Neural Network Model Based On Wavelet Multi-resolution Analysis Theory

In contrast to Fourier analysis,wavelet analysis can not only decompose the signal orthogonally,but also show the local characteristics of the signal in both time and frequency domains.In addition,the theory of multi-resolution analysis provides a bridge for the application of wavelet analysis theory in practice.Therefore,based on the wavelet multi-resolution analysis theory,this paper first proposes a new two-layer orthogonal multi-resolution wavelet neural network prediction model and algorithm,so that the iteration results of the weight parameters of the model are unique and the number of hidden layer nodes is determined.Then,we theoretically prove the rationality and feasibility of the model,i.e.,we demonstrate that the two-layer orthogonal multiresolution wavelet excitation function can approximate any function in the L~P(μ),and the algorithm of the model has consistent convergence,i.e.,the numerical algorithm of the model can converge to the objective function.Finally,we apply the model to the study of stock price prediction in finance.In order to improve the accuracy of the prediction,we first propose a threshold function on the continuity of the input signal and apply it to the data denoising preprocessing.Then,the two-layer orthogonal multiresolution wavelet neural network is trained to obtain the stock price prediction.The experimental results show that the wavelet denoising effect of the proposed threshold function is more prominent,and the established two-layer orthogonal multiresolution wavelet neural network has higher prediction accuracy in stock prediction.