前馈网的知识扩充及故障恢复

Knowledge Extension and Fault Recovery of Feed Forward Neural Networks

针对前馈网扩充性差的问题 ,提出了一种知识扩充方法 .在维持原有网络结构不变的基础上 ,通过添加一个新的子网 ,达到既保存现有训练结果 ,又可以学习新知识的目的 .同时 ,本文对神经网络的故障恢复策略进行了研究 ,提出了相应的补偿算法 .最后通过仿真实验对算法的有效性进行了验证 .

Aiming at the problem of poor extensibility of feed forward neural networks,a knowledge extension method is proposed in this paper.Preserving the original neural networks,we can both retain existing training result and learn new knowledge by adding a new subnet.Simultaneously,the strategy of fault recovery of neural networks is studied and a fault compensation algorithm is given.The effectiveness of proposed algorithms is verified by numerical simulations.