基于改进LightGBM的电力系统暂态稳定评估方法

实际获得的电力系统运行数据,普遍存在由广域测量系统(WAMS)测量误差造成的噪声,并且具备类别不平衡的特点,导致基于机器学习的电力系统暂态稳定评估算法的分类性能受到很大的影响。提出了一种基于改进轻梯度提升机(modified light gradient boosting machine)模型的暂态稳定评估方法,采用直方图算法对数据进行离散化,增强模型对噪声的鲁棒性;在训练中对失稳样本赋予更高的权重,平衡样本数量差异造成的影响;并在损失函数中引入正则项来控制模型复杂度,减少过拟合,从而适应电力系统多样的运行情况。在新英格兰10机39节点系统和美国南卡罗莱纳州500节点实际电网上的仿真结果表明,与其他机器学习方法相比,所提方法在噪声干扰下不容易过拟合,具有更好的鲁棒性:在保持较高总体评估准确率的同时,对失稳样本具有更高的识别率;与其他集成学习模型相比,所提方法在速度上也具有明显优势。

Power System Transient Stability Assessment Method Based on Modified LightGBM

The actual available operation data of power system generally contain the noise caused by the measurement error of wide area measurement system(WAMS), and are also characterized by class imbalance, leading to great influence on the classification performance of the power system transient stability evaluation algorithms based on machine learning. This paper proposes a transient stability assessment method based on modified Light GBM(Light Gradient Boosting Machine). This histogram algorithm is used for data discretization to strengthen the robustness to noise, and larger weights are given to unstable samples in training process to mitigate the impact of imbalanced samples. Meanwhile, a regularization term is introduced into the loss function to reduce overfitting. The simulation results on New England 10-machine 39-bus system and an actual 500-bus grid in State of South Carolina, USA show that, compared with other machine learning methods, the proposed model is hardly prone to over-fitting under noise interference with better robustness. It has a higher recognition rate for unstable samples with satisfactory overall accuracy. Additionally, the proposed method has obvious superiority in computation speed compared with other ensemble models.