粗糙集属性约简算法在热工系统中的应用研究

发布时间：2018-03-26 17:05

本文选题：粗糙集　切入点：属性约简　出处：《华北电力大学》2017年硕士论文

【摘要】：粗糙集理论作为处理不确定信息的有效工具,在多个领域得到广泛应用,为复杂热工系统建模和故障分类提供了新的方法。本文描述了粗糙集和神经网络结合方法,该方法的实现步骤为原始数据的预处理、属性约简,到最后网络模型的构建。其中,针对连续属性的离散化,提出了无需指定聚类数的改进k-means算法,在确定算法的惩罚参数时引入了相容度的概念,充分考虑了属性间的相互关系。针对决策表的属性约简,提出了一种基于近似决策熵的约简算法,将近似精度与传统信息熵相结合,解决了传统信息熵单纯从信息论观点出发的不足。针对火电厂热工对象实时建模困难、模型精度不高,以及神经网络因输入量增多导致收敛速度大幅降低的问题,本文提出一种基于近似决策熵的BP网络建模方法。利用该方法对主汽温和NOx排放浓度建模,结果表明,该方法模型精度高,而且有效降低了BP网络输入层的维数,从而使网络的结构更加简单,训练速度更快,对热工系统建模具有重要的实用价值。为了实现电厂中设备故障类型的快速、准确判断,将基于近似决策熵模型的属性约简算法结合BP网络进行故障分类。该方法用于汽轮机的振动故障分类,结合方法有效降低了网络的输入维数,简化了网络结构,获得了较高的分类精度,同时降低了计算开销,对快速判断故障类型,提高维修效率,具有工程实用价值。
[Abstract]:Rough set theory as a effective tool to deal with uncertain information, is widely used in many fields, and provides a new method for complex thermal system modeling and fault classification. This paper describes the combination of rough sets and neural network methods, steps of pretreatment, the original data of the attribute reduction method to construct the network, finally model. Among them, the discretization of continuous attributes, this paper proposes an improved k-means algorithm without specifying the number of clusters, in determining the penalty parameter algorithm introduced the concept of consistency degree, full consideration of the relationship between attributes. The attribute reduction of decision table, proposes a reduction algorithm of approximate entropy decision based on the approximation accuracy and traditional information entropy, solves the shortcomings of traditional information entropy only from the viewpoint of information theory. According to the thermal process of power plant real-time modeling difficulties, the accuracy of the model is not Is high, and the neural network for input to increase convergence speed greatly reduces the problem, this paper presents a BP network modeling method based on approximate entropy decision. By using the method of main steam temperature and NOx emission modeling results show that the method of model accuracy, but also effectively reduces the dimension of BP network input layer, thereby the network structure is more simple, faster speed of training, has important practical value for thermal system modeling. In order to realize the fault type of power plant in the fast, accurate judgment, the approximate entropy model of decision attribute reduction algorithm based on the combination of BP network for fault classification. The method for vibration fault classification of steam turbine, combined with effective method to reduce the input dimension of the network, the network structure is simplified, obtain higher classification accuracy, while reducing the computational overhead to quickly judge the fault types and improve maintenance Efficiency is of practical engineering value.

【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TM621

【参考文献】