INTERACTINGPARTICLE-BASEDMODEL FOR MISSINGDATAINSENSORNETWORKS: FOUNDATIONSAND APPLICATIONS

Missing dataisunavoidable insensor networks duetosensorfaults, communication malfunctioning andmalicious attacks. There isaverylittle insight inmissing data causes and statistical andpattern properties ofmissing dataincollected data streams. Toaddress this problem, weutilize interactingparticle modelthat takes into account both patterns ofmissing data atindividual sensor data streams aswellasthecorrelation between occurrence ofmissing data atother sensor data streams. Themodelcanbeusedinalgorithms andprotocols forenergy efficient data collection andother tasks inpresence ofmissing data. We usestatistical intersensor models forpredicting the readings ofdifferent sensors. Asadriver application, weaddress theproblem ofenergy efficient sensing byadaptively coordinating thesleep schedules ofsensor nodeswhile we guarantee that values ofnodes inthesleep modecanberecovered fromtheawakenodeswithin auser's specified errorboundandprobability ofmissing data atawakenodesis less thanagiven threshold. Thesleeping coordination isaddressed bycreating themaximal number ofsubgroups ofdisjoint nodes, eachofwhosedata issufficient torecover thedata oftheentire network inpresence ofmissing data. Onsimulated andactually collected data fortemperature andhumidity sensors inIntel Berkeley Lab, weshowthat byusing sleeping coordination that considers missing data, wereduce thetypical 40%missing data rateoftraditional sleeping techniques toless than7%.