Back-propagation(bp) neural network has been widely used for dealing with large-scale nonlinear problems. However, the bpalgorithm is powerless, when in the face of singular sample data, which with high characteristi...
详细信息
Back-propagation(bp) neural network has been widely used for dealing with large-scale nonlinear problems. However, the bpalgorithm is powerless, when in the face of singular sample data, which with high characteristic dimensional and small sample size. Too many input makes the network structure is difficult to determine, cause slow convergence rate;fewer number of samples makes the network training is not complete, thereby affecting the recognition precision of the neural network. Aiming at these deficiencies, in this paper proposed the optimized bp neural network algorithm based on the partial least squares(pls) algorithm(pls-bp algorithm), firstly the new algorithm reduce feature dimension for singular sample data used pls method, which can fully take into account the level of correlation characteristic variables and the dependent variables, then get the low-dimensional data is used for network training and simulation. New algorithm simplifies the network structure and improves the network training speed and recognition precision.
暂无评论