论文部分内容阅读
提出一种船舶横摇时间序列预测方法.该方法使用在隐层具有2个反馈权值的对角递归神经网络进行预测,给出了此网络易于实现的动量梯度学习算法(DBP),并对其收敛性进行了验证.运用该模型对我国某型船舶在横浪中航行情况进行预测,结果表明:本网络可以储存更多的历史数据,有更好的记忆性能,所使用的模型比DRNN模型及前向网络BP模型能快速、准确地预测船舶横摇运动时间序列,仿真实验验证了该方法的可行性与有效性.
A method of forecasting ship roll time series is proposed.The method uses the diagonal recurrent neural network with two feedback weights in hidden layer to predict and gives a momentum gradient learning algorithm (DBP) which is easy to implement for this network. Its convergence is verified.Using the model to predict the ship sailing in a certain type of ship in China, the results show that the network can store more historical data and have better memory performance, and the model used is better than DRNN The model and the forward BP model can predict the time series of ship rolling quickly and accurately. The simulation results show the feasibility and effectiveness of this method.