| With the development of communication and sensor technology,wireless sensor networks have gradually become the focus of attention in the industry,and their application fields have become more and more extensive.Obtaining complete and accurate sensory data through wireless sensor networks is the basic premise for data analysis and application decision-making.However,the inherent deficiencies of wireless sensor networks lead to data loss directly or indirectly due to many different factors,such as node power,storage space,computing ability,and wireless communication.Most of the existing problems for sensory data recovery are based on traditional machine learning or only based on a single data attribute,which makes it difficult to break through the bottleneck.In response to the above problems,this article explores how to recover the missing data in the sensor node efficiently from the following two aspects.Firstly,this paper studies the data recovery algorithm based on generative adversarial network in wireless sensor network.According to the characteristics of sensory data and combining the theoretical advantages of generative adversarial networks,a generative adversarial network model suitable for time series data is constructed,of which basic model is convolutional neural network.The input length suitable for the model and the dataset is selected through experiments,and the sequence with missing data is used as the input of the model.The missing data is recovered by using the trained model.Experiments show that the proposed wireless sensor network data recovery algorithm based on generative adversarial network can improve the effect of data recovery.Different from other machine learning algorithms used in data recovery,the proposed algorithm in this paper is more in line with the distribution of real data and ensures the local continuity of the recovered data.Secondly,this paper studies the data recovery problem based on self-attention multi-attribute assistance generative adversarial network in wireless sensor network.The Generative Adversarial Network is good at learning to generate new data with the same statistics at the training set,but it has the following shortcomings,shuch as easy to fall into Nash equilibrium;no global guarantee on the data continuity;poor in generalization ability caused by the data size limit.This paper introduces the Two-Timescale Update Rule,self-attention mechanism,and mixup data enhancement technology to optimize the model,and further uses the mean square error to optimize the loss function.In this paper,the different attribute data collected by the nodes are converted into multi-dimensional input data.Simulation results show that the proposed scheme has better performance compared with related works in various datasets.According to the data recovery problem in wireless sensor networks,the generative adversarial network used in computer vision is migrated to the wireless sensor network data recovery problem.Experiments show that the generative adversarial network achieves better results.Based on the above model,we use different strategies to optimized the model,and realized the multi-attribute simultaneous recovery in the single node,which further improves the effect of data recovery. |
