粗决策树神经网络预报模型研究及优化

发布时间：2018-10-10 13:22

【摘要】：针对神经网络预报模型效率和精度低下的问题,提出利用粗决策树算法和粒子群算法分别优化神经网络结构和参数,以此提高预报模型的效率和精度。首先,针对静态算法对大数据和增量数据处理不足的问题,构造了基于粗决策树的动态规则提取算法。应用粗糙集进行预处理,降低数据维数,将粗集与决策树结合,用增量方式实现样本抽取,经过动态约简、决策树构造、规则提取与选择、匹配的循环过程实现了数据的动态规则提取。将粗决策树算法用来确定神经网络隐含层,使隐含层选取更加科学合理,精简模型结构。该算法可以获得更多的数据隐式信息,可以用于数据挖掘领域,它提供了一种在线规则提取的研究方法。然后,针对BP网络初始化参数设置随意性大的缺陷,应用粒子群算法对网络初始参数进行优化,从而避免网络学习过程中过早收敛于局部极小值;后续关于模型的在线优化方面,还有待进一步研究。最后,采集2014年空气质量和汽车评估(相关数据),将粗决策树算法与粒子群算法结合起来建立神经网络预报模型,并应用MATLAB编程实现模型的训练和仿真。分别与不同隐含层的神经网络、未进行优化的神经网络预报模型进行分析和对比。结果表明,该方法构建的模型,大大提高了预报精度和速度,为预报问题的处理提供了新的研究思路。
[Abstract]:Aiming at the problem of low efficiency and precision of neural network prediction model, the rough decision tree algorithm and particle swarm optimization algorithm are used to optimize the neural network structure and parameters, respectively, in order to improve the efficiency and accuracy of the prediction model. Firstly, a dynamic rule extraction algorithm based on rough decision tree is proposed to solve the problem of insufficient processing of big data and incremental data by static algorithm. Rough set is used to preprocess, reduce the dimension of data, combine rough set with decision tree, realize sample extraction by incremental way, through dynamic reduction, decision tree construction, rule extraction and selection. The matching loop process realizes the dynamic rule extraction of the data. The rough decision tree algorithm is used to determine the hidden layer of neural network, which makes the selection of hidden layer more scientific and reasonable, and simplifies the model structure. The algorithm can obtain more data implicit information and can be used in the field of data mining. It provides an online rule extraction research method. Then, in order to avoid the premature convergence to the local minimum in the learning process of BP network, the particle swarm optimization algorithm is applied to optimize the initial parameters of BP network. Further research on the online optimization of the model is needed. Finally, the air quality and vehicle evaluation (related data) in 2014 are collected, the rough decision tree algorithm and particle swarm optimization algorithm are combined to establish the neural network prediction model, and the training and simulation of the model are realized by MATLAB programming. The prediction model of neural network with different hidden layers is analyzed and compared with that of unoptimized neural network. The results show that the model constructed by this method greatly improves the prediction accuracy and speed, and provides a new research idea for the processing of forecasting problems.
【学位授予单位】：华北理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183

【参考文献】