Link Quality Prediction Of Wireless Sensor Networks Based On XGBoost

Wireless sensor networks are networks composed of a large number of cheap and miniature sensor nodes deployed in the monitoring areas.The sensor nodes communicate by radio.The communication of the radio is easily affected by signal interference,noise,and multi-path propagation effects,which lead to the unreliability of communication links.In addition,sensor nodes carry limited battery energy.Effective link quality prediction can help us choose high quality links for communication,so as to improve the probability of point-to-point communication,reduce the packets retransmission caused by poor links,improve the throughput of the networks,and extend the life of networks.By analyzing link characteristics,the average of the received signal strength indication,the average of the link quality indicator,the average of the signal to noise ratio,and the asymmetry index are selected as the link quality parameters.Considering the uncertain thresholds of dividing link quality grades and the subjective impact of manually dividing link quality grades,an improved fuzzy C-means clustering algorithm is proposed to adaptively divide the link quality grades according to the distribution characteristics of link quality samples.Calculate the imbalance ratio of link quality samples,and use the adaptive synthetic sampling algorithm to synthesize samples for minority link quality samples,so as to make link quality samples of different link quality grades reach balance.Benefiting from extreme gradient boosting(XGBoost)in classification,analyze the mapping relationship between link quality parameters and grades,and construct the link quality estimation model(XGB＿LQE)based on XGBoost classification algorithm to estimate the current link quality grade.Based on the estimated results of the XGB＿LQE,link quality time series sample sets are built by the sliding time window.The link quality prediction model(XGB＿LQP)is constructed based on the XGBoost regression algorithm,which can predict the link quality grade at the next moment.Deploy sensor nodes and collect data in three scenarios of square,laboratory and grove with different interference.The experimental results show that the estimation model XGB＿LQE has high accuracy,recall ratio,F1 score,G＿mean and AUC in scenarios with interference including moving objects,electromagnetic interference and shelters respectively.Compared with the exponentially weighted moving average based link quality prediction model,the 4C based link quality prediction model,and the support vector regression based link quality prediction model,the proposed link quality prediction model XGB＿LQP has higher prediction accuracy.