Studies On Data Fusion In Wireless Sensor Networks

In recent years,the amount of data in sensor networks has increased exponentially.Data fusion technology is the main means to reduce the amount of data transmission and save node resources.Considering the limited computing power and energy resources of existing sensor nodes,this study designed data fusion models for two different data structures,time series data and multimedia data.The specific work is as follows:1.Aiming at the problem of high communication cost caused by a large amount of space-time redundancy in existing sensor data fusion technology,a data fusion model based on clustering and prediction algorithms is proposed to ensure the data accuracy of wireless sensor networks and save node energy.In the training stage,an online recurrent extreme learning machine is trained with historical data,and nodes are clustered to obtain strong links based on historical data and actual geographical distance.In the process of data fusion,the end users and sensor nodes both use the online recurrent extreme learning machine to predict the future sensing data,which can guarantee that the data sequence in the end users and sensor nodes are synchronous.Finally,cooperative transmission mode based on strong the link of nodes can reduce redundant transmission when prediction fails.The experimental results show that the proposed data fusion model can effectively predict sensor data and reduce communication costs.2.Aiming at the problem of low clustering accuracy and data incompleteness caused by data corruption during transmission in multimedia data clustering algorithm,the optimally designed variational autoencoder networks is proposed for extracting features of incomplete data and using high-order fuzzy C-means algorithm to improve cluster performance of incomplete data.In the initialization process,the feature extraction model based on variational autoencoder is optimized to learn and extract the low-dimensional features of incomplete data.In order to capture the non-linear correlation in the feature space of low-dimensional data,the tensor distance is used as the distance measure.Specifically,the low-dimensional features are clustered by a high-order tensor based fuzzy C-means algorithm.Finally,the decoder outputs the final clustering results and the image data is recovered by using low-dimensional features.Experiments on real datasets show that the proposed algorithm not only can effectively extract the features ofincomplete data,but also improve the clustering performance and get better data reconstruction results.