Grey Kernel Partial Least Squares-based prediction for temporal data aggregation in sensor networks

Data aggregation is a current hot research area in sensor networks. Aiming at the time series data in sensor networks, we present GRBFKPLS (grey RBF kernel partial least squares), a novel prediction model data aggregation of sensor networks. In this model, grey model prediction theory is introduced into partial least squares. By the approach, the input data are firstly mapped to a nonlinear higher dimensional feature space, a linear partial least squares model is then constructed by RBF kernel transformation. Moreover, moving widow method is utilized to update samples continuously in this dynamical prediction model. The model is validated with fuel pressure data of injector. The results show that the model can execute dynamic multi-step prediction, and it has high precision prediction and flexibility. Thus, it can observably reduce the number of transmissions in sensor networks and save energy. Besides, it also has better performance in latency and computation. Comparing with RBFKPLS (RBF kernel partial least squares), GRBFKPLS is more effective for senor networks, so it has a good foreground to improve the prediction performance of data aggregation.