Objective Aiming at the operational bottleneck of far-offshore wind power O&M under high sea conditions (wave height: 2.5–3.5 m), this study addresses the problems of conventional wave compensation trestles, such as low prediction accuracy, delayed response, and insufficient sea state window utilization.

Methods Taking a far-offshore wind farm in Guangdong as the research carrier, an adaptive compensation algorithm based on Fractal Reinforcement Learning (FRL) is proposed by integrating wave fractal self-similarity and Deep Deterministic Policy Gradient (DDPG) algorithm. A six-degree-of-freedom (6-DOF) hydraulic trestle system integrated with feedforward-feedback dual-mode control is developed. Its performance is verified through six-month full-scale field experiments (covering typhoon season) and ablation experiments.

Results  The results show that under high sea conditions, the system's compensation deviation is stably within 0.5 m, with a response time of only 0.08–0.09 s. There were no personnel boarding accidents; the sea state window utilization rate increased to 81.0% (9.1 times higher than that of conventional systems), and the personnel transfer efficiency reached 31 persons/day (3.4 times higher). The 5-year static Return on Investment (ROI) is 338.1%, with a payback period of 0.98 years.

Conclusions This scheme constructs an integrated "algorithm-system-verification-benefit" architecture, which can realize the unity of accuracy, reliability and economy. It provides high-robustness technical support for far-offshore wind power O&M and has universal reference value for the adaptive control of marine engineering equipment.