摘要
数据已经成为信息自动化与智能化的核心,由于数据集中数据的类别不均衡,系统误差和随机误差很容易导致数据产生异常点,因此提出基于时序CNN的FPGA数据处理异常点识别方法。为了保证不同数据关联的有效性,对错误数据中的噪声错误和缺失错误进行对齐处理,使用GloVe方法对原始数据中的数据浅层特征进行提取。在模型训练过程中,为了解决数据梯度爆炸和过度拟合问题,对数据归一化处理,提高网络收敛速度,并采用散度计算输入参数定点化尺度的方法减少计算量、保证精度。最后将CNN算法映射到FPGA中实现硬件加速,FPGA算子图切分按粗粒度切分和细粒度切分进行处理。实验结果表明,所提数据处理异常点识别方法可高效读取数据,能够充分识别数据处理中的异常点,具有较高的准确率和精度。
Data has become the core of information automation and intelligence.Due to the unbalanced categories of data in the data set,systematic errors and random errors can easily lead to abnormal points in the data.Therefore,an FPGA data processing abnormal point identification method based on time sequence CNN is proposed.In order to ensure the effectiveness of different data associations,the noise errors and missing errors in the wrong data are aligned,and the glove method is used to extract the shallow features of the data in the original data.In the process of model training,in order to solve the problems of data gradient explosion and overfitting,the data are normalized to improve the convergence speed of the network,and the divergence method is used to calculate the fixed-point scale of input parameters to reduce the amount of calculation and ensure the accuracy.Finally,the CNN algorithm is mapped to FPGA to realize hardware acceleration.FPGA operator graph segmentation is processed according to coarse-grained segmentation and fine-grained segmentation.The experimental results show that the proposed method can efficiently read the data and fully identify the abnormal points in data processing,and has high accuracy and precision.
作者
李政清
穆继亮
莫小琴
LI Zheng-qing;MU ji-liang;MO Xiao-qin(College of Technology,University of Sanya,Hainan Sanya Hainan 570200,China;School of Instrument and Electronics,North University of China,Taiyuan Shanxi 030000,China)
出处
《计算机仿真》
北大核心
2022年第5期409-412,422,共5页
Computer Simulation
基金
2019年海南省基础与应用基础研究计划(619QN244)
2019年海南省基础与应用基础研究计划(619QN245)。
关键词
数据处理异常点
对齐处理
归一化
散度计算
定点化
Abnormal points of data processing
Alignment processing
Normalization
Divergence calculation
Fixed point
