神经网络串行训练并行处理

NEURAL NETWORK SERIAL TRAINING AND PARALLEL PROCESSING

作者在文中提出一种神经网络串行训练处理方法 ,它能提高神经网络学习收敛的速度 ,并达到提高训练精度的要求。此外 ,该算法还可依据实际问题的需求 ,自适应地扩展神经网络的容量 ,使得可最大限度地充分利用已知信息 ,达到增强神经网络预测结果可靠性的能力。作者在文中还给出了本文算法运用于大庆油田某开发区块储层油气预测的应用实例 。

In this paper a novel approach of serial training and parallel processing in neural network training is presented, which can greatly speed up the convergence and improve the accuracy of the neural network training. Through this method, the capacity of neural networks can be adjusted automatically and the credibility of prediction results is greatly improved. A case study of hydrocarbon prediction in Daqing oil field with satisfied result is also given in the paper.