传感器输出时间序列实时预测方法的比较研究

为了提高传感器输出时间序列的实时预测精度,分析了时间序列可以预测的内在机理,认为噪声是影响预测精度的主要因素,提出了小波滤波与神经网络相结合的组合预测法,研究了最小二乘支持向量机的预测法,并选用了典型的过程变量信号,将两种方法进行了比较.仿真实验结果表明,小波滤波与神经网络相结合的组合预测法能够在消除测量噪声对预测精度...

Comparison study on real time prediction method of sensor output time series

In order to improve the real time prediction precision of sensor output time series,the predictable inherent mechanism of time series is analyzed,and the result of analysis indicates that noise is the main factor.In the paper,a combination method of wavelet filtering and neural network and a prediction method based on Least Square Support Vectors Machine(LSSVM) is proposed and analyzed,then typical process variable signal is used to compare the above two methods.The experimental results show that the combination method of wavelet filtering and neural network can both eliminate the effect of measurement noise on prediction precision and achieve a better fitting effect and prediction precision,which proves noise is the main factor to prediction precision and provides a new idea to increase prediction precision of sensor output time series,while the prediction method of LSSVM has higher real-time character.