Research of Time Series Processing Function in Chaotic Neural Network Prediction Model
Zhiguo Wang, Ting Zhang, Jiwei Yang, Shufang Li, Jinling Fu · International Conference on Electric Information and Control Engineering · 2012
Based on the chaos theory, a kind of Chaotic Neural Network (CNN) has been built to forecast the daily Average sediment concentration and the daily suspended load concentration. In this paper, a series of processing measures has been adopted, such as difference, subtracting tendency, smoothing and Proportional amplifier. After analyzing the effect of every processing measure and the training date, it shows that the forecast accuracy of the processed series are better than un-processed series. The smoothing improved the prediction accuracy, while the subtracting tendency has a small effect on the results under our training date, and other measure s ' effects are associated with data sequences and the network model. All the measures have reduced the randomness of sequences, thus improved the CNN forecast accuracy.