摘要
针对专家系统等单一模型解决发动机故障诊断存在的算法复杂度高、诊断准确率低等问题,提出了BP神经网络集成与DS证据推理相融合的神经集成融合模型,不仅实现了发动机不同部位的专家经验与实际观测数据的特征级融合,还实现了多个模型的优势互补。通过对该方法和传统的专家系统方法比较得出,神经网络集成融合方法提高了7.1%的诊断准确率。
A new fusion model is proposed, which is the combination of integration BP neural networks models and DS evidence reasoning model, to solve the problems of low precision rate in automotive engine fault diagnosis by traditional expert system. The method of this paper not only realizes feature level fusion of all subjective observation data and expert experiments on different parts of engineer, but also realizes the predominance compensation of different models. In simulation experiment, by comparison between the two methods, this method proposed can improve diagnosis precision by 7.1% while reduces time and complexity.
出处
《计算机工程》
CAS
CSCD
北大核心
2007年第14期210-212,共3页
Computer Engineering
基金
国家自然科学基金资助项目(69873007)
关键词
故障诊断
神经网络集成融合模型
特征级融合
诊断准确率
fault diagnosis
neural network integration fusion model
feature fusion
diagnosis precision
