【摘要】 【目的】为提高电力系统暂态稳定评估效果，解决样本不平衡问题下的评估有效性，提出一种基于注意力机制与卷积门控循环单元的多任务暂态稳定一体化评估方法。【方法】所提方法融合卷积门控循环单元与卷积注意力模块，构建表征暂态功角稳定与暂态电压稳定问题的综合特征集。通过对传统二分类交叉熵损失函数的改进，实现动态权重调整，使模型在训练过程中更加关注失稳样本。同时，分析分类决策阈值对模型性能的影响，确定适合暂态稳定评估的最优分类决策阈值，以降低关键失稳事件的误判风险。【结果】仿真验证表明，所提出的融合卷积注意力机制并改进损失函数的卷积门控循环单元多任务模型，能够有效提升对暂态功角稳定和暂态电压稳定问题的综合评估准确性，明显降低失稳样本的漏判风险，在处理样本不平衡问题方面表现出较强的有效性与鲁棒性。【结论】所提方法通过空间与通道双重注意力机制有效增强了模型对关键特征的关注能力，实现了电力系统暂态功角与暂态电压稳定的高效一体化评估，可为电网稳定运行提供新的技术支撑。更多还原

【Abstract】 [Objective] To improve the effectiveness of transient stability assessment in power systems and address evaluation performance degradation under class-imbalance conditions, this paper proposes an integrated multi-task transient stability assessment method based on attention mechanisms and convolutional gated recurrent units(ConvGRU). [Methods] The proposed approach combines ConvGRU with the convolutional block attention module(CBAM) to construct a comprehensive feature set representing transient rotor-angle stability and transient voltage stability. The traditional binary cross-entropy loss function is modified to dynamically adjust weights, guiding model-training process toward unstable samples. Additionally, the impact of classification decision thresholds on model performance is analyzed to determine optimal thresholds for transient stability evaluation, thereby reducing misclassification risks for critical instability events. [Results] Simulation results demonstrate that the proposed multi-task model, enhanced by the convolutional attention mechanism and modified loss function, significantly improves the accuracy of integrated assessments for transient rotor-angle stability and transient voltage stability, substantially reduces missed detections of unstable samples, and exhibits robust performance under class imbalance conditions. [Conclusions] By enhancing attention to critical features through spatial and channel dualattention mechanisms, the proposed method achieves efficient integrated assessment of transient rotor-angle and voltage stability in power systems, providing new technical support for stable grid operation.更多还原