The Research On Parallel Time Series Retrieval Method Based On GPU

This paper first analysis the existing algorithm, including high-dimensional time series similarity metrics and the data stream analysis and processing techniques, then point out the problems between various methods and summarize the way to solve them.At the same time, in recent years, GPU-based high-performance parallel computing technology in recent years has been rapid progress in many areas, and has become general method to improve efficiency of algorithm in many area, this paper analyzes the principles of general purpose GPU-based parallel computing, and then analyze some successful cases of GPU used in all major areas, use the parallel computing capability and higher memory bandwidth characteristics, then proposed a unified device architecture(CUDA) time series model for data stream pattern mining method and parallel computing methods for time series similarity matching method.For the processing of data stream, early researchers have raised the classical spring algorithm, but the classic spring algorithm have the redundant computing phenomenon. We first analysis the existing improve strategies for it, then propose a GPU-based data stream processing algorithm. Experiments show that, relative to the spring algorithm, the algorithm proposed in this paper not only improved the processing speed, but also give a good solution to the problem above.