【摘要】 在油水层识别中,单纯使用神经计算存在因输入信息空间维数较大而使网络结构复杂、训练时间长,以及因冗余属性使网络拟合精度不高等缺点,为此基于属性约简和最优化原理提出一种简化的神经计算方法,主要包括基于粗糙集的样本属性约简算法,基于LM方法的稳定学习算法,以及基于黄金分割的隐含层节点数确定的优化算法等。仿真试验和实际应用表明,这种简化的神经计算方法不仅满足识别系统的精度要求,而且起到节省成本、提高处理速度等功效,在油水层识别中效果显著。更多还原

【Abstract】 In oil and water layer identification,using neural computing has disadvantages including complex network structure and long training time caused by large input information space dimension,and low matching accuracy of network caused by redundant attribute.Based on attribute simplification and optimization theory,this paper proposes a simplified neural computing method,including sample attribute simplification method based on coarse aggregation, stable study method based on LM method,and implicit layer count determination method base on gloden section.imitation test and actual application show that this simplfied neural computing method can not only meet the demand of accuracy of identification system,but also save the cost and improve processing speed.更多还原