針對深度學習的應用於商業領域及時間序列預測，本研究提出了門控遞迴單元雙GRU模型作為預測黃金價格，這種方法包括資料蒐集與前處理，GRU模型的設計和訓練，測試和評估。 黃金的價格的預測受到石油、美元、通膨以及經濟政策和意外事件等眾多因素的影響。想要對這樣一非線性非穩定時間序列進行準確預測有相當難度，為了預測的準確盡可能的將它們加入到預測模型中。因此本文也加入其中具有代表性的指標，油價、VIX指數加入到模型調變中。利用門控遞迴單元GRU解決了傳統遞迴神經網絡無法解決長期依賴的問題，本文試圖使用它們建立模型來預測黃金的價格趨勢。 本文分別設計多個基於GRU的遞迴神經網絡模型，經由充分的訓練和調變及優化，使用VIX及原油價格相關數據，作為訓練時的數據特徵，對隔天的黃金收盤價進行預測，最終雙GRU模型獲得了更好的預測效果，證明雙GRU神經網路模型在金融市場預測上的效能，具有一定的創新性，且模型可以提供時間序列相關研究參考，具有一定的價值。 ;According to the application of deep learning to the prediction of business field and time series, this research proposes a gated recurrent unit of dual GRU model as a gold price prediction. This method includes data collection and pre-processing, design and training of GRU models, and testing. And evaluation. The price prediction of gold is affected by many factors including oil, the US dollar, inflation, and economic policies and contingency events. It is quite difficult to accurately predict such a non-linear and unstable time series, and add them to the prediction model as much as possible for the accuracy of the prediction. Therefore, this paper also incorporates the representative indexes among them, oil price and VIX index to add to the model adjustment. Using the gated recurrent unit GRU solves the problem that the traditional recurrent neural network cannot solve the long-term dependence. This article attempts to use them to build a model to predict the price trend of gold. In this paper, we design multiple GRU models based on recurrent neural network. After sufficient training, adjustment and optimization, we use VIX and crude oil price-related data as data characteristics during training to predict the closing price of gold the next day. The 3 dimensions of dual GRU model has got better prediction results, which proves the performance of the dual GRU recurrent neural network model in the prediction of financial market. It has certain innovation, and the model can provide reference for time series related research and has certain value.
