Table of contents
3. Comparison of data-driven and other methods
1. Data-driven overview
The main goal of data-driven application in power system stability analysis is to establish a power system stability analysis model from the perspective of power grid operation data, and to mine the dynamic stability characteristics of the power system from the perspective of correlation analysis between data, which is different from the purely causal model based on physical models in the past. Analysis [7-8]. All kinds of massive data in the power system provide data sources for data-driven methods. When generated offline, the intrinsic relationship between dominant dynamic features and stability is mined from a large amount of simulation and operation data, and a safety and stability evaluation model is constructed. It is essentially a mapping relationship between the space-time dynamic characteristics of the complex large power grid (specifically reflected in the safety and stability of the system) and the measurement data. In the online application, the safe and stable state can be quickly given by inputting the measurement data, and the data-driven power grid stability can be quickly evaluated. The logic diagram of the data-driven method is shown in Figure 1.
Figure 1 Logic diagram of data-driven method
2. Data-driven features
Wide Area Measurement System (Wide Area Measurement System, WAMS) The measurement data of WAMS is the real reflection of the operating state of the power system, and it is also the external performance of the physical system model, which provides the data basis for the data-driven method. Similarly, the data-driven approach also includes many advantages of methods such as data mining and machine learning.
Compared with traditional analysis methods based on purely physical models, data-driven analysis methods have the following characteristics:
(1) Data-driven analysis methods do not depend on the physical model of the power system, so they are not affected by randomness, uncertainty, and complexity. There are no unreasonable assumptions and over-simplified analysis methods.
(2) From the perspective of data, there is no distinction between linear and nonlinear data, so the data-driven analysis method can be better applied to strong nonlinear systems such as power systems.
(3) For methods based on physical models, the mechanism needs to be clearly understood before modeling. A data-driven approach, on the other hand, does not need to understand or explain the mechanism after the fact.
(4) Faced with different scenarios and needs, data-driven methods such as data mining and machine learning have the ability to generalize to multiple scenarios. Scholars have successfully applied different machine learning technologies to various aspects of life, such as weather forecasting , search engines, autonomous driving, etc., the generalization ability of the model is increasing with the maturity of the technology [9].
(5) The accuracy of the analysis results of the traditional analysis method based on physical modeling relies too much on the accuracy of the physical model, and there are always various unrealistic assumptions as the analysis premise before modeling; the data-driven analysis method is to mine data There is a certain relationship between the data that is not clear but exists in fact. The correlation between the data is different from the causal relationship obtained by the physical model, but the combination of the two will have a greater impact on improving the cognition of the stability characteristics of the power system. help. Therefore, data-driven is an effective supplement to physical modeling methods, and the combination of data-driven and physical modeling methods can provide more valuable technical solutions [10].
3. Comparison of data-driven and other methods
At present, big data technology and machine learning methods are the mainstream algorithms driven by data. With the application of artificial intelligence technology represented by big data and machine learning in weather forecasting, search engines, automatic driving, astronomical data, biotechnology, credit card fraud identification, Successful applications in character recognition, web applications, network intrusion detection, etc. [7-8], artificial intelligence is considered to be the most disruptive technology at present. At present, artificial intelligence technology in the field of electric power is in the ascendant. Some domestic and foreign experts and scholars have always attached great importance to the research and application of artificial intelligence technology, and actively carry out big data, machine learning and other technologies to realize the stability assessment of large-scale online power systems, and have made great achievements. There are few achievements [11], and the common representative methods are: put forward a series of questions to decompose the data, so as to make a decision tree method [12], generalized linear classifier support vector for binary classification of data Machine method [13-15] (Support Vector Machine, SVM), neural network method [16-17], which is composed of many adjustable neuron connection weights and has the characteristics of large-scale self-organization and self-learning ability, etc., these Methods of data processing have their own characteristics. The neural network method has poor interpretability, and the many internal layers of the algorithm lead to high calculation and complexity of the model, which is not suitable for the analysis and calculation of large systems, and is prone to over/underfitting problems. The support vector machine algorithm has a very large amount of computation in the training phase, especially for the analysis of nonlinear systems, and the demand for the number of support vector machines is also huge. Although the decision tree method has the advantages of simplicity, strong interpretability, and easy understanding, it is a single classifier with SVM and neural network, and there are bottleneck problems of over-fitting and performance improvement. In particular, in terms of artificial intelligence algorithm selection, algorithm optimization is a key research content.
Four. Summary
Therefore, the data-driven method is affected by the stability, convergence, accuracy, and efficiency of the algorithm. It is worth further development and exploration to analyze the static scene of the power system and find a more optimized algorithm for application and integration.
5. References
[5] Zhang Rui. Static stability assessment of large power grid based on wide-area information [D]. Northeast Electric Power University, 2017. [6] Yang Shengchun, Tang Biqiang, Yao Jianguo, et al. Architecture and key technologies of automatic intelligent dispatching of power grid based on situational awareness[J]. Power Grid Technique, 2014,38(01):33-39.
[7] Xue Yusheng, Lai Yening. Integration of Big Energy Thinking and Big Data Thinking (1) Big Data and Power Big Data [J]. Electric Power System Automation, 2016,40(01):1-8.
[8] Xue Yusheng, Lai Yening. Integration of Big Energy Thinking and Big Data Thinking (2) Big Data and Power Big Data [J]. Electric Power System Automation, 2016, 40(08):1-8.
[9] Zhou Zhihua. Machine Learning: Machine learning[M]. Tsinghua University Press, 2016.
[10] Zhang Dongxia, Miao Xin, Liu Liping, Zhang Yan, Liu Keke. Research on the development of smart grid big data technology [J/OL]. Chinese Journal of Electrical Engineering, 2015,35(01):2-12.
[11] Cheng Lefeng, Yu Tao, Zhang Xiaoshun, Yin Linfei. Application and prospect of machine learning in the field of energy and power system[J]. Power System Automation, 2019, 43(01):15-43.
[12] He Miao, Zhang Junshan, Vittal V. Robust online dynamic security assessment using adaptive ensemble decision-tree learning [J]. IEEE Trans.on Power Systems, 2013,28(4):4089-4098．
[13] Zhao Wanming, Huang Yanquan, Chen Guihui. Static voltage stability assessment of power system based on support vector machine [J]. Power System Protection and Control, 2008,36(16):16-19.
[14] Moulin LS, Alves DSAP, El-Sharkawi MA, et al. Support vector machines for transient stability analysis of large-scale power systems[J]. IEEE Transactions on PowerSystems, 2004, 19(2): 818–825.
[15] Dai Yuanhang, Chen Lei, Zhang Weiling, et al. Power system transient stability assessment based on multi-support vector machine synthesis[J]. Chinese Journal of Electrical Engineering, 2016, 36(5): 1173-1180.
[16] Li Yanglin, Jiang Quanyuan, Yan Rong, et al. Small Disturbance Stability Evaluation of Power System Based on Convolutional Neural Network [J]. Automation of Power Systems, 2019, 43(02):50-59.
[17] Yao Dequan, Jia Hongjie, Zhao Shuai. Power system transient stability assessment and margin prediction based on compound neural network[J]. Electric Power System Automation, 2013, 37(20): 41-46.