【作者】 李晖；

【导师】 郭晨；

【作者基本信息】 大连海事大学 ， 轮机工程， 2003， 博士

【摘要】 船舶减摇鳍是船舶与海洋工程中的一种重要系统,目前已在多种船舶中广泛应用。减摇鳍对于提高船舶耐波性,增加船舶使用寿命,改善设备与人员的工作条件,提高舰艇的战斗力具有重要作用。减小船舶横摇是目前船舶运动控制领域的重要课题之一。本文以船舶减摇鳍系统作为研究对象,对于船舶减摇鳍系统智能控制算法及其虚拟现实环境下的可视化仿真进行了系统深入的研究。 减摇鳍系统目前大多采用基于力矩对抗原理的PID控制器。控制器的性能对船舶自然横摇周期和无因次横摇衰减系数有着很大的依赖关系。由于船舶横摇运动的复杂性、非线性、时变性和海况的不确定性,经典PID控制难以获得满意的控制效果。采用先进的控制策略是解决这一问题的有效方法。 本文提出了基于逆模式小波神经网络的船舶减摇鳍系统自适应控制方法。由于小波神经网络结合了小波特有的多分辨特性和神经网络固有的自学习和自适应的能力,能够实现良好的动态非线性映射的特点,使得小波神经网络具有学习收敛速度快,逼近精度高、参数(隐层节点数和权重)的选取有理论指导,有效避免了局部最小值问题等优点。仿真实验表明减摇效果有较大的提高,能够克服传统PID控制器适应性差的缺点,具有较好的容错性和较强的适应非线性的能力。在仿真过程中引入一个常量信号乘以一个富含频率的信号、一个频率在期望输出附近的伪随机二元序列信号作为海浪作用于船舶的波倾角输入信号,目的是为了使所建横摇模型对各种不同的输入信号都具有很好的泛化能力。为了获取最佳的常量信号并能减少系统仿真量,增强系统适应性,针对不同海情依有义波高和风速不同可分为三个不同范围的特点,上述常量信号的大小只需取三个不同的值。仿真表明采用此方法后,对各种海况都能获得满意的效果。 由于船舶减摇鳍系统是一个关系到船舶安全、适航的实际工程系统,它除了需要控制精度高以外,还需要保障系统运行的可靠性。尽管神经网络控制在理论研究上取得很大进展,但目前大部分工作还只是处于仿真阶段,完全基于神经网络的控制应用到实际船舶工程系统中还待时日。而常规PID控制在船舶减摇鳍系统实际应用中已非常成熟,所以选择二者的结合不失为一个能有效解决船舶减摇等实际控制更多还原

【Abstract】 As an important system of marine engineering, ship stabilizing fins have been widely used in many different kinds of ships. Stabilizing fins have played an important role for improving ship behavior in waves, increasing operational life-span of ship, ameliorating work conditions of equipment and crew and advancing battle effectiveness of naval ship. Reducing ship roll motion is one of the important tasks of ship motion control. The researches on the purpose of intelligent control strategies and its visual simulation environment based virtual reality for ship stabilizing fin system are systemically investigated in the thesis.Generally, PID controller based anti-moment principle has been commonly adopted in ship stabilizing fin system. The performance of the controller depends mainly on natural period of ship roll motion and non-dimensional roll damping coefficient. Because of complexity, non-linearity, time-varying of ship roll motion and uncertainty of sea condition, satisfied control effect is very difficult to be obtained with conventional PID controller. The effective measure to solve the problem is that advanced control strategies should be introduced.The method of inverse mode wavelet neural network adaptive control based for ship stabilizing fin system is proposed in the thesis. Combining own multi-resolution characteristic of wavelet analysis and self-adaptation and self-learning ability and favorable dynamic nonlinear mapping characteristic of neural network, wavelet neural network possesses the advantages such as rapid learning convergence speed, high approximation accuracy, the choice of parameters (for instance, number of hidden nodes, value of net weights, etc.) based on theoretic guidance and effectively avoiding local minimum value. The simulation results show that the roll reducing effects are obviously improved, the method can overcome poor adaptability of conventional PID controller and provide better characteristics of fault tolerance and stronger nonlinear adapting ability. In the simulation studies, by introducing a signal consisting of a constant value multiplied by pseudo random binary signal as simulated input signal of wave slope angle, the built roll model gets better extensive ability for different input signals. In order to obtain the most optimal constant, avoid too heavy calculations and improve the system adaptability, the size of the constant signal can only be taken three different values since the different sea conditions can be divided into three different ranges according to the different wave height and wind speed. The simulation results indicate that satisfying effect is obtained for different sea conditions.As a practical engineering system closely related to navigation safety of ship, high control accuracy and reliability are necessarily required for ship stabilizing fin system. Although neural network control is made great progress in theoretical research, so far, most algorithms are still proved effective in simulation, the application of the control system to a practical ship completely based on neural network may take longer time. Since conventional PID control is widely adopted in the practical ship fin system, it is good choice to combine PID control with neural network to resolve the issues of practical ship fin system. Therefore, four intelligent control algorithms which integrate wavelet neural networks and PID control are presented in the thesis. In these control algorithms, neural networks are utilized in modeling of dynamic system, acted as process model or served as optimal calculation in traditional control system respectively, so choosing wavelet neural network can ensure the engineering requires such as high approximation precision, good real time characteristic, high reliability.Sea wave is the main cause of ship rolling, it is necessary to build an effective mathematic model to disturbing of sea wave for the research of ship roll reducing. Due to randomicity of sea wave, it is extraordinary difficulty to gain exact sea wave model. Thus a compositive prediction approach of sea wave based on wavelet decomposition and ANFIS model is presented in the thesis. Firstly, multilevel 1-D wavelet decomposition of irregular sea wave is completed to gain approximative regular period signals, thus, it can translate time series prediction of unregulated multi-period sea wave into time series prediction of relative simple and regular period signals. The movement trend and detail information of sea wave can be observed, the difficulty of predication can be reduced obviously. Next, the main regular period signals are predicated by using the ANFIS predication model, the final predication value can be gained by integrating the outputs of ANFIS. The simulation results illustrate the effectiveness of the method, high accuracy is acquired.Because of the complicated mechanical mechanism, serious non-linearity and time-varying of ship roll motion, precise model is hard to be obtained, the modeling and predicating methods of ship rolling time series prediction based on wavelet neural network is proposed in the thesis. Wavelet neural network combining wavelet transform time frequency localization with self-adaptation and self-learning ability and stronger robustness and generalization ability of neural network, it breakthroughs the spectrum analysis method of traditional singleness frequency. Comparing the experiment results with the simulation results based on RBF and Multi-layers Perceptron, it demonstrated that with the same network structure and the same number of hidden nodes and the sameiterating times, wavelet network possesses the characteristics of faster training and convergence speed, especially stronger approximation ability in situation of curve saltation, better predication precision has been obtained.In order to obtain realistic dynamic effect for the simulation of control for ship stabilizing fin system and realize the natural interaction between users and environment, visual simulation environment for ship stabilizing fin system is developed by using MATLAB Virtual Reality Toolbox and Virtual Dials & Gauges Blockset. It includes: ? virtual reality visual simulation of ocean waves is completed on the basis of wave spectrum analysis; ? 3-D model of ship is established by using VRML language; (D realistic-looking instruments panel is designed by simulating instruments panel of a real ship; ? the integrated system of virtual waves, virtual ship model and virtual instruments panel is set up by using MATLAB Simulink control simulation environment, the triune simulation effect is realized.更多还原