粗糙集理论及粗糙混合智能方法在船舶电力系统中的应用研究

【作者】 张腾飞；

【导师】 王锡淮；

【作者基本信息】 上海海事大学 ， 电力电子与电力传动， 2007， 博士

【摘要】 目前，船舶控制向着自动化和智能化的方向发展，发电机系统的容量不断增大，船舶电力系统越来越复杂，对船舶电力系统的研究也提出了更高的要求。采用新的人工智能技术、方法研究船舶电力系统，进行智能化信息处理、动态建模和智能控制，并应用于船舶电力系统是有现实意义的。波兰数学家Z.Pawlak于1982年提出的粗糙集理论(Rough Set Theory)是一种较新的软计算方法，它能有效地分析和处理不精确、不一致、不完整等各种不完备信息，并从中发现隐含的知识，揭示潜在的规律，在某些方面可以弥补其它软计算方法的不足。粗糙集理论的发展以及粗糙集与神经网络的集成研究为船舶电站的建模与控制提供了一种新的思路和方法。目前，粗糙集理论在船舶电站电力系统中的研究和应用还较少，尚处在起步、探索阶段。本文利用粗糙集理论在智能信息处理方面的突出特点、强大的数据分析能力，以及神经网络的自组织学习、容错性能，对粗糙集理论以及粗糙集和RBF神经网络混合系统在船舶电力系统建模与控制中的应用进行探索研究。主要研究内容包括：●粗糙集理论算法研究针对粗糙集方法只能处理量化数据以及传统离散化方法存在的缺点，提出了一种基于微粒群优化(PSO)的连续属性离散化方法，降低了连续属性离散化后决策表的不相容度；在深入研究粗糙集理论的基础上，发现了正区域的一些有用性质，提出了一种基于正区域的直接求核方法，并给出了两个利用正区域求取属性相对约简的算法；为了在属性约简的过程中省去对属性正区域的多次计算，引入了广义决策表这一工具，发现了一些度量属性核与相对约简的性质，提出了一种基于广义决策表的属性求核与约简算法，可以非常简单地求取属性的核与相对约简，所给算法既适用于相容决策表也适用于不相容决策表；由于从实际系统中采集到的数据所构成的信息系统往往是不完备的，通常要对其进行预处理，提出了一种不完备信息系统的属性相对约简算法；决策规则的约简是基于粗糙集智能信息处理的一个重要内容，基于二进制可辨矩阵给出一个简单的直接求取决策规则核的方法，并提出一种决策规则的约简算法。●基于粗糙集的径向基函数神经网络(粗糙-RBF网络)粗糙集理论与神经网络都是智能信息处理较为有效的方法，但都有着各自的局限性，同时两者之间存在着许多互补之处，为两者的集成研究提供了理论基础。在深入分析径向基函数特性和网络结构的基础上，提出了一种基于粗糙集理论的RBF神经网络(粗糙-RBF网络)设计方法，利用粗糙集简化神经网络输入样本，确定RBF网络的中心向量候选集和扩展常数，结合正交最小二乘(OLS)算法构造RBF网络的结构。●基于粗糙-RBF网络的船舶发电机动态建模方法粗糙集理论和神经网络的集成运用为处理包含不确定、不完整信息的复杂系统提供了一个强有力的工具，也为复杂系统的建模提供了一种新的思路。本文在对粗糙集理论与RBF神经网络集成研究的基础上，针对现有复杂非线性系统神经网络建模存在的缺点与不足，提出了一种基于粗糙-RBF网络的连续系统动态建模方法，并利用这种方法对具有复杂动态特性和不确定性的船舶同步发电机进行建模。●基于粗糙混合智能方法的船舶发电机励磁控制粗糙控制是近年来兴起的一种新的智能控制方法，粗糙集理论处理不确定性问题的独特方式以及它与其它理论较好的融合性将有利于它在智能控制领域的进一步发展。尽管对粗糙控制已经进行了一些研究，但数量和所占据的地位仍然相对较小，并且都是学术上的而不是真正意义上的应用。在计算智能领域，通过各种理论与方法相结合的方式解决实际问题是目前的主要研究方向之一。本文探索了基于粗糙集理论的混合智能算法，并应用于船舶同步发电机的励磁控制系统中。针对船舶电力系统的特点，首次提出了基于粗糙-RBF网络辨识的发电机励磁神经PID控制和粗糙-神经网络逆前馈补偿的发电机励磁复合控制两种混合智能粗糙控制方法，仿真结果验证了所设计方法的有效性。更多还原

