| 摘要: |
| 深度学习在暂态稳定评估中发挥着越来越重要的作用，然而电网规模逐渐扩大导致数据出现维数灾难，这对模型的性能提出了更高的要求。目前，暂态稳定特征构建需要依靠人工经验，具有主观性；深度学习的模型在设计和训练上耗时、耗力。针对以上两点，结合极限梯度提升(XGBoost)算法和实体嵌入(EE)网络，提出了一种基于XGBoost-EE的电力系统暂态稳定评估方法。首先通过XGBoost算法的路径规则生成类别特征，将原始特征进行降维。然后采用EE网络对新的特征进行分类，从而完成快速、精准的暂态稳定评估。该方法充分利用了机器学习算法处理速度快和神经网络评估精度高的优点，能够直接面向底层量测数据。最后，在IEEE新英格兰10机39节点和IEEE 50机145节点系统的仿真结果表明，所提方法相比于其他方法具有更高的预测精度和更好的抗噪性能，且在训练时不容易过拟合。 |
| 关键词: XGBoost算法 实体嵌入 暂态稳定评估 深度学习 大数据 |
| DOI：10.16081/j.epae.202011032 |
| 分类号:TM712 |
| 基金项目: |
|
| Power system transient stability assessment method based on XGBoost-EE |
|
WU Chunming1,2, REN Jihong2
|
|
1.Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education, Northeast Electric Power University, Jilin 132012, China;2.School of Electrical Engineering, Northeast Electric Power University, Jilin 132012, China
|
| Abstract: |
| Deep learning plays an increasingly important role in transient stability evaluation. However, the increase of power system scale generally results in dimension disasters. In this case, an efficient and tractable computation model is highly desirable. Currently, the construction of transient stability features generally relies on the experience of power system operators, which is more or less subjective. However, the deep learning approach is generally time-consuming and labor-intensive in aspects of design and training. Based on the above two points, a transient stability assessment method of power system based on XGBoost-EE is developed by combining XGBoost(eXtreme Gradient Boosting) algorithm and EE(Entity Embedding) network. Firstly, the path rules of the tree are extracted and the category features are generated by XGBoost algorithm. In this way, the original features are dimensionally reduced. Then, the EE network is used to classify the new features, which provides a fast and accurate assessment. The proposed method, hence, takes full advantage of the fast processing speed of machine learning algorithms and the high accuracy of neural network evaluation. Simulative results based on IEEE New England 10-machine 39-bus system and IEEE 50-machine 145-bus system show that the proposed method exhibits higher prediction accuracy and better anti-noise performance than other approaches. Additionally, the proposed method is not easy to become over-fit during the training process. |
| Key words: XGBoost algorithm entity embedding transient stability assessment deep learning big data |
