赵健^1,2, 刘展^1,2, 樊彦国^1,2, 丁宁¹
1.中国石油大学(华东) 地球科学与技术学院, 山东 青岛 266580;2.青岛海洋科学与技术国家实验室海洋矿产资源评价与探测技术功能实验室, 山东 青岛 266071

摘要:
在对BP算法进行深入分析的基础上，将测量数据处理与误差理论中的精度评定方法应用到BP神经网络的精度估计中，通过分别计算BP神经网络学习训练过程及预测过程的输出层中误差，实现对神经网络模型的精度评定。最后以海洋油气资源预测为例，结合实测资料建立了BP神经网络预测模型并分别进行了学习训练过程及预测过程的精度评定，以期为神经网络模型结构的优化设计提供有效参考，为提高神经网络模型的适用性提供科学依据。
关键词:神经网络BP算法精度估计中误差
DOI：10.11759/hykx20180427001
分类号:P631
基金项目:中央高校基本科研业务费专项资金（18CX02066A）；山东省自然科学基金青年基金项目（ZR2014DQ008）；中国石油科技创新基金项目（2015D-5006-0302）

Precision estimation of BP neural network and its application in ocean oil and gas resources prediction
ZHAO Jian^1,2, LIU Zhan^1,2, FAN Yan-guo^1,2, DING Ning¹
1.School of Geosciences, China University of Petroleum(East China), Qingdao 266580, China;2.Laboratory for Marine Mineral Resources, Qingdao National Laboratory for Marine Science and Technology, Qingdao 266071, China
Abstract:
Based on an intensive analysis of the back-propagation (BP) algorithm, we use the accuracy estimation method for surveying error data processing and the error theory to estimate the precision of a BP neural network. We calculate the mean square errors of the output layer in the learning and prediction processes of the BP neural network, respectively, to evaluate the accuracy of the neural network model. Lastly, taking ocean oil and gas resources prediction as an example, we use measured data to establish a BP neural network prediction model and calculate the precision of the learning and prediction processes of the BP neural network. The results indicate that the proposed precision analysis can provide an effective reference for the optical design of the BP network structure and provide scientific basis for improving the applicability of the neural network model.
Key words:neural networkBP algorithmprecision estimationerror model





PDF全文下载地址:

点我下载PDF