An improved data streams processing algorithm for sensor networks based on training weights

In order to enhance the performance of SPIRIT, in this paper we propose an improved multi-streams processing algorithm for data streams processing of sensor networks. The improved algorithm adapts the orthogonalization step and the training step for tracking weights vectors. Simulation experimental results show that compared to the original algorithm, the improved algorithm can reduce the reconstruction error, increase the energy fraction of reconstruction, and decrease the number of hidden variables, so it extract the principal components among streams more effectively.