The Study Of Active Power Security Correction Algorithm Based On Machine Self-Learning Methods

The active power security correction is a very important aspect of security control of the power system. It is mainly used to prevent or alleviate overload of transmission lines and tie line groups. Through adjusting the operation mood of a power system, when some transmission lines are overloaded, and we can present a strategy of load-shedding if those adjustments not work effectually. Traditionally, the method based on sensitivity analysis is usually used to solve active power security correction problem. In such a method, the sensitivity relationship between generators and lines is calculated in advance, and then the active power output of every generator is regulated properly to alleviate overloads in lines under the condition of meeting the prescribed constraint conditions. However, this method can not be self-learning, it can not achieve a satisfactory error requirement in a complex power system operating mode. When the system is in the non-secure state, the adjustion of the operation mood should be adjusted to reduce artificial means, so that the calculated active correction can be also changed by the changes of operation mood,and it can be automatically modified based on historical data, such as some network parameters. Therefore, we need an intelligent method, and machine learning method can solve these problems. Since the current machine learning method is used in many aspects, especially in artificial intelligence is more intelligent and more mature technology, but the method used in power securitycorrection is very little. So the research done in this paper is necessary. With the use of the machine self-learning method it can provide an accurate active correction as much as possible to ensure power system secure and economic.This paper analyzes the traditional sensitivity method and its existing problem, and consider this question from machine self-learning methods. The paper applies BP neural network method and support vector machine in active power security correction control, and the methods are proved to be valid and superior in calculation error by the calculation results for the IEEE-30 node system.