神经网络专家系统在电源系统诊断中的应用

The Neural Network Expert System Applied to Power System Diagnosis

为了解决卫星这类复杂系统采用单一的智能故障诊断技术的诊断局限性问题,提出一种基于模糊综合评判的神经网络和专家系统相结合的诊断方法。选用改进的BP神经网络,添加动量因子和可变学习速率,并通过计算十个相同结构神经网络输出的标准差获得诊断结果的可信度。采用基于可信度的专家系统不确定性推理,给出了可信度计算方法。采用代数积与代数和模糊算子对两种诊断方法的结果进行综合评判。成功地应用于某卫星电源系统的故障诊断中。结果表明诊断的可信度和诊断的正确率得到提高。

Aiming at resolving the limitation problems of complex system such as satellite diagnosed by single intelligent fault diagnosis technology, a combined diagnosis method combining neural network and expert system and based on fuzzy synthesis evaluation is proposed. The BP neural network modified is selected by using momentum factor and variable learning speed rate. The diagnosis result confidence is obtained by computing the root mean square error of the outputs of ten same structure BP neural networks. The uncertain reasoning based on confidence is adopted for expert system, and the computation method for confidence is given. The results of two diagnosis methods are synthesis evaluated by fuzzy operator of algebra product and algebra sum. The technology is applied to the fault diagnosis of a satellite power system, and the result shows that the diagnosis confidence and accuracy is increased.