用分形和神经网络算法改善仪表精度和响应时间

An Improvement on Precision and Responding Time of Instruments Using Fractal Interpolate and Neural Network

研究在仪器仪表中利用分形插值和人工神经网络算法改善测试精度和响应时间的方法。在仪表检测中有时会遇到检测的分辨率与实时性相冲突的情况 ,此时对数据进行插值是一个很好的解决办法。利用分形和人工神经网络算法插值是一种可以进行多点数插值的优越方法 ,因为它可以通过训练学习不断修正网络的权值 ,使检测误差的方差控制在预定的范围。还研究了利用神经网络做谱分析来求取主频的方法 ,它在运算速度和分辨率方面都优于 FFT。文章还提出了一些减少人工神经网络学习训练时间的方法。

A method on improving precision and real time response for the instrument is studied using fractal and neural network(NN). We may meet the detecting resolution in conflict with real time response of some instruments. Fraction and interpolation is a better way to solve these problems. Using fraction and NN arithmetic to modify the weight value while training to control the square root of error within a scheduled area. The method for spectrum analysis with NN is studied, it is better than FFT in efficiency and resolution. The way to reduce training time of NN is also discussed.