| Two phases of a Radio Frequency Identification (RFID) Localization method are
explored for industrial processes including precast concrete structures. The first phase utilizes a
three-dimensional weighted Euclidean distance ๐-nearest neighbor (๐-NN) algorithm. This phase
localizes the target tags in motion. Average error is calculated to be 10.5 cm in the direction of
movement. Modifying the number of nearest neighbors from 4 to 6 decreases average error by
10%. The second phase of the RFID localization uses an artificial neural network that predicts
error based on the target tag location predictions done by the ๐-NN algorithm. By predicting the
error that is calculated in phase one, phase two is able to use a โcontrol systemโ that feeds the
error result back into the ๐-NN algorithm location prediction. An average Mean Absolute Error
(MAE) of 3.4 is seen using the ANN. The MAE shows that error is reduced from 10.5cm to
3.4cm.
Keywords: RFID Localization, k-NN algorithm, Activity Recognition, Artificial Neural
Networks, Machine Learning |