近年来移动互联网已经进入一个快速发展的时期，随着手机用户流量的数量级从GB级迈向TB级甚至PB级，海量手机用户流量的出现给各运营商在数据分析处理上带来挑战。如何对海量手机用户流量下阶段使用情况进行有效预测，进而拉动企业业务量增长，已成为研究热点。由于传统数据分析方法无法对海量数据进行快速分析处理，处理海量数据较慢，无法高效分析非结构化数据，且扩展性较差。因此建立一个基于大数据预测算法的平台去处理海量数据势在必行。Hadoop是一个分布式框架，可以方便实现对海量数据挖掘与处理，所以基于Hadoop平台的处理框架为解决上述问题提供了一种新的解决方案。 本文通过对比常用智能算法对手机流量的预测效果，得到BP神经网络算法优于其他各算法。但BP算法也存在着不足，为了克服BP神经网络算法陷入局部极小值问题，本文结合了BP神经网络与相空间重构技术，计算出最佳嵌入维数和延迟时间，来提高手机流量预测的准确率。对于训练过程需要消耗大量时间的缺点，本文提出了基于Hadoop平台对BP算法进行并行化处理，缩短训练时间，以此来提高处理海量数据的效率。为了能够更加直观反应数据的预测效果，本文搭建了基于Hadoop平台下利用MapReduce分布式编程框架与CC-BP算法相结合的仿真平台，采用了SSM框架技术，从而对手机流量数据预测结果进行可视化。 通过对比实验结果，证明本文所设计出基于Hadoop的BP改进算法处理海量手机流量数据效率更高，同时也保持着较高的预测精度。通过对比不同节点下的数据处理效率，可知随着平台节点数目的增加，海量数据的处理效率会随之提高，从而验证了仿真平台的可行性。

In recent years, mobile Internet has entered a period of rapid development, with the number of mobile phone users from the GB level to the level of TB or even PB-level, the emergence of mass mobile phone user traffic to the operators in the data analysis and processing to bring challenges. How to predict the use of mass mobile phone users in the next stage, and then to stimulate business growth has become a research hotspot. Because the traditional data analysis method can not analyze the massive data quickly, it is imperative to set up a platform based on large data prediction algorithm to deal with massive data.Hadoop is a distributed framework, which can easily realize the mining and processing of massive data, so the processing framework based on the Hadoop platform provides a new solution for solving the above problems. By comparing the prediction effect of common intelligent algorithms on mobile phone traffic, the BP neural network algorithm is superior to other algorithms. But BP algorithm also has some shortcomings. In order to overcome the problem that BP neural network algorithm falls into local minimum, this paper combines BP neural network and phase space reconstruction technology, calculates the best embedding dimension and delay time to improve the accuracy of mobile phone traffic prediction. For the shortcoming that training process needs a lot of time, this paper proposes a parallel processing of BP algorithm based on Hadoop platform to shorten the training time, in order to improve the efficiency of processing massive data.In order to reflect the forecasting effect of data more intuitively, this paper builds a simulation platform based on Hadoop platform, which combines the MapReduce distributed programming framework with the CC-BP algorithm, and adopts the SSM framework technology to visualize the forecasting results of mobile traffic data. By comparing the experimental results, it is proved that the improved BP algorithm based on Hadoop is more efficient and has higher prediction accuracy. By comparing the data processing efficiency of different nodes, we can see that with the increase of the number of platform nodes, the processing efficiency of massive data will be improved, thus verifying the feasibility of the simulation platform.