.png){kind=icon}
.png){kind=icon}
Return
BERTC: A new Bayesian exponential regularized tensor completion method for sparse geomagnetic time series data
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.132509.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Geomagnetic data is vital for predicting earthquakes and magnetic storms. However, the data is often incomplete due to hardware failures and uncontrollable interference. Tensor decomposition can capture the potential relationship among different data sets, and it has been widely used to address the problem of missing entries in sparse tensors in recent years. However, since this kind of method is prone to overfitting, the imputation accuracy is low when the missing rate of geomagnetic data is relatively large. In this regard, a new Bayesian exponential regularized tensor completion framework for sparse geomagnetic data, i.e. BERTC, is proposed to address this problem in the study. First, the spatiotemporal geomagnetic data is reshaped into a 3D tensor with days and hours that features random missing elements. Second, a Gibbs sampling algorithm is developed to achieve probabilistic inference on matrices’ factors and corresponding parameters in this model. Thus, the sparse tensor can be gradually optimized to fill the missing entries during iterations. Third, an exponential regularizer is proposed to reduce oscillations before and after iterations to enhance imputation quality further. Finally, the derived factor matrices are aggregated from Gibbs sampling to complete the sparse tensor. Numerical geomagnetic datasets from 18 different cities are employed, and extensive comparison experiments are conducted to evaluate the imputation performance of the BERTC. The results show the clear advantage of the proposed BERTC compared to the state-of-the-art methods in terms of imputation accuracy, with an approximate improvement of the imputation accuracy as no less than 20 %.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W