Research On The Key Techniques Of Massive High Speed Rail Data Based On Hadoop And GPU Hybrid Model

Nowadays with the growing demand for train speed and transport capacity, security and comfort has become an important issue that must be solved. In order to guarantee the safety of running trains, the security system is taken as the key part of the research. Nowadays people use the modern sensor technology to collect a large number of data from the train, including vibration data, pressure data and noise data, which will have a direct impact on the safety system of the train. Furthermore, researchers have applied data mining technology to analyze and deal with a large number of complex high-speed rail vibration data. However, with the rapid growth of data, the traditional methods are inefficient and can’t meet the requirement of real-time processing. Cloud computing and GPU provide a method to solve this problem.Cloud computing is a computing method based on the Internet. It is the product of distributed processing, parallel processing and grid computing, which gathers software and hardware resources together to provide a variety of IT services.GPU has a high parallel computing ability. With the wide applications of CUDA developed by NVIDIA corporation, GPU has been applied to many areas. In this thesis, we combine cloud computing with GPU to extract features and classify features of high-speed rail vibration data in order to improve the real-time processing capacity.The main works are as follows:1. Analyze and reorganize the high-speed rail data. A software is designed to process high-speed rail data in order to get data from 58 channels under different dates, conditions, speeds and items.2. Establish the Hadoop platform and analyze the pretreatment algorithms of vibration signal, including outlier removal algorithm and linear trend removal algorithm. The pretreatment algorithm for the running gear data of high speed rail is developed based on Hadoop. The effectiveness of the proposed method is verified by the experiments and the parallel evaluation indexes are used to evaluate its performance.3. Establish the CUDA platform, analyze the feature extraction algorithms of nonlinear and non-stationary vibration signals, and design the method of fault diagnosis algorithm based on CUDA architecture. Firstly the vibration signal is processed by EMD algorithm, and multiple components are obtained. Then the fuzzy entropy of each component is calculated. Next KNN is used to classify the vectors which are composed by multiple fuzzy entropy. Finally, the efficiency of the proposed fault diagnosis algorithm based on CUDA is demonstrated by the experiments.4. Establish the hybrid platform of Hadoop and CUDA and design the method of fault diagnosis algorithm based on hybrid architecture. Experiments for fault diagnosis on hybrid architecture and Hadoop are carried out. Its efficiency is verified by comparing it with the fault diagnosis approach on CUDA architecture..The experimental results show that the method of processing the running gear data of high speed rail based on hybrid platform is efficient, and can meet the requirement of processing large scale of high-speed rail data.