析取置信规则库系统参数优化的深度神经网络模型

Deep neural network model for parameter optimization of disjunctive belief rule base system

基于深度学习的置信规则库系统在参数优化方面存在可移植性差、应用效率低等问题,为此,将深度神经网络与析取置信规则库结合,有效减少模型的规则和参数的数量;引入梯度下降算法优化模型参数,提高模型构建和优化的效率.通过非线性函数拟合、北京市空气质量污染预测和多个UCI公共分类数据集的实验,对提出的方法进行验证,并将实验结果与现有的置信规则库系统和传统的机器学习方法进行对比.结果表明,所提出的方法比传统的方法具有更高的推理精度和更快的训练速度.

Belief rule base based on deep learning has some problems in parameter optimization,such as insufficient portability and inefficiency in the application.This paper combined deep neural network with disjunctive belief rule base,which reduced the number of rules and parameters and improves the efficiency of model construction and optimization via the gradient descent algorithm.Finally,this paper fitted nonlinear functions,Beijing air pollution prediction and experimented on public datasets from UCI,and compared the experimental results with some existing EBRB systems and conventional machine learning methods.The results showed that the proposed method in this paper has higher accuracy and faster training velocity than the conventional methods.