Abstract: Objectives Single physical models struggle to adapt to complex propulsion profiles and nonlinear aging characteristics of power batteries in all-electric ships during long-endurance missions. Meanwhile, the data-driven approach is limited by the issue of the limited computing resources available on the onboard BMS.. To address these issues, this paper proposes a ship-shore collaborative State of Charge (SOC) estimation framework based on fuzzy adaptive weights. Methods A second-order RC equivalent circuit model is adopted at the end (edge) of the ship, combined with Particle Swarm Optimization (PSO) for parameter identification and an Adaptive Cubature Kalman Filter (ACKF) for low-complexity real-time state tracking. At the shore-side (cloud), a Transformer-based PatchTST deep learning model is deployed, utilizing its patching mechanism and channel-independence strategy to capture nonlinear features and aging trends from long-sequence operational data. A fuzzy logic controller is designed to dynamically adjust the fusion weights between the ship and shore estimations, using voltage residuals and current change rates as inputs. Results Experiments based on the CALCE dataset, simulating complex ship maneuvering load profiles, demonstrate that the collaborative method effectively overcomes the limitations of physical models in low SOC regions and under drastically fluctuating conditions and the issue of aging parameter mismatch. Compared with the standalone shipborne ACKF method, the collaborative estimation reduces the Root Mean Square Error (RMSE) by 29.44% and the Mean Absolute Error (MAE) by 35.71%. Furthermore, under an extreme initial error of 30%, the convergence time is shortened from 153 s to 13 s, significantly enhancing system robustness and convergence speed. Conclusions This architecture provides an effective technical pathway for achieving high-precision ship-shore integrated state estimation, remote outlier fault diagnosis, and online parameter Condition-Based Maintenance (CBM).