The traceability relationship modeling of heterogeneous metering information in electric power test based on the knowledge graph-neural-symbolic reasoning self-evolution management

With the development trend of complexity and diversification in power systems, the power test’s Heterogeneous Measurement Information (HMI) traceability has become a key link to ensure the safe operation of the power system. The HMI traceability technology mostly relies on traditional database or single algorithms at present, which has some outstanding problems, such as loose association of multi-source data and a lack of dynamic adaptive ability. To encounter the above problems, this study adopts the Knowledge Graph-Neural-Symbolic Reasoning (KG-NSR) collaborative application system and introduces the self-evolution management mechanism to build a traceability relationship modeling framework. It innovatively improves and optimizes the hierarchical association KG construction method of HMI, and designs the fusion reasoning mechanism of NSR and self-evolution to carry the reasoning rules. Then, it constructs the KG-NSR collaborative self-evolution management framework to dynamically update the model parameters and reasoning rules. Due to this technology’s ability in effectively solving the core pain points of loose association of multi-source HMI and the lack of self-adaptation ability in the traceability system, it realizes the precise association and efficient traceability of multi-format HMI. The experimental findings demonstrate that the proposed technology enhances the accuracy of HMI traceability to 98.5%±0.5%, which is 8.9% points higher than that of the traditional technology. Besides, it shortens the inference response time to 0.4 ± 0.1 s, reduces the data redundancy rate to 4.8%±0.3%, achieves a robustness index of 0.97, and still maintains stable performance in 30% data distribution change scenarios. Therefore, compared with the other comparison methods, the comprehensive performance of the proposed technology has been improved by 10.5% at most. To sum up, this technology builds a complete technical link from modeling to application, which provides a new technical path for power test measurement traceability. Lastly, it upgrades the precision and efficiency of power test measurement traceability, thus having important engineering application value. The proposed framework is applicable to distribution substations, renewable energy grid-connected points, and large-scale industrial power consumption scenarios.