参数自适应VMD与CNN-BiGRU耦合的海平面变化预测

Sea Level Change Prediction Based on Coupled Parameter-Adaptive VMD and CNN-BiGRU Model

摘要: 月平均海平面预测是海岸带风险管理与海洋环境预报的重要依据。针对现有方法难以兼顾海平面序列多尺度非平稳特性与时序依赖关系、预测精度有限的问题，本文构建参数自适应VMD与CNN-BiGRU混合预测模型。基于南海北部闸坡验潮站实测数据，依托样本熵与灰狼优化算法自适应优化VMD参数，解耦非平稳海平面时序分量；利用CNN提取局部波动特征，结合BiGRU挖掘双向长程时序关联，实现月尺度海平面精准预测。试验结果表明，相较于传统熵准则，样本熵优化的VMD分解性能更优，预测均方误差降低15%以上、平均绝对误差降低9.3%。特征步长为6个月时CNN-BiGRU性能最优，可精准捕捉序列突变特征，全周期预测结果稳健，季风转换时段误差可控，可为海岸带防灾减灾与海洋环境预报提供技术支撑。

Abstract: Monthly mean sea level prediction is an essential basis for coastal zone risk management and marine environmental forecasting. Aiming at the deficiencies of existing methods in balancing the multi-scale non-stationary characteristics and temporal dependence of sea level sequences with limited prediction accuracy, this study constructs a hybrid prediction model combining parameter adaptive variational mode decomposition (VMD) and CNN-BiGRU. Based on the measured data from the Zhapo tide gauge station in the northern South China Sea, the sample entropy and grey wolf optimization algorithm are adopted to adaptively optimize VMD parameters and decouple non-stationary sea level time series components. CNN is used to extract local fluctuation features, and BiGRU is applied to explore bidirectional long-term temporal correlations to realize accurate monthly sea level prediction. The results show that VMD optimized by sample entropy outperforms traditional entropy criteria, reducing the mean square error by more than 15% and the mean absolute error by 9.3%. The CNN-BiGRU framework achieves optimal performance at an input step size of six months, effectively capturing sequence mutation features. The model presents robust full-cycle prediction with controllable errors during monsoon transition periods, providing technical support for coastal disaster prevention and marine environmental forecasting.