模糊神经网络预测算法改进及应用
发布时间：2019-04-01 10:47
【摘要】：模糊神经网络是人工神经网络领域的一个重要分支,并且广泛的应用于系统控制、建模等问题当中,并取得了不俗的成就。不过面对维度高、关联性低的数据时,因为某个结果只和数据中的一项或几项相关,所以模型的预测结果差,生成过多的模糊规则也不易于被人理解。就存在的一些相应问题,本文对其进行相应讨论与研究,包括:输入向量的选取方式、模糊神经网络的学习能力、模糊集合的划分、模糊规则的生成方式及误差等问题。本论文的工作内容包括:1、研究了国内外模糊神经网络的现状,研究了国内外教育数据的挖掘技术,并指出了理论与应用中存在的一些问题。2、针对存在的问题,比较了多种模型,在自适应模糊神经推理系统(Adaptive Network-based Fuzzy Inference System)的基础之上,对模糊神经网络模型进行了适当的改进,使其更适用于高维度、低关联的数据,可以找出数据中关联性较大若干数据项,描述其关系,并生成易于人理解的知识。3、使用修改后的自适应神经模糊推理系统,进行计算机验证,生成若干输入变量,其中一部分与输出变量有关,一部分与输出变量无关,寻找出与输出变量有关的输入量并做出相应解释。4、将模型应用在大学生的教育数据上,根据各科成绩的相关性,做出相应的成绩预测。以某些先导课程的成绩来预测学生将来某些科目的成绩,解决了传统神经网络,在高维度输入时,训练时间过长,预测结果较差的弊端。新的方法能够寻找各种的科目成绩之间的关系,并做出了相应解释,预测的精度更高,误差更小,而且有较强的解释性。
[Abstract]:The fuzzy neural network is an important branch in the field of artificial neural network, and it is widely used in the problems of system control and modeling. However, in the case of data with high dimension and low correlation, because a result is only related to one or several of the data, the prediction result of the model is poor, and the generation of too many fuzzy rules is not easy to be understood. In this paper, some corresponding problems are discussed, including the selection of input vector, the learning ability of the fuzzy neural network, the division of the fuzzy set, the generation of the fuzzy rule and the error. The working contents of this paper are as follows:1. The present situation of the fuzzy neural network at home and abroad is studied, the mining technology of the domestic and foreign educational data is studied, and some problems in the theory and application are pointed out. Based on the Adaptive Network-based Fuzzy Inference System, the fuzzy neural network model is modified appropriately, so that it is more suitable for high-dimension and low-associated data, which can find a number of data items in the data and describe the relation. a modified adaptive neural fuzzy inference system is used for computer verification to generate a plurality of input variables, a part of which is related to an output variable, a part of which is independent of the output variable, And finding out the input quantity related to the output variable and making corresponding explanation.4, applying the model to the education data of the college students, and making corresponding achievement prediction according to the correlation of the results of the subjects. The results of some pilot courses are used to predict the performance of some subjects in the future, and the traditional neural network is solved. In the case of high-dimension input, the training time is too long and the prediction result is poor. The new method can find the relation between the subject's achievements and make a corresponding explanation. The accuracy of the prediction is higher, the error is smaller, and there is a strong explanation.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183
本文编号：2451497
[Abstract]:The fuzzy neural network is an important branch in the field of artificial neural network, and it is widely used in the problems of system control and modeling. However, in the case of data with high dimension and low correlation, because a result is only related to one or several of the data, the prediction result of the model is poor, and the generation of too many fuzzy rules is not easy to be understood. In this paper, some corresponding problems are discussed, including the selection of input vector, the learning ability of the fuzzy neural network, the division of the fuzzy set, the generation of the fuzzy rule and the error. The working contents of this paper are as follows:1. The present situation of the fuzzy neural network at home and abroad is studied, the mining technology of the domestic and foreign educational data is studied, and some problems in the theory and application are pointed out. Based on the Adaptive Network-based Fuzzy Inference System, the fuzzy neural network model is modified appropriately, so that it is more suitable for high-dimension and low-associated data, which can find a number of data items in the data and describe the relation. a modified adaptive neural fuzzy inference system is used for computer verification to generate a plurality of input variables, a part of which is related to an output variable, a part of which is independent of the output variable, And finding out the input quantity related to the output variable and making corresponding explanation.4, applying the model to the education data of the college students, and making corresponding achievement prediction according to the correlation of the results of the subjects. The results of some pilot courses are used to predict the performance of some subjects in the future, and the traditional neural network is solved. In the case of high-dimension input, the training time is too long and the prediction result is poor. The new method can find the relation between the subject's achievements and make a corresponding explanation. The accuracy of the prediction is higher, the error is smaller, and there is a strong explanation.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183
【参考文献】
相关期刊论文 前7条
1 李婷;傅钢善;;国内外教育数据挖掘研究现状及趋势分析[J];现代教育技术;2010年10期
2 李岩;;神经网络的应用及其发展[J];商情(教育经济研究);2008年05期
3 黄勇;李玉华;;面向知识发现的数据分类技术在网络教学中的应用研究[J];南华大学学报(自然科学版);2006年02期
4 蒋玉兰;周磊;;关于开放教育学生流失情况的调研报告——以宁波电大2000级金融专业为例[J];宁波广播电视大学学报;2006年01期
5 周黎辉,韩璞,孙海蓉;一种有约束的神经网络预测控制方法[J];系统仿真学报;2003年05期
6 曾宪华,赖晓风;补偿模糊神经网络的改进及实例分析[J];四川师范学院学报(自然科学版);2002年03期
7 朴营国,何希勤,张化光;一种多输入-多输出模糊自适应控制方法的研究[J];自动化学报;2000年01期
，本文编号：2451497
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2451497.html
