Research On Drift Calibration And Data Reconstruction Methods In Wireless Sensor Networks

In recent years,with the tremendous advancement of wireless communication,computer network and sensor technology,wireless sensor networks(WSNs)have been an important research field and widely applied in medical,military,environmental monitoring and other fields.For environmental monitoring,sensors are often deployed in the unattended outdoor area for a long time.In the process of long-term environmental monitoring application,the sensor data may encounter drift and loss phenomenon,which is a serious problem for final users who need accurate data to make scientific decisions.This paper focues on the sensor data drift and data loss in the environmental monitoring application of WSNs,three different algorithems of sensor data drift calibration and data reconstruction are presented.The main contributions of this paper are as follows.(1)A data drift blind-calibration algorithm based on genetic algorithm optimized BP neural network and Kalman filter(GABP-KF)is proposed.Aiming at the problem that the conventional BP neural network is prone to fall into the local optimal solution,the global search ability of the genetic algorithm is combined with the local search ability of the BP neural network,which effectively improves the prediction accuracy of the network.The simulation results show that GABP-KF maintains a high degree of model fitting,and is superior to the traditional algorithms in calibration accuracy.(2)A data drift blind-calibration algorithm based on the integration of optimized extreme learning machine(ELM)with Kalman filter(CELM-KF)is proposed.Aiming at the problem that the classification results are unstable when the parameters of the traditional ELM are randomly selected,the weights and thresholds of the network are optimized by the difference constraint among the samples to improve the stability of the results.The simulation results show that CELM-KF has obvious advantages over the exsiting similar algorithms in terms of training time,model fitting degree and calibration accuracy.(3)A missing data reconstruction algorithm based on adaptive K-means algorithm and fuzzy neural network(KM-FNN)is proposed.The traditional data reconstruction method mainly relies on the temporal correlation of the sensor node's perceived data or the spatial correlation of the deployment location,resulting in low data reconstruction accuracy in real application scenarios.To solve this problem,this paper introduces an adaptive mechanism to update the training model and use fuzzy neural network for data reconstruction.Simulation experiments show that KM-FNN has better performance than the exsiting similar algorithms.