central frequencies, thereby effectively decoupling multi-scale features. Subsequently, the Mamba model is utilized to predict the decomposed error sequence. By virtue of its state space modeling capability with linear complexity and input-adaptive selective mechanism, Mamba can accurately capture the temporal dependencies of diverse input features and achieve precise prediction of the error sequence. Finally, the predicted error compensation values are superimposed onto the preliminary predicted power to correct the initial prediction results. Simulations are carried out using actual measured data from a distributed PV power station in a southern Chinese city to verify the effectiveness of the proposed error correction method for short-term distributed PV power prediction. The simulation results show that the proposed method presents significant superiority in enhancing the accuracy of preliminary prediction and possesses certain engineering application value.