CAJ | 학술논문

针对氧化铝配料过程波动较大的非平稳时间序列参数预测问题，提出了一种基于小波分析的 ARMA(自回归移动平均)模型、BP 神经网络模型和 Holt-Winters 非季节模型的组合模型。首先通过小波分解，将原始时间序列依尺度分解为同长度不同频率的数据，然后对经过处理的高频数据和低频数据分别采用不同的模型进行预测，最后将不同频率的所有预测数据进行重构得到原始时间序列的组合预测模型。预测结果表明，所建立的模型对波动较大的非平稳时间序列的预测具有很大的优势。
침대양화려배료과정파동교대적비평은시간서렬삼수예측문제，제출료일충기우소파분석적 ARMA(자회귀이동평균)모형、BP 신경망락모형화 Holt-Winters 비계절모형적조합모형。수선통과소파분해，장원시시간서렬의척도분해위동장도불동빈솔적수거，연후대경과처리적고빈수거화저빈수거분별채용불동적모형진행예측，최후장불동빈솔적소유예측수거진행중구득도원시시간서렬적조합예측모형。예측결과표명，소건립적모형대파동교대적비평은시간서렬적예측구유흔대적우세。
In order to effectively realize the prediction of non-stationary time series with large fluctuation in the alumina blending process, a hybrid prediction model based on wavelet analysis is proposed,which is composed of the ARMA (autoregressive moving average) model, BP neural network model and Holt-Winters non seasonal model. Firstly, by wavelet decomposition, the original time series is decomposed into the same length of different frequency series according to the scale. Then, the high frequency series and the low frequency series are respectively predicted by the different models. Finally, the hybrid prediction model of the original time series is constructed by combining every series model. The prediction results show that the proposed model has great advantages for the prediction of non-stationary time series with large fluctuation of the process industry.