基于磨光函数的权值直接确定双输入BP神经网络

Direct Determination of BP Neural Network with Dual-Input by Weights Based on Smoothing Function

利用磨光函数与采样数据建立了双输入单输出的权值直接确定的BP神经网络。改进后的网络依据训练数据重要性,选择以数据为中心的磨光函数作为激励函数,同时网络结构可以根据数据多少得到相应的调整。根据误差反向传播学习算法得到了改进后网络的权值直接确定算法。在仿真中,利用2个不同的目标函数验证了改进算法的有效性与建模精度。实验结果表明,三次磨光函数建立的网络性能优于二次磨光函数建立的网络。

The BP neural network with double input single output directly determined by weights, is construc- ted by using the smoothing function and the sampled data. Based on the importance of training data, the selected data are used as the center of the polishing function as an activation function. At the same time, the network structure can be adjusted according to the number of data. According to the error back propagation learning al- gorithm, the weights of the improved network are determined directly. In the simulation experiment, two different objective functions are used to verify the effectiveness of the improved algorithm and the accuracy of mod- el. The experimental results show that the network performance established by cubic polishing function is better than that of quadratic polishing function.