Robust sensor selection based on maximum correntropy criterion for ocean data reconstruction

Selecting an optimal subset of sensors that can accurately reconstruct the full state of the ocean can reduce the cost of the monitoring system and improve monitoring efficiency. Typically, in data-driven sensor selection processes, the use of Euclidean distance to evaluate reconstruction error is susceptible to non-Gaussian noise and outliers present in ocean data. This paper proposes a Robust Sensor Selection (RSS) evaluation model based on the Maximum Correntropy Criterion (MCC) through subspace learning, enabling the selection of robust sensor measurement subsets and comprehensive data reconstruction. To more accurately quantify the impact of varying noise magnitudes, noise weights were incorporated into the model’s objective function. Additionally, the local geometric structure of data samples is utilized to further enhance reconstruction accuracy through the selected sensors. Subsequently, the MCC_RSS algorithm is proposed, which employs the Block Coordinate Update (BCU) method to achieve the optimal solution for the proposed model. Experiments conducted using ocean temperature and salinity datasets validate the proposed MCC_RSS algorithm. The results demonstrate that the sensor selection method proposed in this paper exhibits strong robustness, outperforming comparative methods under varying proportions of outliers and non-Gaussian noise.