Heterogeneous Data Feature Extraction Technology based on Long-Short Term Memory

Power data resources have complex problems such as heterogeneous, multi-source dispersion, high dimensionality, and diverse forms. This paper proposes one heterogeneous data feature extraction framework based on long-short term memory graph convolutional neural network, which realizes feature extraction and fusion of distributed heterogeneous data with numerical, image and text information's mixture. Experiments show that this method is more suitable for feature extraction of heterogeneous data in grid business than other graph convolutional networks or long-short term memory neural networks. It provides further monitoring and early warning, state analysis and professional management for the power data business technical support.