Objective Offshore wind farm operation and maintenance (O&M) is subject to complex marine environmental conditions, including wind, waves, and tides. Traditional manual scheduling suffers from low efficiency, high O&M costs, and substantial risks of offshore operations that violate safety requirements. Existing scheduling algorithms struggle to handle multiple constraints while achieving global optimization. To meet the intelligent scheduling needs of a 300 MW offshore wind farm, this study develops a multi-source data-driven joint optimization model for O&M scheduling.

Method Four types of heterogeneous data, namely meteorological data, marine environmental data, vessel records, and fault work orders, were integrated and preprocessed through data cleaning, smoothing, and structuring. A multilevel procedure was established to calculate weather windows suitable for offshore operations. A bi-objective mixed-integer programming model was formulated to minimize total O&M costs and maximize operation time utilization, with explicit constraints on meteorological conditions, vessel capacity, task sequencing, and complete task coverage. A hierarchical LNS-MIP-PSO algorithm was developed. Large neighborhood search (LNS) performed coarse-grained neighborhood exploration, mixed-integer programming (MIP) determined a feasible baseline schedule subject to hard constraints, and particle swarm optimization (PSO) refined vessel departure times and navigation routes to correct travel-time deviations caused by ocean currents and waves. A closed-loop parameter iteration mechanism was designed, and multi-scenario simulations and baseline comparisons were conducted. Independent-samples t-tests were used to assess the statistical significance of the results.

Result A 300 MW offshore wind farm comprising 50 turbines rated at 6 MW each and three O&M vessels is used as the case study. Thirty sets of test samples are generated from actual fault work orders recorded at the wind farm from 2024 to 2025. The coverage and accuracy of data preprocessing reach 95.0% and 98.0%, respectively. Compared with standardized manual greedy scheduling, the scheduling time is reduced from 120 min to 30 min, and operation time utilization increases from 70.2% to 85.1%. Labor costs decrease by 30.1%, and total O&M costs decrease by 25.3%. The proportion of offshore operations violating meteorological thresholds decreases from 20.3% to 12.2%, a relative reduction of approximately 40%. Compared with standalone MIP and PSO algorithms, the proposed hybrid algorithm reduces total costs by 11.6% and 14.2%, respectively. Sensitivity analysis indicates that the wind speed threshold and vessel drag coefficient have the greatest influence on scheduling outcomes, and that the model parameters have identifiable ranges suitable for engineering applications. The t-test results show that all optimization indicators are statistically significant (P < 0.05).

Conclusion The proposed data-driven joint optimization model effectively accommodates the multiple constraints of offshore O&M operations. The hierarchical LNS-MIP-PSO algorithm combines global search capability with constraint feasibility, reducing overall O&M costs and the risk of noncompliant operations. The model features adjustable parameters and reproducible results, making it suitable for engineering implementation. Future research may integrate real-time internet of things (IoT) data and reinforcement learning to enable dynamic, real-time optimization of offshore O&M scheduling.