基于改进的概率神经网络的分类预测方法的理论、算法与应用研究

项目名称： 基于改进的概率神经网络的分类预测方法的理论、算法与应用研究

项目编号： No.71271070

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 管理科学

项目作者： 吴冲

作者单位： 哈尔滨工业大学

项目金额： 54万元

中文摘要： 针对概率神经网络的优缺点特征，对该方法进行改进，设计相应的机制发挥该方法的优点，在一定程度和应用背景下，克服其缺点，以提供一个高效、准确、适用性强的新分类预测方法。本项目建立三种可通过训练来修正的网络参数得到椭圆概率神经网络，推导椭圆概率神经网络的学习规则。研究模糊值概率神经网络在一个紧集上能否以任意精度逼近在水平收敛，send-图度量导出的收敛等常见收敛结构下的连续模糊值函数。如果可以或在一定条件下可以，给出这种实现的具体步骤。在概念漂移的概念基础上，研究适用于椭圆概率神经网络的多分类器集成方法，并对算法的误差率、可行性、准确性及复杂度进行理论分析，基于该分类预测新方法进行实证研究. 讨论椭圆概率神经网络与模糊理论相结合的机制与算法，讨论在模糊语义下模型的训练问题，并进行实证分析。其研究成果将发展分类预测方法，提高分类预测的信息提取能力，提高其准确率, 解决分类对象属性时变性的问题.

中文关键词： 分类预测；椭圆概率神经网络；多分类器集成；模糊概率神经网络；财务危机预测

英文摘要： For the advantage and disadvantage characteristics of the probabilistic neural network, we improve this method, design the appropriate mechanisms for its advantages, and to a certain extent and the application context, overcome its shortcomings in order to provide an efficient, accurate and adaptable new classification prediction method. This project establishes three network parameters that can be corrected through training to get the oval probabilistic neural network and derives the learning rules of the oval probabilistic neural network. We explore whether the fuzzy probabilistic neural network can approximately convergence horizontally in a compact set in an arbitrary precision, and studies the common continuous fuzzy value functions like the send-diagram metrics exported convergence in the structure of convergence. We will give the specific steps of this implementation under certain conditions. Based on the definition of the concept drift, we study multiple classifier ensemble applied to the oval probabilistic neural network, and make the theoretical analysis of the error rate of the algorithm, the feasibility, accuracy and complexity, and do the empirical research based on the new classification prediction method. We discusse the mechanisms and algorithms to combine the oval probabilistic neural network an

英文关键词： Classification prediction；oval probabilistic neural networks；multiple classifier ensemble；fuzzy probabilistic neural network；financial distress prediction