基于ARIMA与WASDN加权组合的时间序列预测

Time series forecasting based on weighted combination of ARIMA and WASDN

为了提高时间序列预测方法的预测精度以及增强其适用性,提出一种ARIMA-WASDN加权组合方法。该方法同时使用差分自回归移动平均(autoregressive integrated moving average,ARIMA)模型与配备权值及结构确定(weights and structure determination,WASD)算法的幂激励前向神经网络(WASDN)对时间序列进行建模、测试以及预测。根据测试结果,将ARIMA与WASDN进行加权组合。数值实验结果显示,所提出的ARIMA-WASDN加权组合方法的预测精度高于ARIMA或WASDN单独使用时的预测精度,验证了该方法在时间序列预测方面的有效性和优越性。

In order to improve the forecasting accuracy and enhance the applicability of the time series forecasting approach, this paper proposed a novel weighted combination method, namely ARIMA-WASDN method. This method simultaneously exploited the ARIMA model and WASDN （ short for the power-activation feed-forward neuronet equipped with the WASD algo- rithm） to model, test and forecast the time series. According to the results of testing, two models could be combined into one model in a weighted manner for time series forecasting. Numerical experiment results indicate that the ARIMA-WASDN method can improve the accuracy achieved via either of the models used separately, and the results further illustrate the effectiveness and superiority of the proposed ARIMA-WASDN method in terms of time series forecasting.