基于CNN-GRU模型的USV运动姿态多步预测方法

Multi-step motion attitude prediction method of USV based on CNN-GRU model

针对单一模型在预测无人水面艇(Unmanned Surface Vessel,USV)运动姿态时精度不高的问题,提出一种基于卷积神经网络(CNN)和门控循环单元(GRU)的USV姿态多步预测模型。首先,使用滑动窗口法构造运动姿态数据集作为模型输入;然后,使用CNN模块挖掘时序数据的局部特征;最后,使用GRU模型进行多步预测。使用实测USV运动姿态数据进行预测实验,实验结果表明,该模型比XGBoost模型、单一LSTM模型和单一GRU模型具有更高的预测精度,各项评价指标表现更佳,具有重要的应用价值。

Aiming at the problem of low accuracy of single model in predicting the ship motion of Unmanned Surface Vehicle(USV),a multi-step prediction model based on Convolutional Neural Network(CNN)and Gate Recurrent Unit(GRU)is proposed.Firstly,the sliding window method is used to construct the motion data set as the model input.Then,the CNN module is used to mine the local features of time series data.Finally,the GRU network is used for multi-step prediction.The experimental results show that the model has higher prediction accuracy than XGBoost model,single LSTM model and single GRU model,and its performance of each evaluation index is better,which has important application value.