摘要
为了开发显示卡超频仿真软件,提出了一种估计超频结果测试值的算法;首先根据显示卡的主要性能指标确定了超频的设置参数、主要结果及其测试方法,然后以设置参数为输入、结果测试值为输出,构建了一个广义回归神经网络;采用交叉验证方法计算网络的拟合误差,并以此为指标函数,引入粒子群优化算法对网络参数进行了优化,从而提高了网络的拟合精度;经采集MSI R6850显示卡的25个超频样本实测,该算法使拟合误差降低了60%,其有效性得到了验证。
To develop the simulation software of graphics card＇ s overclocking, an algorithm to estimate the test value of overclocking result is presented. First, the overcloeking＇ s set parameters, main results and its test method is determined according to the main perform- ance index of graphics card. Then a general regression neural network is constructed with settings parameters and result value as input and output. The paper uses cross--validation to calculate the network＇ s fitting error and acting it as the index function. It optimizes the network parameters by introducing particle swarm optimization algorithm. So it improves the estimation accuracy of the network. Testing by collecting 25 overclocking samples of MSI R6850 graphics card, the algorithm reduces the fitting error by 60%, and its validity is verified.
出处
《计算机测量与控制》
北大核心
2014年第4期1263-1266,共4页
Computer Measurement &Control
关键词
粒子群优化算法
广义回归神经网络
显示卡超频
结果测试
particle swarm optimization algorithm
general regression neural network
graphics card
s overclocking
result test
