A Time Series Analysis Based Data Aggregation Method In WSN

Data aggregation of WSNs is a method that processes the data of multi-sensor toeliminate redundant data transmission, and sends the aggregation result to base station.Data aggregation could lower energy consumption thus to prolong the lifetime of WSNsby reducing the amount of data communication. Prediction based data aggregation is amethod that forecast the future data of a sensor by building a predictive model with itshistorical data. If the deviation of real value and the predictive value exceeds a giventhreshold, the node sends its actual sensed data and updates its predictive model,otherwise, it does nothing.However, when a sensor is affected by man-made or environmental factors, thesensed data itself maybe abnormal and this may also lead to the predictive error exceedsthe error bound. Existing algorithms are mainly sending the data to an aggregate nodefor fusion. But it will not only affect the final fusion results, but also increase the burdenof network and shorten the lifetime of the network. Otherwise, as sensor nodes arelimited in power, computational capacities, and memory, if the complexity of modelingor prediction is high, the aggregation algorithm will be infeasible in practicalapplication. Therefore, the prediction based data aggregation should lower complexityof the predictive model under the premise of guaranteeing the accuracy of prediction.Based on the above analysis, this paper proposed an ARMA time series modelbased data aggregation method which is designed to reduce energy consumption ofWSN and improve the accuracy of aggregation results. By exploiting the temporalcorrelation of data series generated by a node, the proposed algorithm builds apredictive model with the historical data of the node and predicts its future value.Considering that the deployment strategy of sensor nodes makes the data are spatialcorrelated the proposed algorithm analysis the reliability of the raw data and eliminatesthe abnormal data. After analysis the ARMA model based data aggregation, we foundthat the cost of modeling and prediction are connected with the step of the model.Therefore, the algorithm optimized the step of the ARMA model in order toaccommodate to WSN.The proposed algorithm is tested on the dataset of the collected temperature anddeflection data of sensors deployed on a bridge in Chong Qing. Results of experimentsshows that the algorithm proposed in this paper could reduce energy consumption of WSN and get rid of abnormal data; the ARMA model which is determined by BIC&Ftest could better adapt to WSNs.