电工技术学报2026，Vol.41Issue(2)：541-557,17.DOI:10.19595/j.cnki.1000-6753.tces.250180
模型-数据可信集成的低惯量电力系统频率预测及主动减载方法
Frequency Prediction and Active Load Shedding of Low Inertia Power System Based on Confidence-Driven Model-Data Integration
摘要
Abstract
When a low-inertia power system is subjected to significant active power disturbances,the reserve capacity is abundant.Still,the adjustment lag and the"unpredictability"of the existing under-frequency load shedding schemes cause the overcutting problem.Active load shedding methods based on frequency prediction have been proposed.However,most encounter contradictions between the low-frequency transient minimum point frequency and the steady-state frequency safety constraints.This paper proposes a transient frequency prediction and active load shedding method based on confidence-driven model-data integration.By conducting confidence-aware assessments on the frequency prediction results and actively implementing optimal load reduction predictions,the load reduction cost can be reduced. In the transient frequency part,during the offline stage,a data-driven frequency prediction model is trained using historical operational data.Paired with a rolling-updated system frequency response(SFR)equivalent model,a confidence threshold is determined based on deviations between the equivalent model and data-driven predictions across sample sets.During the online application,real-time measurement data are fed into the comparable model to estimate power imbalances,while the data-driven model predicts frequency nadirs.A confidence evaluation module then outputs validated frequency predictions. In the active low-frequency load shedding quantity prediction part,during the offline stage,an optimal load shedding strategy database is iteratively generated based on the set minimum point frequency constraint and quasi-steady-state frequency constraint.Then,the data-driven optimal load shedding quantity predictor is trained,and the threshold for the startup frequency of the active load shedding strategy is set.During the online application,active load shedding is executed based on the frequency prediction results. The ablation experiments on the attention module demonstrate that incorporating feature attention and channel attention significantly enhances the overall prediction performance of the 1DCNN(1 dimensional convolutional neural networks)model.The prediction results on the IEEE 10-machine 39-bus system show that compared with the ELM model,the GRU model,and the 1DCNN-DA(dual dimensional attention)model,the proposed method improves the MAPE error of the minimum point frequency by 0.292%,0.012%,and 0.022%respectively.The performance of the active deactivation strategy reveals that the proposed method ensures a lower deactivation cost under conditions that meet both the minimum point frequency constraint and the quasi-steady-state frequency constraint. The following conclusions can be drawn from the simulation analysis.(1)The framework guarantees accurate frequency nadir prediction and provides field operators with confidence-based decision-making tools during emergencies.(2)The proposed active under-frequency load shedding strategy achieves the optimal load shedding decision under the frequency safety constraint through the load shedding prediction model.Compared to the traditional low-frequency load shedding method,the proposed strategy exhibits a better frequency response.(3)Active load shedding reduces the operation time of load shedding in advance.It raises the system nadir frequency above the threshold with less load shedding,which can alleviate the increase of load shedding caused by the delay of traditional methods.关键词
低频减载/频率预测/可信度评估/模型-数据驱动/低惯量电力系统Key words
Under frequency load shedding/frequency prediction/confidence evaluation/model-data driven/low inertia power system分类
信息技术与安全科学引用本文复制引用
王翔宇,张庚午,陈武晖,郭小龙,刘德福..模型-数据可信集成的低惯量电力系统频率预测及主动减载方法[J].电工技术学报,2026,41(2):541-557,17.基金项目
新疆维吾尔自治区重大科技专项(2022A01007-1,2022A01007-2)资助. （2022A01007-1,2022A01007-2）
