基于RF-WOA-VMD-BiGRU-Attention的神经网络模型在海浪预测中的应用

海上风电场的海况数据极其复杂导致用于海浪高度预测的输入参数极其不稳定，筛选出关键信息，提高输入参数的质量可以极大地提高海浪高度预测的准确性。以乐亭菩提岛风电场近一年的海上数据为基础，构建了一种基于随机森林(random forest, RF)、鲸鱼优化算法(whale optimization algorithm, WOA)、变分模态分解(variational mode decomposition, VMD)和双向门控循环单元(bidirectional gated recurrent unit, BiGRU)的海浪预测模型。该模型利用随机森林对环境特征等输入变量进行筛选，有效减少数据冗余，然后基于WOA-VMD模型自适应确定最优参数和自适应分解原始序列，提高数据质量并消除数据噪声的干扰。此外，针对海浪预测提出了一种基于注意力机制优化的BiGRU算法，随机森林的注意力机制将为BiGRU的隐藏层状态分配不同的权重，加强关键信息的影响。实验结果表明该模型和其他模型对比，输入质量更高，预测精度更高，拟合程度更准确，对风电场海浪预测有着重大意义。

Application of neural network model based on RF-WOA-VMD-BiGRU-Attention in wave prediction

The extremely complex sea state data of offshore wind farms makes the input parameters used for wave height prediction extremely unstable, and screening out key information and improving the quality of input parameters can greatly improve the accuracy of wave height prediction. Based on the offshore data of the wind farm in Laoting Bodhi Island for nearly one year, a wave prediction model based on Random Forest, Whale Optimization Algorithm, Variational Mode Decomposition and Bidirectional Gated Recurrent Unit was constructed. The model uses Random Forest to screen input variables such as environmental characteristics to effectively reduce data redundancy, and then adaptively determines the optimal parameters and adaptively decomposes the original sequence based on the WOA-VMD model to improve data quality and eliminate the interference of data noise. In addition, a BiGRU algorithm based on Attention Mechanism optimization is proposed for wave prediction, and the Attention Mechanism of Random Forest will assign different weights to the hidden layer state of BiGRU to strengthen the influence of key information. The experimental results show that compared with other models, the model has higher input quality, higher prediction accuracy and more accurate fitting, which is of great significance for the prediction of wind farm waves.