欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
传统人工神经网络的软件实现速度慢,可以利用FPGA的并行性加速其实现。本文在软件上采用PSO优化ANN,得到最优权值,最优权值和测试样本保存在FPGA片上ROM,然后用FPGA-ANN实现股票预测。Sigmod函数的逼近采用分段近似和查找表相结合的方法。利用ANN层与层之间在FPGA中的流水线处理以及每一层内神经元与神经元之间的并行处理,使用国际通用股票预测数据集Nasdaq-100 index of Nasdaqsm进行仿真实验得知,该方法显著提高了股票预测的速度。
英文摘要：
We employ the parallelism of FPGA （Field Program Gate Array） to accelerate the speed of traditional ANN （Artificial Neural Network）. We initially train an ANN by PSO algorithmon a traditional computer and get the optimal weight parameters, which are preserved in a ROM of a FPGA with some test samples. This approach of FPGA-ANN is then applied to the forecast of a stock index. We approximate the function of Sigmod by the integration of LUT and piecewise approximation. The experiment, in which parallel processing between different layers of ANN and between different neurons of the same layer and the international standard data set, Nasdaq-100 index of Nasdaq^sm, are employed,shows that the speed of stock index forecast of FPGA-ANN is much higher than that of a PC-ANN.
同期刊论文项目
同项目期刊论文
期刊信息
位置：