Futures Trading Forecast Based On Hybrid Neural Network

After nearly 30 years of development,China's futures trading market has got rid of the chaotic situation,and attract more and more investors to join the futures market for trading.By predicting the range of changes in futures trading prices,this research will not only provide investors with trading advice,but also have great significance for the government to regulate the market.Aiming at the problem of forecasting the futures price change range,this thesis proposes a forecasting model of futures price change amplitude based on hybrid neural network.In the process of model design,dataset selection and neural network structure are taken into consideration.In terms of datasets,taking the investor's investment returns and the actual value of the forecast results into account,the transition point and the transition point related data are selected from the continuous time series dataset as the data set for training and prediction.The current forecasting model usually only considers the closing price,which will lose a lot of useful information.This thesis chooses to take the highest price,the lowest price,the volume into consideration,and uses the principal component analysis method to reduce the dimensionality of the multidimensional data.In terms of neural network structure,considering the multi-scale features of financial time series,this paper use the hybrid neural network structure to make the prediction.In the training process,the auxiliary loss function is introduced,so that the composition network can optimize itself during the training process.According to the traditional prediction method,auxiliary loss functions are given a weight in order to reflect the difference between different types of data.In order to verify the validation of the model,the data of the soybean five,the rebar No.10 contract,and the CSI 300 stock index five contract were selected for verification.The comparison model are separate 1D-CNN,LSTM network and a cascade structure ofConvNet and LSTM networks.The results show that the average prediction accuracy of the model on the three data sets is 6% higher than that of the single network,and 4%higher than that of the cascade network,which verifies the effectiveness of hybrid network model.