【作者】 沈敏；

【导师】 金心宇；

【作者基本信息】 浙江大学 ， 电路与系统， 2013， 硕士

【摘要】 GSM-R无线通信系统已经被选为中国列控专用通信系统。2003年6月至现在,国内多条铁路干线已经使用GSM-R网络。CTCS-3是中国目前的列车信号安全控制系统,该系统规定必须要保证GSM-R网络的可靠性,因此GSM-R网络优化工作已经成为网络维护的主要工作之一。GSM-R网络优化主要包括网络数据采集分析和网络参数调整。但采集数据存储格式不统一导致数据融合困难,单机存储导致存储量有限,查询性能低。本文提出云存储方案解决这些存储问题。同时基于单机处理的数据分析有很大的局限性,它只能处理简单、小规模的数据,算法运行速度慢,很难进行深层次的数据挖掘,软件安装维护繁琐,软硬件资源利用率低。因此本文结合云计算技术设计了GSM-R数据挖掘平台来解决这些问题,将单机的数据挖掘算法进行并行化,提高算法的时效性,同时显著增加了算法的数据处理规模。本文还提出了云计算的若干个优化方案,显著提高了计算效率。本文最后成功地将GSM-R场强覆盖模型预测应用到GSM-R数据挖掘平台。根据现有的场强覆盖理论模型提取出场强覆盖的影响因素,并考虑高速环境下的速度因素,在数据挖掘平台上以实测数据为训练样本,用并行化的BP算法训练场强覆盖预测神经网络,测试结果表明该模型相比理论模型能更好地预测GSM-R场强覆盖。更多还原

【Abstract】 GSM-R wireless communication system has been selected as the dedicated communication system of China Train Control System. Several railway lines has been using GSM-R since June2003. CTCS-3is China’s current train signal safety control system, which requires the reliability of the GSM-R network. So optimization of GSM-R network has become one of the main job of the network maintenance, which includes network data acquisition, data analysis and network parameter adjustment. But the data storage format is always not unified, which increase the difficulty of data fusion and decrease the query performance. And single server storage limits the storage capacity. So the paper deploys a cloud storage system to solve these storage problems. At the same time, one server based data processing has great limitations, it can only deal with simple, small data, and the algorithm will run slowly. It’s difficult to carry on deep level data mining. Deployment and maintenance of software is also very tedious and usage of software and hardware is low. So this paper design the GSM-R data mining platform based on cloud computing to solve these problems, which will improve the efficiency of data mining algorithm and greatly increase the processing scale. This paper also puts forward several optimization solutions for cloud computing system, which greatly improves the calculation efficiency. At last, this paper successfully applies the field strength coverage prediction model to GSM-R data mining platform, extracts the factors of field strength coverage according to the existing field strength coverage theories, and takes high speed into consideration, and takes measured data as training sample and trained an field strength coverage prediction neural network using parallel BP algorithm. The test result shows that the prediction model is better than theory model.更多还原