EMD-BP神经网络预测模型及应用

时间序列分析是根据客观事物的连续性和规律性推测未来发展趋势的预测方法,分析时设法过滤除去不规则变动,突出反映趋势性和周期性变动。为了提高预测精度,构建了EMD-BP神经网络预测模型,利用Hilbert-Huang变换中的经验模态分解将时间序列分解为有限个本征模函数,重构后进行BP神经网络预测。通过对中国石化的股票资料进行实验仿真,表明该模型降低了被预测数据的非平稳性,其精度比直接用神经网络预测有较明显的提高。

EMD and BP neural network forecasting model and its applications

Time series analysis is a forecasting method to speculate future trends according to the continuity and regularity of development of objects. During the analysis, the irregular movements are tried to be removed while the trends and cyclic movements are highlighted. In order to improve forecast accuracy, EMD-BP Neural network prediction model is built and the empirical module decomposition in Hilbert-Huang transform is applied in the model to decompose the time series data into a finite number of intrinsic modular functions. After getting restructured, BP neural network prediction starts. The experiments are conducted and simulated by using Sinopec stock data. The results show that the non-stationarity of the forecast data is greatly reduced, and the prediction accuracy is improved compared with the direct neural network.