故障预测中基于模糊神经网络的规则发现方法

摘要 为分析和发现故障预测中表征网络运行状态的多指标和故障间的关联关系,提出了一种基于模糊神经网络的规则发现方法. 该方法利用模糊神经网络具有的学习能力和模糊推理能力,分析和发现网管系统中多指标和故障的关联关系,实现基于多指标的在线故障预测. 仿真实验结果表明,有效的参数初始化确定了算法收敛方向,从而加速了收敛速度;新方法能够准确预测故障的发生,并且优化了预测准确度、真正率、误判率等性能指标.
Aiming at analysis on correlations between failures and various indices characterizing network states in failure prediction, an approach for rule discovery based on fuzzy neural network (ARD_FNN) was proposed. It discoveries the correlations in network management system using the abilities to learn and infer by fuzzy logic in FNN. Simulations show that the effective initialization of parameters indicates the convergence directions that accelerate the convergence rates; thus the approach can predict the occurrence of failures accurately and such performance indicators as precision, true positive rate and false positive rate are optimized.