In renewable-dominant hybrid alternating current/direct current (AC/DC) power systems, cascading failures pose significant threats to security and stability, severely limiting the active power transmission of high-voltage direct current (HVDC) systems. A research team from Shandong University has proposed a rapid risk assessment and critical line identification method based on gradient boosting decision tree (GBDT) and frequent pattern growth (FP-Growth) algorithms, published in Engineering.

Cascading failures in such systems can trigger security and stability events including successive commutation failure, sending-end overvoltage, and equivalent power angle instability. These events interact with HVDC systems and renewable energy components, reducing the maximum DC power transmission capacity. Traditional risk assessment methods focus on load loss and ignore DC power limitations, while critical line identification approaches often fail to distinguish the importance of each line in failure patterns.

The proposed method first defines a cascading failure risk index centered on DC power limitation, calculated as R=p⋅(P_rated,HVDC​−P_{limited,HVDC​}), where p is the failure probability, and P_rated,HVDC​ and P_{limited,HVDC​} are the rated and limited DC power respectively. To handle the strong nonlinear relationship between cascading failures and maximum DC power, a GBDT with an update strategy is utilized for rapid prediction, enhancing generalization to renewable energy uncertainty.

For critical line identification, an improved FP-Growth algorithm mines frequent patterns in cascading failures. The algorithm constructs an FP-Tree through two dataset scans, then extracts conditional pattern bases and conditional FP-Trees to identify frequent patterns. An importance index for each fault in these patterns is defined, considering fault probability, impact factors, and cascading failure risk, enabling accurate identification of critical lines.

Validation on a modified Ningxia–Shandong hybrid AC/DC system, which features 268 buses, 92 machines, 462 AC lines, and a 4000 MW HVDC link with over 60% renewable energy penetration, demonstrates the method’s effectiveness. The GBDT model achieves an average prediction error of 3.523 MW for maximum DC power, outperforming the stacked denoising autoencoder (SDAE) method. The FP-Growth algorithm shows higher computational efficiency than Apriori, requiring only 0.534 seconds to process 10,000 cascading failure samples compared to Apriori’s 404.167 seconds.

This method provides a reliable tool for power system operators to assess cascading failure risks rapidly and identify critical lines accurately. It supports proactive measures to prevent cascading failures and improve the efficiency of renewable energy transmission through HVDC systems. Future research will focus on early warning of high-risk cascading failures and preventive control based on the identified critical lines.

The paper “Frequent Pattern Growth-Based Identification of Critical Lines in Cascading Failures for Renewable-Dominant Hybrid AC/DC Power Systems,” is authored by Tianhao Liu, Jiongcheng Yan, Yutian Liu. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.06.033. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.