| With the arrival of the era of big data, data mining for time series has become a new research direction. In time series data mining, the similarity measure of time series is the most basic and important link. In a lot of time series similarity measure algorithm, dynamic time warping algorithm is widely recognized by the industry as the best time series metric algorithm.With the popularization and development of embedded devices, the demand of time series data mining is increasing. The resources, performance and power consumption of the embedded platform are relatively high. But it is important to calculate the similarity of time series and improve the performance and energy efficiency of the embedded platform.The main work of this paper is to implement a DTW algorithm acceleration system based on System-on-a-Programmable-Chip(SoPC) platform by the method of hardware and software collaboration. The purpose of this system is to improve DTW computing performance and reduce consumption without reducing system flexibility and availability. The proposed system makes full use of the parallelism of the algorithm itself, realizes the fine-grained parallel computing, and has high computational efficiency; this design supports configurable features that support the DTW distance calculation for all sequence lengths; the proposed system is based on the SoPC platform, so the system task can be divided into hardware and software, thus we can take the DTW-FPGA module as a software layer subroutine to be called, which enhance the flexibility of application design.We have carried out a large number of experiments on the proposed system. The experiment covers a variety of data sets, for different applications, and treat performance and power consumption as the main evaluation system. In the experiment, the system has achieved the average 6.91 x speedup and 3.88 x energy efficiency improvement, and can adapt to a wide variety of data sets, with a certain stability and flexibility. |
PDF Full Text Request