Fault Diagnosis Methods in Power Systems

Fault diagnosis in power systems encompasses a range of techniques designed to detect, locate and classify electrical failures rapidly and accurately in transmission and distribution networks. Traditional model-based approaches rely on detailed topological descriptions and physical models of protective devices, using analytical frameworks to interpret relay and breaker signals. More recently, data-driven techniques have proliferated, leveraging advances in machine learning, neural networks and big-data analytics to mine vast streams of operational and alarm data. Hybrid strategies are emerging that integrate model-based reasoning with data-driven pattern recognition, combining the interpretability of physical models with the adaptability of learning algorithms. Across all methods, real-time performance, robustness to missing or erroneous alarms and the ability to cope with high penetration of renewable sources are key criteria. Globally, improved fault diagnosis supports grid resilience, reduces outage durations and facilitates smart-grid self-healing, thereby enhancing system reliability and operational efficiency.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

A 2023 study introduced a fault-segment location method for distribution networks incorporating spiking neural P systems and Bayesian estimation. The network is decomposed into single-branch circuits, modelled by spiking neural P systems to generate preliminary fault-location candidates, which are then refined by Bayesian estimation to correct misclassifications. Tests on standard network models demonstrate improved localisation accuracy and rapid response suitable for systems with high distributed-generation penetration.

In 2021, researchers developed a data-driven long time-series fault prediction scheme using an enhanced stacked-Informer network. By tailoring the transformer-based Informer architecture to extract deep temporal features and integrating a gradient-centralised optimisation routine, the approach achieves superior prediction accuracy for electrical line-trip events. Validation with real-world wind–solar substation data shows significant gains in forecasting horizon and training efficiency over conventional sequence models.

A foundational 2019 work proposed temporal constrained fuzzy Petri nets for fault diagnosis, addressing uncertainty and timing in relay alarms. By encoding alarm truth degrees and temporal constraints within a Petri-net framework, the method computes fault probabilities and time-point constraints via matrix algorithms. Case studies on multiple network configurations confirm the approach’s capability to tolerate false or delayed alarms while maintaining diagnostic precision.

Fault Diagnosis Methods in Power Systems publication trend

The graph below shows the total number of articles in fault diagnosis methods in power systems across all publications each year (not limited to Nature Index journals).

Technical terms

Spiking neural P system: A bio-inspired computational model using neuron-like cells and synapses to process discrete spiking events for pattern recognition in network segments.

Bayesian estimation: A statistical inference technique that updates fault-location probabilities by integrating prior information and new observational data.

Stacked-Informer network: A transformer-based deep learning model optimised for long sequence forecasting, employing self-attention mechanisms with improved efficiency and gradient centralisation.

Fuzzy Petri net: A graphical modelling tool combining Petri-net structure with fuzzy logic to handle uncertain and imprecise information in diagnostic reasoning.

References

1. An Analytic Method for Power System Fault Diagnosis Employing Topology Description. Energies (2019).
2. A fault segment location method for distribution networks based on spiking neural P systems and Bayesian estimation. Protection and Control of Modern Power Systems (2023).
3. A Data-Driven Long Time-Series Electrical Line Trip Fault Prediction Method Using an Improved Stacked-Informer Network. Sensors (2021).
4. Fault Diagnosis of Power Systems Based on Temporal Constrained Fuzzy Petri Nets. IEEE Access (2019).