<p>Accurate short-term wind speed forecasting is crucial for the scheduling, management, and stable operation of power systems. Due to the high randomness of the marine environment and the instability of wind speed, this presents a greater challenge to the accuracy of offshore short-term wind speed predictions. To further enhance the accuracy of wind speed forecasting, this paper introduces a novel hybrid model for offshore short-term wind speed prediction, named TMI-ResTKAN. The model first employs Time-Variant Filtered Empirical Mode Decomposition (TVF-EMD) to decompose the original wind speed series. The bandwidth threshold and the order of the B-spline function in the TVF-EMD are then optimized using the Ivy Algorithm (IVYA) and the Maximal Information Coefficient (MIC). Subsequently, a three-layer one-dimensional residual convolutional network is used to extract features from the decomposed wind speed series. Finally, the extracted features are input into Temporal Kolmogorov-Arnold Networks (TKAN) for prediction, with the final forecasted wind speed for the next time step obtained by superimposing and reconstructing the results. To validate the performance of the model in offshore wind power short-term forecasting, simulation experiments were designed based on multiple hybrid models across three offshore wind farm datasets. The experimental results demonstrate that the TMI-ResTKAN model outperforms other hybrid models in prediction accuracy, with its MAPE showing improvements of 17.30%, 20.03%, and 27.51% compared to the baseline models across the three datasets, respectively. This confirms the predictive potential of the model in the field of offshore short-term wind speed forecasting.</p>