Research On Parallel Approaches For Processing Massive High Speed Rail Noise Data Based On Cloud Computing

With the rapid development of high-speed rail, the safety and comfort of the high-speed railway has become a hot research topic. The data collected from sensors which are installed on high-speed rail reflects the operational state of the train, and it is related to the safety of the train. For the affect of different factors, the data collected contains interference data with different frequencies and characteristics. Research indicates that preprocessing and filtering process can remove the interference data effectively. But with the increasing of the data, and traditional preprocessing and filtering running on single computer cannot fulfill the practical requirements. Cloud computing is a key technology to deal with these problems. The MapReduce model can be used for parallel computing on large-scale data. Due to its good parallel efficiency and unnecessary to know the underlying architecture, many researchers have applied MapReduce to design algorithms and achieved good results. So this thesis intends to apply cloud computing technology to preprocessing and filtering to improve the efficiency of processing the train noise data, it is of good practical value.At first, we introduce the present research situation of preprocessing, filtering and cloud computing. Then we summarize the cloud computing technology and preprocessing methods, we study the parallelization of the preprocessing method and put forward parallel methods for preprocessing high-speed railway noise data. The Speedup and Sizeup parallelization indicators are used to evaluate the performance of the algorithm. Experimental results show that the efficiency of the parallel preprocessing algorithm is good. Then we introduce the research of high-pass filter, low-pass filter, moving window filter and median filter. We also parallelize the filter methods. Experimental results on waveform display and filtering accuracy show that the filtering effect is obvious. Optimal parameters of high-pass and low-pass filtering are achieved through SNR and average variance. The three parallel parameters (e.g., Speedup, Sizeup and Scaleup) are used to evaluate the performance of parallel filtering algorithm. Experimental results show that the parallel movable window filtering and parallel median filtering algorithm perform well. The efficiencies of parallel high-pass filtering and parallel low-pass filter are not so ideal due to their use of public variables and the impact of time complexity of algorithms themselves.