| 引用本文: | 杜一星,胡志坚,陈纬楠,王方洲,张翌晖.基于改进CatBoost的电力系统暂态稳定评估方法[J].电力自动化设备,2021,41(12): |
| DU Yixing,HU Zhijian,CHEN Weinan,WANG Fangzhou,ZHANG Yihui.Transient stability assessment method of power system based on improved CatBoost[J].Electric Power Automation Equipment,2021,41(12): |
|
| 摘要: |
| 在实际电网的运行过程中，通过同步相量测量单元实时采集到的电网动态参数通常含有部分噪声，且有时会因通信故障造成数值的随机缺失，对基于人工智能的电力系统暂态稳定评估模型造成很大影响。为此，提出一种基于改进CatBoost的暂态稳定评估方法。通过分箱算法对输入特征数据进行离散化处理，提高模型对噪声的鲁棒性；采用加权的焦点损失函数代替交叉熵损失函数，提升模型的可信度并减少模型对失稳样本的漏判；将量测数据部分缺失的样本划分到单独的节点中继续建模，从而充分挖掘不完整样本中的暂态信息。在新英格兰10机39节点上的实验结果表明，所提方法的准确率和查全率均优于其他几类机器学习算法，而且所提方法对噪声和数值缺失表现出良好的鲁棒性且具有较快的训练速度和预测速度。 |
| 关键词: 机器学习 人工智能 电力系统 暂态稳定评估 集成学习 CatBoost算法 |
| DOI：10.16081/j.epae.202107026 |
| 分类号:TM712 |
| 基金项目:国家自然科学基金资助项目(51977156) |
|
| Transient stability assessment method of power system based on improved CatBoost |
|
DU Yixing1, HU Zhijian1, CHEN Weinan1, WANG Fangzhou1, ZHANG Yihui2
|
|
1.School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China;2.Electric Power Research Institute of Guangxi Power Grid Co.,Ltd.,Nanning 530023, China
|
| Abstract: |
| In the practical operation process of power grid, the dynamic parameters of power grid collected in real time by phase measurement units usually contain some noise, and sometimes the values are randomly deletion due to communication failures, making great influence on the transient stability assessment models of power system based on artificial intelligence, for which, a transient stability assessment method based on improved CatBoost is proposed. The binning algorithm is used to discretize the input feature data for improving the robustness of model to noise. The weighted focal loss function is used for replacing cross-entropy loss function, which improves the confidence of the model and reduces the misjudgment of model to unstable samples. The samples with part of the measurement data missing are divided into separate nodes for continue modeling, thus the transient information can be fully exploited from incomplete samples. The experimental results of New England 10-generator 39-bus system show that the accuracy rate and recall rate of the proposed method are superior than other machine learning algorithms, and the proposed method performs good robustness to noise and values missing and has fast training speed and prediction speed. |
| Key words: machine learning artificial intelligence electric power systems transient stability assessment ensemble learning CatBoost algorithm |
