【作者】 江娜；

【导师】 刘胜；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 船舶的航向控制是一个复杂而重要的问题,如果船舶航行时航向控制系统出现故障,就可能造成重大的影响,甚至会造成人员的伤亡,所以仅在故障发生后进行故障诊断是远远不够的。船舶航向控制系统故障预报技术的开发,可以避免不必要的停航,防止故障进一步发展,并且可以提前做好准备,缩短检修时间,对于获得最大的经济效益具有积极的意义。因此进行船舶航向控制系统故障预报技术研究具有非常重要的意义。本文采用针对小样本问题的支持向量机算法(Support Vector Machines,SVM),对船舶航向控制系统的故障预报技术进行研究。在分析船舶航向控制系统故障模式及故障原因的基础上,构建了该系统的故障树模型,采用下行法对故障树进行了定性分析,并对船舶航向控制系统进行了故障建模与仿真。推导出了基于支持向量机的时间序列预报方法,并将该方法应用于sinc函数和Lorenz混沌映射这两个时间序列的预报中,同时将移动平均法和指数平滑法的预报结果与之进行比较,仿真结果表明支持向量机回归算法能获得较高的预报精度。在分析系统故障发生和演变过程的基础上,建立了基于支持向量机回归算法的故障预报模型。针对将支持向量机应用于故障预报时,故障特征选择优化、加权SVM的加权系数优化、SVM的参数优化是相互关联的,而不是孤立的,提出将三者并行优化的方法;将动物的捕食搜索策略引入到基本鱼群算法中,形成了一种改进人工鱼群算法,并用这种改进的鱼群算法并行优化。仿真结果验证了并行优化方法和改进鱼群算法的有效性和优越性。针对支持向量机在故障预报时应同时追求训练精度高和训练速度快,提出了从多目标优化的角度对SVM算法的这两个性能指标进行综合考虑,并采用直接对多个目标同时优化的方法来求解Pareto近似解集;将免疫算法引入鱼群算法中,形成了改进的免疫鱼群算法,并采用改进的免疫鱼群算法来求取Pareto近似解集。仿真结果验证了多目标优化方法和改进的免疫鱼群算法的有效性和优越性。针对支持向量机回归算法中,核函数对其回归效果的影响,提出选用多项式核函数、径向基核函数和Sigmoid核函数构造不同的SVM算法,建立了三种船舶航向控制系统故障预报的SVM单预报模型,并根据组合预报原理建立了基于小波网络的SVM组合预报模型,仿真结果验证了SVM组合预报模型的优越性。对预报出的航向角偏差进行统计分析,并将统计值与给定阈值作比较,实现了船舶航向控制系统的故障预报。以所建故障树为依据建立专家知识库,在C++Builder 6.0语言环境下实现了故障预报的可视化。更多还原

【Abstract】 Ship heading control is a complex and important issue. If the heading control system presents malfunction during navigation, it possibly have serious influence, and even causes casualties. Therefore, it is not enough to have the failure diagnosis only after the malfunction. The development of heading control system fault prediction technology could avoid the nonessential suspension, prevents further exacerbation of malfunction, and make the preparation ahead of time to reduce the overhaul time and maximize the economic efficiency. So the study on fault prediction technology of ship heading control system has vital significance. The fault prediction technology of ship heading control system, using SVM with small samples, is investigated.Based on the analysis of ship heading control system fault mode and the cause of the fault, the fault tree model of the system was established which uses downlink method to carry on the qualitative analysis. And it carries on failure modeling and simulation on ship heading control system.Time-series prediction method based on support vector machine was derived. It was applied on the prediction of two time series, sinc function and Lorenz chaotic mapping, and compared with the results of moving average and exponential smoothing method. The simulation results show that support vector machine regression algorithm can get a better predicting precision. On the basis of analysis system malfunctions and the evolutionary process, fault prediction based on support vector regression algorithm model was established.Considering of fault characteristics optimization, the weighted coefficient of weighted SVM optimization and the parameters of SVM optimization are interrelated rather than isolated when Support Vector Machine was applied to fault prediction, the parallel optimization method was proposed; predatory search strategy of animals was introduced to the basic artificial fish swarm algorithm in order to create an improved artificial fish swarm algorithm, which was used in parallel optimization.The simulation results show the effectiveness and superiority of the parallel optimization method and the improved artificial fish swarm algorithm.In the pursuit of training accuracy and training speed on Support Vector Machine for fault prediction, the two SVM algorithm performance indicators were considered from the perspective of multi-objective optimization, and the Pareto approximate solution set was solved using the direct multiple targets optimized method; immune algorithm was introduced into fish swarm algorithm to form the improved immune fish swarm algorithm. And the Pareto approximation solution set was gained by using the improved immune fish swarm algorithms. The simulation results show the effectiveness and the superiority of the multi-objective optimization method and the improved immune fish swarm algorithm.Consider that regression results of different kernel functions are different in support regression algorithm, different SVM algorithms, using polynomial kernel function, RBF kernel function and Sigmoid kernel function, were constructed to establish three SVM-forecasting models of ship heading control system fault prediction, and SVM combination forecasting model based on wavelet network was established in accordance with combination prediction principle.The simulation results demonstrated the superiority of SVM combination forecasting model. After carrying on statistical analysis of the predicted heading angle deviation, the statistical value was compared with the set thresholds. Thus the ship heading control system fault prediction came true. Expert repository was established according to the fault tree, and the fault prediction was visualized depend on C++ Builder 6.0.更多还原