摘要
改进的BP算法把检验网络得到的性能指标J反馈回神经网络训练中,利用J和testJ构造出新的性能指标newJ,通过训练和检验的性能指标加权得到神经网络最优权值和阈值。用检验样本集间接地调整网络参数,使网络既能搜索更优解又不过早陷入局部极小值,从而提高网络的泛化能力。
Gained performance index testJ from tested network was returned to neural network training by improved BP algorithm. A new performance index newJ was constructed with J and testJ, the best weight and threshold of neural network was gained with training and testing performance index. Network parameters are adjusted indirectly with testing sample set, it make network not only search optimal solution but also does not early get in local minimal value, thereby, improve the generalization of network.
出处
《兵工自动化》
2004年第5期63-64,共2页
Ordnance Industry Automation
关键词
神经网络
泛化能力
改进BP算法
Neural network
Generalization
Improved BP algorithm