电信科学 ›› 2023, Vol. 39 ›› Issue (1): 108-116.doi: 10.11959/j.issn.1000-0801.2023004

• 研究与开发 • 上一篇    下一篇

基于KPCA-GA-BP神经网络的POI质量预测研究

刘璐1, 杨丹2, 陈睿杰2, 李嘉2, 周熹2   

  1. 1 中国移动通信集团设计院有限公司重庆分公司,重庆401121
    2 中国移动通信集团云南有限公司,云南 昆明 650228
  • 修回日期:2022-11-09 出版日期:2023-01-20 发布日期:2023-01-01
  • 作者简介:刘璐(1986− ),男,中国移动通信集团设计院有限公司重庆分公司工程师、高级咨询设计师,主要从事无线网络智能优化业务及相关咨询设计工作
    杨丹(1986− ),男,中国移动通信集团云南有限公司工程师,主要从事无线优化大数据分析及相关管理工作
    陈睿杰(1987− ),男,中国移动通信集团云南有限公司工程师,主要从事无线优化大数据分析、智能运维工作
    李嘉(1989− ),男,现就职于中国移动通信集团云南有限公司,主要从事无线优化大数据分析、智能运维工作
    周熹(1993− ),女,现就职于中国移动通信集团云南有限公司,主要从事网络优化研究工作

Research on POI quality prediction based on KPCA-GA-BP neural network

Lu LIU1, Dan YANG2, Ruijie CHEN2, Jia LI2, Xi ZHOU2   

  1. 1 Chongqing Branch of China Mobile Communications Group Design Institute Co., Ltd., Chongqing 401121, China
    2 Yunnan Branch of China Mobile Communications Group Co., Ltd., Kunming 650228, China
  • Revised:2022-11-09 Online:2023-01-20 Published:2023-01-01

摘要:

目前移动网络优化一般基于小区进行网络质量评估及预测,遵循“升维研究,降维实施”的研究思路,提出了兴趣点(point of interest,POI)网络质量的柔性评价体系,但其涉及较多网络关键绩效指标(key performance indicator,KPI),导致POI网络综合质量评价体系较为庞杂且预测精度不高,为提高POI网络质量预测精准性,采用核主成分分析(kernel principal component analysis,KPCA)算法对反向传播(back propagation,BP)神经网络的输入变量进行相关性压缩,简化了BP神经网络结构,然后通过遗传算法(genetic algorithm,GA)优化了BP神经网络连接权值及阈值参数。与传统BP神经网络预测结果进行对比,在预测准确度方面提高了10.90%,均方误差性能显著降低,对研究POI网络质量的预测可起到较好的支撑作用。

关键词: POI柔性评价体系, 核主成分分析, 遗传算法, BP神经网络, POI质量预测

Abstract:

At present, in network optimization, network quality evaluation and prediction are generally based on communities, and a flexible evaluation system for POI network quality was proposed following the research idea of “research on dimensionality increase and implementation of dimensionality reduction”.However, it involves many network KPI, resulting in a relatively complex evaluation system for POI network comprehensive quality and low prediction accuracy.In order to improve the prediction accuracy of POI network quality, KPCA was used to compress the correlation of input variables of BP neural network, the structure of BP neural network was simplified, and then the connection weights and threshold parameters of BP neural network were optimized through GA.Compared with the prediction results of traditional BP neural network, the prediction accuracy is improved by 10.9%, and the mean square error performance is significantly reduced, it can play a better supporting role in the prediction of POI network quality.

Key words: POI flexibility evaluation system, kernel principal component analysis, genetic algorithm, BP neural network, POI quality prediction

中图分类号: 

No Suggested Reading articles found!