【作者】 李占英；

【导师】 王科俊；

【作者基本信息】 哈尔滨工程大学 ， 模式识别与智能系统， 2012， 博士

【摘要】 船舶在海浪中会产生六自由度摇荡的复杂运动,具有很强的随机性和非线性,因此船舶摇荡运动预报对于船舶航行有着重要的意义。目前由于对海浪中船舶摇荡运动的机理认识不清,故实船摇荡的时域预报仍限于10秒之内,制约了其应用。本论文立足于舰船的横摇运动预报,通过发掘横摇运动中所蕴含的非线性动力学特性,证实其可预报性及混沌特性,以期有效提高预报精度和增加预报时长,使之能用于船舶航海实践。主要完成的工作有：1.分析船舶横摇运动时间序列的可预报性及混沌特性。建立了基于相空间重构理论的前向神经网络和递归神经网络的预报模型,使网络本身融入了混沌的确定性规则,提高了神经网络用于船舶横摇运动时间序列预报的效果。将混沌增加到网络中,除了上述方法外还可以选用混沌神经网络来映射船舶横摇时间序列所蕴含的混沌特性,进行预报,本文针对一种混沌神经网络进行优化修正后将其用于船舶横摇预报。但与基于相空间重构的神经网络预报相比,该网络预报精度低,而且目前对于混沌神经网络的研究尚不成熟,实现起来比较困难,而基于相空间重构的预报方法简单可行。2.在对角递归神经网络研究的基础上,优化了二阶对角递归网络参数,提出了基于相空间重构的对角递归和二阶对角递归网络的预报模型,并用于船舶横摇预报,预报效果好于未重构的网络,且优化后的二阶对角递归网络的具有更好的预报效果。3.针对递归网络训练复杂而且存在记忆渐消等问题,提出了一种新的预报方法—使用回声状态网络进行船舶横摇运动预报,该方法有效预报时间能达到17秒以上,且比已有的预报方法的预报精度要高近4倍。4.针对船舶在海浪中行驶的高度复杂性,传统的单一预报方法自适应能力较差,提出了运用了非负约束的冗余方法和协整理论方法对模型进行筛选的组合预报,并给出单项模型的筛选过程,避免了目前仅凭个人经验对模型进行选取。达到提高预报精度的目的。最后给出了船舶摇荡运动预报的性能指标及评价准则,对本论文中所用的方法进行了定量的评价对比。证实本论文所提方法的有效性。更多还原

【Abstract】 Ships in the waves will have complex movements of six degrees of freedom, which has strong randomness and the non-linearity. So the prediction of ship sway motion for ship navigation has an important significance. Currently, due to unclear understanding of the ship motion mechanism in waves, the prediction of time domain for ship sway motions is limited to10seconds, hereby restricts its application. This paper based on the ship roll motion prediction aims to find out the nonlinearity dynamics held in the ship roll motion and confirm the characteristics of predictability and chaos, in order to improve the accuracy of forecast and prolong the time length of prediction, which can be used in the practice of ship navigation. The main research work is as follows:1. This paper proposed that the ship roll motion of time series is a chaotic system through the analysis of its characteristics of predictability and chaos. The feedforward neural networks and recurrent neural networks prediction model based on chaos theory are introduced and it makes the network possessing a chaotic deterministic rules, which improves the accuracy of prediction for ship roll motion by the application of neural network. In addition to the above methods of the chaos added to the network, it can also use the chaotic neural network to map the chaos characteristics rooted in time series of ship rolling motion to predict. This paper introduces a chaotic diagonal recurrent neural network(CDRNN), which is used to ship rolling forecast after optimization and correction. However, compared with neural network based on phase space reconstructionprediction, the prediction accuracy of this network is low. Moreover, the current chaotic neural network for the study is not mature yet and more difficult to achieve. While the method of prediction based on phase space reconstruction is simple and feasible.2. Based on the research of diagonal recurrent neural network (DRNN), optimizing the parameters of second diagonal recurrent neural networks(SDRNN). The models of DRNN and SDRNN based on phase space reconstruction are proposed to forecast the ship rolling motion and results are better than non-reconstruction network. The optimized SDRNN is better than SDRNN in ship rolling motion prediction.3. Because the training for recurrent networks complex and issue of memories fading exists, a new method is going to be brought forward that is used echo state networks(ESN) to predict the ship rolling motion. This method can effectively forecast time of17seconds more and accuracy of prediction is4-fold higher than other existing means.4. As the highly complex nature of ships sailing in the waves, the traditional single prediction method has a problem of poor self-adapt. One way by using redundant of non-negative constraints and co-integration theory for screening the model is proposed to carry on a combination forecasts, and giving a single model of the screening process to avoid the current selection of model by personal experience. It achieves the purpose of improving the prediction accuracy. At last, the performance index and evaluation criteria are given for ship sway motion prediction. And the methods in this article are used to evaluate by quantitative contrast. The validity of proposed method in the paper is confirmed.更多还原