Research On Interpolation Method For Missing Data Based On Nonparametric Bayesian Dictionary Learning

Wireless sensor network(WSNs),as a kind of brand-new information perception technology,which has widely used in the many areas,like the intelligent household,industrial3.0,military defense,remote monitoring,et al,and plays an important role of information acquisition in the IoT.However,sensor nodes are susceptible to some non-artificial factors such as severe climate and bad environment or applying the compressed sensing(CS)technology into wireless sensor network for guarantee the network life,which both are inevitably to cause the original data lost.Therefore,in order to ensure that WSNs perception data keep in high accuracy and completeness,this article considering the signal itself contains abundant prior information,two kinds of efficient missing data interpolation method is proposed.The main work is as follows:1.Study the nonparametric Bayesian method and its classical model Dirichlet process.Because the size of traditional orthogonal basis or redundant dictionary are fixed,so it lacks of adaptability in the face of the signal which sparsity dynamic change.This paper puts forward a kind of loss of data interpolation method based on nonparametric Bayesian dictionary learning.On the basis of statistical learning,we introduce a nonparametric Bayesian prior that is no limited on the number and size of parameters,and we combine a prior of wide distribution with the samples.Then,the optimal parameters is obtained by Gibbs sampling method.Finally,we can interpolation the lost data by these optimal parameters.This article adopts the real temperature data for simulation experiments,the experimental results show that the proposed interpolation algorithm can effectively recover lost data.2.On account of data components in WSN is more and more diverse,its structure is becoming more and more complex,the single structure of the traditional dictionary has difficult to optimally represent this kind of signal.Therefore,we introduced an interpolation method based on structural dictionary learning,and based on this algorithm to improve the way of training dictionary,we proposed an interpolation method of nonparametric Bayesian learning dictionary based on structural information.By introducing the wavelet transform and TV transformation to fully dig the structural features of signal itself,and utilize the nonparametric Bayesian dictionary learning method to obtain the optimal structure of dictionary and sparse coefficient and realize interpolation lost data.The simulation results indicate that this improved algorithm can further improve the interpolation accuracy and shorten the operation time.