【Abstract】 At present, automatic and intelligent control is the important developing direction for modern ships. The electrical capacity of ship generators is growing large, which impels ship power system more and more complex. So the demand for the ship power system research is becoming much higher. It will have the practical significance that adopting the new artificial intelligence technologies and methods to realize information process, dynamic modeling and intelligent control in the ships power system.Rough set theory putted forward by Polish scientist Z. Pawlak is a new soft computing method. It can deal with imprecise, uncertain and incomplete data validly, find the underlying knowledge or rules from initial information system, and can overcome weaknesses of other soft computing method in some aspects. The development of rough set theory and the research of integrating rough set with neural networks provide a new way for ship power station modeling and control. Up to now, the research and application of rough set theory in ship power system are very rare. In this dissertation, according to the data analysis ability of rough set theory in intelligent information processing, and the self-organizing and fault-tolerance capability of neural networks, the dynamic modeling and control for ship power system based on the hybrid intelligent method of rough set theory and RBF neural networks are researched mainly. The major innovations in this article are as follows:Rough Set Theory and AlgorithmsIt is a pity that rough set theory can only deal with the discrete attributes. Aiming at the weaknesses of traditional discretization methods, an algorithm for continuous attributes discretization based on particle swarm optimization (PSO) is presented here, which can reduce the incompatibility degree of discretized decision table. The rough set theory is deeply investigated, and some useful properties of the positive region are discovered. A method for attribute core computation directly based on the positive region is proposed, and then, two algorithms for attribute relative reduction based on positive region are given. In order to leave out the repeated computation of positive region in the attribute reduction process, a kind of generalized information table is introduced; on the basis of it a criterion of attribute core and relative reduction is provided. A method for calculating attribute core is presented. And then, an algorithm for relative attribute reduction based on the generalized information table is designed, which is suitable for not only consistent decision table but also inconsistent decision table. In real applications, many information systems are incomplete because of different reasons, and need pretreatment. An algorithm for calculating attribute relative reduction in incomplete systems directly is put forward. Moreover, the decision rules reduction is an important topic in the research on intelligent information processing based on rough set theory. A method for calculating decision rules core based on binary discernibility matrix directly is presented. And then, an algorithm for decision rules reduction is designed.RBF Neural Networks Constructing based on Rough Set Theory (RS-RBF Neural Networks)Rough set theory and neural networks, which are both valid methods for intelligent information processing, have respective limitations; meanwhile there are fine complementarities between them, which provide the theoretical foundation for their integration research. After analyzing the characteristics of radial basis function and the structure of RBF neural networks deeply, a RBF neural network configuring method based on rough set theory, namely RS-RBF networks, is presented. Firstly, in order to build RBF neuron center vectors candidate set and spreads, the training samples are reduced based on rough set theory. Then, orthogonal least squares method is used to construct RBF networks.Ship Synchronous Generator Dynamic Modeling Method Based on RS-RBF Neural NetworksThe integration of rough set theory and neural networks supplies a powerful way for information processing of complex nonlinear system with uncertain and incomplete data, and also provides a new approach for complex nonlinear system modeling. By virtue of respective advantage about rough set theory and RBF networks as well as their complementarities, the integration of them was researched, and a dynamic modeling method based on RS-RBF neural networks is presented. The method was applied to model the ship synchronous generator with complex dynamic characteristics and uncertainties. The simulation results prove the validity of this method.Ship Generator Excitation Control based on Hybrid Intelligent Rough Control MethodsRough control is a newly arisen intelligent control method in recent years. It is good for rough set theory development in intelligent control field that the particular way of dealing with uncertain problems and the good properties of fusion with other uncertain theories. Although there have been some researches on rough control, their number and domains are relatively small, and they are all academic rather than real life applications. In the field of computational intelligence, it is one of the most promising ways that synthesizing different theories to solve the practical problems. In this dissertation, hybrid intelligent methods based on rough set theory were pilot studied for excitation control system of ship synchronous generator. Aiming at the characteristics of ship power system, two hybrid intelligent rough control methods were presented for the first time. The one is an adaptive neural PID control strategy based on RS-RBF neural networks identification; another one is excitation compound control method with rough neural inverse system feed forward compensation. The simulation results prove the validity of these methods.更多还原