Improved Recurrent Neural Network Method And Its Application

As an effective method to study the analysis and prediction of large-scale time series,recurrent neural network algorithm has many improvements and variations,it has been applied in many aspects on time series data.However,there are still some problems that need to be further studied in the recurrent neural network.The main problem is that the existing recurrent neural network algorithm and its improvement do not have a good solution to the slow convergence speed and poor prediction accuracy.Therefore,the main content of this paper is the improvement of the initial weight selection of the recurrent neural network model.It is modified by introducing random disturbance term.Firstly,in order to solve the problem that the initial weight of the recurrent neural network model has a great influence on the convergence speed and prediction accuracy after the initial weight is selected.In this paper,the random disturbance term is introduced into the hidden layer of the recurrent neural network algorithm,and an improved recurrent neural network model is proposed.The new weights are obtained by adding the weight matrix connecting the input values and the random disturbance matrix,so that the convergence effect of the improved recurrent neural network algorithm is improved.Then,aiming at the problem that the initial weights of LSTM and GRU have great influence on the convergence speed and prediction accuracy of the algorithm after determining the initial weights,this paper introduces the random disturbance term into the gate structure of LSTM and GRU,and proposes the improved LSTM and GRU.The new weight matrix is obtained by adding the weight matrix connecting the input value and the random disturbance matrix in the gate structure,which improves the convergence effect of the improved LSTM and GRU algorithm.Again,aiming at the program design of recurrent neural network,LSTM and GRU,the program design ideas and specific Steps of improved recurrent neural network,LSTM and GRU are given.Finally,this paper selects the stock data to test the traditional recurrent neural network,LSTM and GRU compared with the improved recurrent neural network,LSTM and GRU.The test results show that the convergence speed and prediction accuracy of the improved recurrent neural network,LSTM and GRU are better than those of the traditional recurrent neural network,LSTM and GRU.