Hybrid Stochastic Deep Neural Network Model And Financial Market Statistical Analysis

The volatility of crude oil futures prices is highly nonlinear,uncertain and dynamic,which has led to its price forecasting facing many difficulties.Based on this problem,this paper proposes a hybrid stochastic neural network model that integrates empirical wavelet transform,random inheritance formula,Elman recurrent neural network,and variational learning rate algorithm.The random inheritance formula is proposed considering the timeliness of historical data on future price fluctuation,and can also simulate the impact of unpredictable events.The novel variational learning rate method optimizes the process of parameter selection for the model.EWT is utilized to extract feature sequences,and then establish a hybrid network model.Further,a more reliable synchronization evaluation method q order dyadic scales complexity invariant distance is proposed and applied to verify the accuracy of the proposed model.Empirical studies show that the proposed model has higher prediction accuracy.A stochastic deep bidirectional long short-term memory model is created by applying the random inheritance function and combining the deep learning algorithm and the bidirectional training structure.In-depth research found that it has higher prediction accuracy for low-frequency sequences decomposed by EWT,while the novel stochastic Elman recurrent neural network model based on variational learning rate has better performance for high-frequency sequences.Therefore,a new multi-step hybrid stochastic deep neural network model is proposed,which selects the appropriate prediction model according to the frequency of subsequences extracted from EWT.The accuracy of the proposed model and the effectiveness of optimization methods are confirmed by model comparison and evaluation analysis.In addition,experimental studies on variational learning rates provide a reliable parameter choice.