BP神经网络计算法及共价键长计算的研究

Study on the BP Neural Network Algorithm and Its Application in the Covalent Bond Length Calculations

提出了BP神经网络权值、阈值调整的双动量算法和学习率调整的批处理半恢复自适应调整法 ,基于此法计算了 15种含氢原子共价键的键长 ,计算键长的平均相对误差为 0 11%。结果表明 。

The calculation method with the double momentum items has been presneted to adjust weight and bias values of BP neural network,and the half\|recovery self\|adaption adjustmemnt method has been offered for adjusting learning rate after a batch of sample data are calculated.Using these methods,the covalent bond lengths in the molecule of containing hydrogen atoms are calculated.The average relative error of the calculated bond length is 0 11%.The calculated results show that he population convergence of the BP neural network is improved by these methods.