PSO-LSSVM based missing data imputation algorithm for wireless sensor network for nuclear power plants' environmental radiation monitor

Sensor nodes' monitoring data missing can cause damage to environment radiation continuous monitoring for nuclear power plant, the missing values should be estimated as accurately as possible. A missing data imputation algorithm based on least squares support vector machine (LSSVM) is proposed, where the optimum parameter set is found using particle swarm optimization algorithm (PSO). This algorithm imputes missing data utilizing node's previous monitoring data and neighbor node's current monitoring data jointly. Experimental results show that the proposed algorithm can impute the missing γ dose rate data with a maximum relative error of 3% and correlation coefficient of 0.955378, much better than the GA-LSSVM and the feed-forward network working with the same data set. (authors)