Wavelet neural network weight initialization method based on Bayes estimation

Abstract

The invention relates to the technical field of optimization of the wavelet neural network, in particular to a wavelet neural network weight initialization method based on Bayes estimation, and the state estimation and search idea is adopted in the wavelet neural network weight initialization method. The wavelet neural network weight initialization method based on Bayes estimation comprises the steps of building a wavelet neural network model, unitizing weights, inputting and optimizing wavelet nerve cell weights, and optimizing weights of nerve cells of an output layer. Wavelet neural network weight parameters are linked with the network structure, wavelet types, input data and output target values, the state estimation idea and theory are introduced into initial setting of the weight parameters, wavelet network learning and training capacity is enhanced, the wavelet network has certain pertinence in the initialization phase, and therefore the adaptability of the weights in follow-up network learning and training is improved. Compared with a traditional weight initialization method, the learning efficiency can be effectively improved, oscillation amplitude of network output can be reduced, the rate of algorithm convergence is improved, and network output divergence caused by improper weights can be avoided.

Description

technical field [0001] The invention relates to the technical field of wavelet neural network optimization, in particular to a wavelet neural network weight initialization method based on Bayesian estimation using state estimation research ideas. Background technique [0002] The artificial neural network is a network system composed of interconnected artificial neurons. It abstracts and simplifies the human brain from the perspective of microstructure and function. It can be regarded as a large-scale highly parallel processor composed of simple processing units. Nature has the property of storing experiential knowledge and making it available. The similarity between the neural network and the human brain is that the knowledge acquired by the neural network is learned from the external environment, and the connection weights between the interconnected neurons are used to store the acquired knowledge. In terms of processing and computing, although the function of each proces...

AI Technical Summary

Problems solved by technology

But the shortcomings of this method are: first, it limits the output layer neurons to a linear structure, and does not initially set the weights between the hidden layer and the output layer; Training makes the output value of the network approach the target value, but the above method relies too much on the initial weight value generated randomly in the design process, ignoring the role of the target value of the network output, so it only speeds up the follow-up learning of the wavelet network to a certain extent speed

However, it is necessary to provide a variety of weight combinations in advance. Too few sample points will also affect the probability density of the weights. At the same time, it is necessary to assume that the weights are in the form of a Gaussian distribution, which has certain limitations.

Images