可解释性分层神经模糊网络的股票价格预测算法  

Interpretable hierarchical ANFIS for stock price prediction algorithm

作　　者：廖宏昊 胡峰[1] 邓维斌[1] LIAO Hong-hao;HU Feng;DENG Wei-bin(School of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《计算机工程与设计》2024年第12期3615-3621,共7页Computer Engineering and Design

基　　金：重庆市自然科学基金项目(cstc2021ycjh-bgzxm0013);重庆市教委重点合作基金项目(HZ2021008)。

摘　　要：针对现有的股票价格预测模型难以兼顾精度与可解释性的问题,提出一种基于分层神经模糊网络的股票价格预测模型。提出一种结合注意力机制的自适应神经模糊网络单元(ANFIS-A),以此单元构建分层自适应神经模糊网络;结合二进制灰狼优化算法(BGWO),提出一种特征子集选择算法;提出一种规则消除的递归算法,进一步减少规则数量,提高规则的可解释性。实验结果表明,该模型在预测股票价格方面具有较高的准确性和可解释性。In response to the challenges of balancing accuracy and interpretability in existing stock price prediction models,a hiera-rchical neural fuzzy network model for stock price prediction was proposed.An adaptive neural fuzzy network unit incorporating an attention mechanism(ANFIS-A)was introduced,and a hierarchical adaptive neural fuzzy network was constructed using these units.A feature subset selection algorithm was proposed by combining the binary grey wolf optimization algorithm(BGWO).A recursive rule elimination algorithm was introduced to further reduce the number of rules and enhance their interpreta-bility.Experimental results demonstrate that this model exhibits high accuracy and interpretability in stock price prediction.

关 键 词：灰狼优化算法 层次自适应模糊神经网络 注意力机制 股票价格预测 可解释性 金融时间序列 规则消除 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]