Research And Application On Interval Wavelet Nerual Networks

Information is the source of understanding of the world and transforming the world.However,due to the diversity,complexity,time-variation of the world,the information people realized is often uncertain and imprecise.The phenomenon of imprecision of measurement data caused by the limit of sensor precision or external disturbance also appear in scientific research and complex industrial process.In the following processing,approximation always exist.So the first issue is how to describe the uncertain information and establish mathematical model of uncertain information.Granule is a concept to describe uncertain,imprecise,fuzzy information.There are various kinds of information granulation methods.One of them is interval theory which has excellent universality and outstanding description ability of inaccuracy data.Meanwhile wavelet neural networks is the combination of wavelet theory and neural networks,which has highlight time and frequency domain analysis ability from wavelet theory and highlight ability of complex system modeling and self-learning.Combining interval theory and wavelet neural networks consisting of interval wavelet neural networks can well solve the problem of inaccurate information modeling.So the research on the structure and learning algorithm of interval wavelet neural networks have a very important theoretical and practical significance.Based on the interval theory and wavelet neural networks theory,structure and gradient descend learning algorithm of interval wavelet neural networks are further studied in this thesis through extensive review of relevant documents.Aim to simplify calculation and classification of first kind of interval wavelet neural networks,an first kind of extension interval wavelet neural networks is proposed.Then applied it to PM2.5 laser monitoring system modeling.The main research work is as follows:First of all,the interval wavelet neural networks can be divided into three categories,and the relevant model is established and deduced gradient descent algorithm is used in first kind and second kind of three layer interval wavelet neural networks.Aim to simplify calculation and classification of first kind of interval wavelet neural networks,an first kind of extension interval wavelet neural networks is proposed.In accordance with partial derivative of extremum points is zero during back propagation process,the derived function translation and adaptive learning rate method is adopted to solve this problem.Secondly,under the background of increasingly urgent needs of air environment quality monitoring,I design and exploit a portable PM2.5 laser monitoring device,describe its working principle and model it based on Fraunhofer diffraction principle.Then design the measurement control system in detail,make theoretical simulation and verification as the hardware test platform.Finally,respectively analyze two kinds of interval wavelet neural networks' background and significance,select the first kind of extension interval wavelet neural networks to model and forecast PM2.5 laser monitoring system.Describe the modeling process in detail,use matlab to simulation and make contrast to model of two traditional wavelet neural networks.It's easily found that,the first kind of interval wavelet neural networks has good performance and application prospect in the aspect of system evaluation and fault monitoring.In conclusion,hope that this paper may be helpful to the structure and formula of the non-monotonic activation function interval neural networks(such as wavelet function,Gaussian funtion,etc).