Research On Transient Stability Assesment Of Power System Based On Improved Support Vector Machines

Nowadays,with the vigorous construction of UHV transmission lines,large-scale intermittent new energy sources are connected to the power system,the structure and dynamic characteristics of the power grid are gradually becoming more complex.How to quickly and accurately determine whether the system can remain stable after a fault occurs is very important for the safe and stable operation of the power system.In recent years,the rise of machine learning methods such as support vector machines has provided new ideas for transient stability assessment of power systems.Their core idea is to find the mapping relationship between the original feature quantity and the stable state through a data-driven method.However,the performance of the existing stability evaluation methods based on support vector machines is restricted by the choice of loss function and kernel function,and the problems of over-fitting and under-fitting are prominent.This paper studies the transient stability assessment method of power system based on support vector machine and improves its shortcomings.The main research contents are as follows:For the classical vector machine hinge loss function used for transient stability evaluation,the farther away from the classification surface,the greater the penalty is for the noise samples,which leads to the problem that the model is sensitive to noise and outlier data and affects the evaluation accuracy,and the generalization of a single kernel function To solve the problem of lack of ability,the classical vector machine,twin vector machine,bump process,kernel function and other methods are studied,the advantages and disadvantages of each algorithm are summarized.A transient stability evaluation method of twin support vector machines based on ramp loss function and mixed kernel function is proposed.The original fault feature set is obtained through a large number of simulation comparisons.The Latin hypercube method is used to randomly generate generator and load data,and three line settings are selected.Phase short-circuit fault constructs a sample set to enhance the generalization ability of the model;uses the ramp loss function and the mixed kernel function to introduce a twin support vector machine,and uses the bump process to solve it,thus forming a combined vector machine algorithm,improving the model’s anti-interference ability and enhancing the model’s performance.Learning ability and generalization ability;use the three indicators of accuracy rate,consistency test coefficient,instability and misjudgment rate to evaluate the model;finally,using the above combination algorithm,through the IEEE-39 node system example simulation to verify the effectiveness of the method proposed in this paper.