基于BP+Bi-LSTM的电力系统广域数据异常检测研究

Study on Anomaly Detection of Wide-Area Data in Power System Based on BP+Bi-LSTM

摘要: 电力系统同步向量测量装置(PMU)产生的数据对电力系统规划和安全起着至关重要的作用。但PMU装置往往充满噪声，除去系统本身能够排除的干扰，仍旧会产生PMU数据丢失、精确度降低等问题。本文提出一种双向长短期记忆网络(Bi-LSTM)和BP神经网络相结合的解决方案。首先通过BP神经网络拥有强大的非线性成像技术和网络结构，针对不同情况改变其特殊性能结构；其次利用Bi-LSTM神经网络解决多变量问题，将这两者相互结合，BP+Bi-LSTM具备比BP神经网络模式更强大的非线性映射能力和泛化学习能力，大大提高数据分析应用范围。最后以新英格兰10机39节点为例，验证本文方法对不良数据检测可行性和正确性。

Abstract: The data generated by the Phasor Measurement Unit (PMU) in power system plays a crucial role in power system planning and safety. However, PMU devices are often filled with noise, and even with interference that the system itself can eliminate, problems such as PMU data loss and reduced accuracy still occur. This paper proposes a solution that combines bi-directional long short-term memory network (Bi-LSTM) and BP neural network. Firstly, BP neural network has powerful nonlinear imaging technology and network structure, and its special performance structure can be changed according to different situations; Secondly, using Bi-LSTM neural network to solve multivariate problems, combining the two, BP+Bi-LSTM has stronger nonlinear mapping ability and generalization learning ability than BP neural network mode, greatly improving the application scope of data analysis; Finally, taking the example of 10 machines and 39 nodes in New England, the feasibility and correctness of our method for detecting bad data are verified.