近年來，預測模型(Prediction Model)被廣泛且成功地應用在各種領域之中，藉由建立與使用系統之預測模型，可預先掌握系統之未來趨勢。而隨著建模對象越趨複雜，傳統以類神經網路(NeuralNetwork；NN)建立之預測模型其預測準確度已無法令人滿意。有鑑於此，本研究提出一新的基於經驗模態分解法之遞迴式赫米特類神經網路(Empirical Mode Decomposition based Recurrent HermiteNeural Network；ERHNN)預測模型，該模型整合了Hermite多項式之非線性訊號近似、EMD之非線性訊號分析與過濾、RNN之非線性系統建模與學習能力，對於非線性與非穩態訊號有優異之建模與預測效果。本研究以美國加州高速高路I-80W所記錄之短時交通流量進行預測驗證，最後經模擬證實，相較於NN預測模型，本研究所提出之ERHNN預測模型確實具有更佳之預測準確度。

In recent years, prediction model has been widely and successfully applied in various kinds of fields.According to the development and employment of prediction models, the trends of the system can bepredicted in advance. However, the performance of the pure neural network (NN) based prediction model isnot favorable in terms of the complication of modeling system. For this reason, an empirical modedecomposition based recurrent Hermite neural network (ERHNN) prediction model is proposed in this studywhich comprises good analysis ability for the non-linear and non-stationary signals through the combination ofthe merits of Hermit polynomial, empirical mode decomposition and recurrent Hermite neural network. Thevalidity of the ERHNN prediction model is verified using all day short-term traffic flow data on route I-80W inCalifornia. Simulation results show that the proposed ERHMM prediction model is with superior performancecompared with the pure NN one.