网络广告投放的神经网络组合预测模型及实现  

Implementation of neural network combination forecasting model based on network advertisement

作　　者：赵蔷蔷 徐晓敏[1] 谷晓燕[1] 

机构地区：[1]北京信息科技大学信息管理学院,北京100192

出　　处：《北京信息科技大学学报（自然科学版）》2015年第6期83-87,共5页Journal of Beijing Information Science and Technology University

基　　金：北京市教委科研计划项目(KM201411232016)

摘　　要：通过将Winters加法预测、GM(1,1)预测、ARIMA预测3种方法作为BP神经网络的输入,点击消费金额的实际值作为输出,构建网络广告投放的BP神经网络组合预测模型并予以实现。用SPSS、Matlab、EXCEL等软件模拟仿真出3种预测模型的预测结果,且调整校验到拟合度相对较高的值。利用Matlab进行网络训练学习,经反复训练1999次后精度达到一定要求停止训练,计算出相对误差进行比较,BP神经网络预测结果的误差最小、精度最高。组合预测降低误差风险,综合各单项预测的特征,提高预测结果的精确度和可靠度,为本行业相关预测分析提供了方法借鉴。By using winters addition forecasting,GM（ 1,1） model and ARIMA forecast as the input of BP neural network,and the click on the consumption amount of the actual value as the output,BP neural network combination forecasting model is constructed and realized based on network advertisement. SPSS,EXCEL,Matlab and other software are applied to simulate 3 kinds of forecasting model,and adjust the value to a relatively high fitting degree. Matlab is used to carry on the network training study,and the training repeated for 1999 times before it stops when the precision achieves certain request. The error accuracy of BP neural network prediction is the highest. Combined forecasting can reduce the risk of error,and synthesize the characteristics of the individual forecasting to improve the accuracy and reliability of the prediction results. It provides a reference for the relevant forecast analysis of the industry.

关 键 词：网络广告 BP神经网络 预测模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]