附加奇数s的动量BP算法在动态流量软测量中的研究与应用

Study on a Improved Momentum BP Algorithm with an Odd Parameter in the Neural Network Soft Measurement System of Dynamic Flow

针对BP算法收敛速度慢以及陷入平坦区的难题,提出了附加奇数s的动量BP算法.该方法在保证训练收敛和训练精度的情况下,通过降低计算量来提高计算速度节省计算时间,通过改进激发函数来增加梯度加快收敛速度.文中详细论证了算法的正确性,并通过实验验证了算法的性能,实验结果表明该算法比传统的附加动量BP算法节省了9.66%的时间,训练步数也减少了31.35%,比较好的适应于动态流量软测量中的实时性要求.

To solve the slow converge efficiency and easier to fall into the flat area in the training of neural network, a improved momentum BP algorithm with an odd parameter was proposed. Under the convergence and train precision which was demanded, it can save time through reducing the calculation, it can also increase the convergence speed through increasing the grads. In this paper, it argument the validity in detail,and its capabilities are validated by experimentations finally. The new algorithm can save the training time about 9.65% and training step was also decreased about 31. 13% in the experimentations than the traditional BP algorithm. It is suitable for demanding about Real-time with the dynamic flow measure.