论文部分内容阅读
目前 ANN的分析中缺乏对硬故障容错性能的分析 ,针对这一问题利用切比雪夫不等式给出了一种容错性分析的估算方法。利用切比雪夫不等式 ,分析了具有可微作用函数的前向神经网络容错性 ,建立了前向神经网络随机故障模型 ,讨论了固定型连接故障和错误输入故障对单个神经元的影响 ,通过分析这种前向神经网络故障传播特点 ,结合神经元容错分析的结论 ,得出了前向神经网络容错性分析的算法和相应公式。通过仿真实验 ,验证了上述结论的正确性。
At present, ANN analysis lacks the analysis of hard fault fault tolerance. To solve this problem, a method of fault tolerance analysis is given by using Chebyshev inequality. By using Chebyshev inequality, the fault tolerance of feedforward neural networks with differentiable function is analyzed. A stochastic fault model of feedforward neural network is established. The effects of fixed connection fault and faulty input fault on single neuron are discussed. By analyzing This kind of forward neural network fault propagation characteristics, combined with the neuronal fault-tolerant analysis concluded that the forward neural network fault tolerance analysis algorithm and the corresponding formula. Through the simulation experiment, the correctness of the above conclusion is verified.