Missing Data Completion Based on Nonlinear Feature Space Mapping for the Power Grid

The power grid is a non-linear, highly-coupled, multi-dimensional mathematical-physical system, and its informatization is mostly dependent on empirical data. Nonetheless, this data is occasionally misplaced because of signal transmission malfunctions, sensor faults, or human errors. Retrieving these missing data is indispensable for the power system digitization, informatization, and intellectualization. This paper proposes a missing data completion method by a non-linear feature space mapping algorithm that maps nonlinearly related variables to linear space to obtain data linearly coupled with the variable, including missing data. Comparative experiments based on actual multi-dimensional grid data show that compared with multi-dimensional linear interpolation, the proposed algorithm can reduce the root mean squared error by 87.95% and the mean absolute error by 88.20% on average. And R
²
is more significant than 0.99. Besides, for all dimensional variables in the data set, the proposed algorithms have relatively high accuracy. Experiments prove that this algorithm is sufficient for the application of the power grid.