摘要
在分析D-S证据理论和神经网络理论各自特点的的基础之上,提出将这两种方法进行融合,并对电控汽车车载自诊断系统的诊断数据流参数进行处理和分析。该融合方法是将各个独立的低维神经网络的输出值处理后作为辨识框架上命题的基本可信度,经过证据理论的再次融合后得到最终的诊断结果。通过电控发动机典型故障的实例分析表明,该方法能够克服单一神经网络诊断中数据源包含信息的不全面性以及模糊性等局限性,并使得证据理论的基本可信度分配不再完全依赖专家的主观化赋值,同时可以充分利用各种故障的冗余和互补信息,从而使得汽车电控系统的故障的识别能力得到提高。
On the basis of analysis about D-S evidence theory and neural network, combination of these two methods was put forward to process and analyze the data flow parameters which from on-board diagnosis system of electronic controlled automobile. The fusing method takes each low-dimension neural network＇ s output value as the basic belief assignment value, then through D-S evidence fusing to get the final result. The example of typical fault of electronic controlled engine shows that （1） this method could reduce the insufficiency and fuzziness of information in data resource in single NN diagnosis; （2） it could make the basic belief assignment not fully depend on expert＇ s subject idea; and （3） it could also make use of various faults＇ redundant and complement information sufficiently to promote the recognition ability.
出处
《公路交通科技》
CAS
CSCD
北大核心
2009年第9期141-145,共5页
Journal of Highway and Transportation Research and Development
基金
黑龙江省自然科学基金面上资助项目(E200817)
关键词
汽车工程
诊断精度
融合
电控系统
神经网络
证据理论
automobile engineering
diagnostic accuracy
fusion
electric controlled system
neural network
evidence theory
