Research On Transmission Optimization Based On Time Series Prediction In Wireless Sensor Networks

Wireless sensor networks consist of sensor nodes with sensing,transmission,computation,and storage capabilities,and have been widely used in various fields such as smart homes,smart cities,and environmental monitoring in recent years.However,the limited energy of sensor nodes restricts the usage period of wireless sensor networks.During the operation period of wireless sensor networks,transmission is the most significant factor affecting energy consumption.Therefore,optimizing transmission in wireless sensor networks to reduce the energy consumption of sensor nodes is particularly important for extending the usage period of wireless sensor networks.Time series prediction is the inference of future data using prediction models based on historical data.Conducting time series prediction on sensor nodes and using predicted values instead of sensing data within an allowable error range can reduce the amount of transmitted data in wireless sensor networks and achieve transmission optimization.On the one hand,sensor nodes have limited computation and storage capabilities,requiring a reduction in the computational cost of time series prediction.On the other hand,the allowable error range is controlled by a threshold and needs to be adaptively adjusted during the operation period of wireless sensor networks.This thesis studies a research on transmission optimization based on time series prediction in wireless sensor networks.The main work is as follows:(1)To conduct time series prediction on sensor nodes with limited resources,a novel Echo State Network based on Improved Knowledge Distillation(ESN-IKD)is proposed.First,the model of ESN-IKD is constructed with the classic Echo State Network(ESN)as a student network,the long and short-term memory network as a teacher network,and the ESN with double loop reservoir structure(ESN-DLRS)as an assistant network.The student network learns the long-term memory capability of the teacher network with the help of the assistant network and achieves model compression.Second,the training algorithm of ESN-IKD is proposed to corrects the learning direction through the assistant network and eliminate redundant knowledge through iterative pruning.It can solve the problems of error learning and redundant learning in traditional knowledge distillation.Finally,ESN-IKD is applied to three typical time series prediction tasks.Simulation results show that ESN-IKD achieves model compression,with a reservoir possessing stronger long-term and shortterm memory capabilities and a more concise structure.Therefore,ESN-IKD has better prediction performance and lower computational cost than the teacher network and other optimized ESNs.(2)To further reduce the computational cost of time series prediction and achieve adaptive threshold adjustment during the operation period of wireless sensor networks,a transmission optimization method in wireless sensor networks based on ESN and threshold adjustment is proposed.First,a combined prediction model based on ESN is constructed with the least mean square algorithm making preliminary prediction,ESN-DLRS making in-depth prediction and ESN-IKD making corrected prediction.Second,a threshold adjustment algorithm based on congestion control is proposed.The algorithm is divided into three stages: fast attempt,slow attempt,and fast startup.The threshold is increased quickly in the fast attempt stage,increased slowly in the slow attempt stage,and contracted as needed in the fast startup stage,achieving adaptive threshold adjustment during the operation period of wireless sensor networks.Finally,the proposed method is simulated on three typical wireless sensor networks dataset.Simulation results show that the proposed method has high prediction performance,low energy consumption,and reduces the amount of transmitted data,effectively achieving wireless sensor network transmission optimization.