Resolving Power Equipment Data Inconsistency via Heterogeneous Network Alignment

Abstract

This paper studies the problem of resolving data inconsistency from multiple sources in managing data related to power equipment for China’s state grid corporation. This paper proposes HENGE, a HEtetrogeneous Network GEneration model, to automatically align inconsistent devices from multiple sources, i.e., the same devices with multiple entries with different values in each source. HENGE builds multiple data sources into a heterogeneous graph, and captures complex physical and semantic relationships among devices. HENGE combines feature and relational information and improves alignment accuracy by feature-enhanced residual graph encoder and disentangled representation learning. HENGE can learn from a small amount of labeled data through a uniformity autoencoder trained on an unsupervised generation task. Experiments on two real-world datasets demonstrate the capability of HENGE in resolving inconsistent device entries in multiple sources.

Keywords

• Data inconsistency
• disentangled representation learning
• graph alignment
• graph neural network
• uniformity autoencoder