Modelling Of Compound Stochastic Neural Network Model And Research On Financial Market Volatility

The prediction of financial energy futures price time series has attracted much attention of the society.The correct prediction of energy futures price index time series can reduce the huge loss caused by energy price fluctuation to the national economy and protect the interests of various economic entities to the maximum extent.Hence this paper proposes two novel compound deep learning neural networks models to improve the prediction accuracy of financial energy futures price index.The stochastic time strength function takes into account the influence of historical data weight changing with time on neural network prediction.The closer the distance from the current and future time,the greater the weight value of historical data to the future forecast value.Ensemble empirical mode decomposition(EEMD)is an algorithm which decomposes the non-stationary and nonlinear time series into simple and independent time series.Thus one is to combine Elman recurrent neural network model(ERNN)with stochastic time strength and then introduce EEMD to propose a novel hybrid stochastic neural network(E-STERNN)so as to improve forecasting performance of energy market neural network system.Another model proposes the ST-LSTM model on the basis of the long short-term memory(LSTM).The stochastic time strength with different effects on the current and future information at different times is introduced to the weight and error of the LSTM model,which makes the model more consistent with the randomness and volatility of the futures market.Through the empirical study of global financial energy futures market price,a series of evaluation methods,such as error class evaluation statistics,trend class evaluation statistics and linear regression fitting method and so on,are adopted.And t-test and W rank test verify that the prediction results of the novel model and the conventional deep learning neural network model are significantly different.Additionally,a new type of multi-scale complex synchronization(q-MCCS)is proposed,which proves that both E-STERNN model and ST-LSTM model improve the prediction accuracy of the model from multi-scale,and that the two models have better prediction performance for long-term volatility than short-term fluctuation series.