基于CATTSTS模型的国际原油价格预测研究

Research on international crude oil price prediction based on CATTSTS model

国际原油价格数据具有复杂的特征变化趋势,直接使用现有模型对其预测往往效果不佳。针对此问题,提出一种分解-预测机制的预测模型。使用由噪声自适应完备总体平均经验模态分解算法对原油价格数据进行分解,将分解获取的子序列和残余趋势序列作为训练数据;基于CNN、LSTM单元和注意力机制模块构建了附有注意力机制的序列到序列深度学习模型。对所有子序列进行训练和预测,将预测结果重构以获取最终的价格预测结果。以布伦特原油日价对模型进行性能测试,结果表明,提出模型的预测结果与真实价格数据拟合情况良好,在平方根均方误差、平均绝对误差、平均绝对百分比误差分三项指标上均达到1.2545、0.9675及1.23%,相比其余几种对比模型有着更优秀的预测性能。

The international crude oil price data has complex characteristics and trends,and the direct use of existing models is often ineffective for its prediction.To solve this problem,this paper proposes a prediction model of decomposition-prediction mechanism.The crude oil price data is decomposed by using the complete ensemble empirical mode decomposition with adaptive noise,and the subsequences and residual trend sequences obtained by the decomposition are used as training data.Based on CNN,LSTM unit and attention mechanism module,a sequence-to-sequence deep learning model with attention mechanism is constructed.All subsequences are trained and predicted,and the prediction results are reconstructed to obtain the final price prediction results.The daily price of Brent crude oil is used to test the performance of the model.The results show that the prediction results of the proposed model fit well with the real price data,and the three indicators of square root mean-square error,mean absolute deviation and mean absolute percentage error all reach 1.2545,0.9675 and 1.23%,which means the proposed model has better prediction performance than other comparison models.