基于多维放缩和长短期记忆网络的期货价格预测

Futures Price Prediction Based on Multidimensional Scaling and Long Short-Term Memory Network

摘要: 结合多维放缩方法构建长短期记忆网络(MDS-LSTM)模型，对黄金主连期货未来10日的收盘价格进行预测。首先选取黄金主连的12个指标，利用多维放缩对12个指标进行降维，然后建立MDS-LSTM、MDS-BP和LSTM神经网络模型对收盘价格进行预测，将三个模型的预测结果进行比较，结果表明MDS-LSTM模型的预测精度较高，能更好的预测期货的走势。

Abstract: Combined with the multidimensional scaling method, the long short-term memory network (MDS-LSTM) model is constructed to predict the closing price of gold main link futures in the next 10 days. First, 12 indexes of the gold main link are selected and dimension-reduction of the 12 indexes is carried out. Then, MDS-LSTM, MDS-BP and LSTM neural network models are established to predict the closing price. The comparison of the prediction results of the three models shows that the MDS-LSTM model has a higher prediction accuracy and can better predict the futures trend.