A smart grid knowledge retrieval framework for heterogeneous operational data: enhancing fault tracing and equipment diagnosis for digital smart grid operation

Abstract The large-scale integration of renewable energy sources, such as wind and solar, has intensified the complexity of power system operations, leading to heterogeneous operational data—spanning SCADA, PMUs, and event logs—being generated across isolated systems. In modern smart grid operation, including renewable-rich environments, heterogeneous operational data from SCADA platforms, monitoring systems, incident reports, and maintenance logs are often distributed across isolated systems. Unified access to this heterogeneous data is critical for data-driven fault diagnosis, predictive maintenance, and grid stability control, yet remains challenging due to incompatible representations and fragmented storage. This paper proposes a unified retrieval framework for multi-source heterogeneous smart grid data. To address the challenges of cross-domain retrieval in power systems, the framework integrates heterogeneous data through source-specific adapters into a unified internal representation. It introduces a lightweight lexical fusion and source-balanced ranking mechanism centered on lexical relevance estimation and cross-source evidence balancing. This design ensures that critical fault signatures and operational patterns distributed across substations and time-series logs can be retrieved efficiently, without the high overhead of complex semantic modeling or deep neural ranking models. We implemented a prototype with RESTful interfaces and evaluated it using a heterogeneous smart grid dataset composed of structured records, JSON reports, textual logs, and time-series monitoring data. The evaluation is conducted on a prototype benchmark consisting of normalized heterogeneous smart-grid-style records, including structured equipment records, JSON incident reports, textual operation logs, and time-series monitoring entries. Unlike ontology-heavy approaches such as CIM or SAREF-based systems, the proposed method prioritizes retrieval efficiency and operational responsiveness. Experimental results demonstrate that the proposed framework improves early ranking quality over keyword-based, BM25, federated, and single-source baselines in retrieval effectiveness. A case study focusing on breaker overload fault tracing validates the framework’s practical value, showing its ability to correlate scattered evidence from asset records, incident reports, textual logs, and monitoring streams for operational evidence association. The proposed approach provides a scalable, retrieval-oriented foundation for smart-grid operational support, enabling more efficient access to critical data for power engineers and contributing to enhanced grid resilience and automated maintenance workflows. However, the current implementation should be interpreted as a research prototype for low-concurrency operational retrieval rather than a production-grade real-time control or protection system.