Research On Power System Transient Stability Assessment Based On Machine Learning Algorithms

Stability is one of the most important performance indexes of power system.With the rapid development of modern power system,the scale of power grid becomes larger and larger,and the topology of power system becomes more and more complex,it is difficult to achieve rapid and accurate transient stability assessment only by traditional methods.In the background of machine learning and wide area measurement system of power system,it is of great significance to use machine learning to deal with transient stability assessment of power system.This thesis studies three kinds of power system transient stability assessment methods based on machine learning algorithm on the basis of consulting and analyzing the research status at home and abroad.The main contents are as following:Firstly,a transient stability assessment method power system based on gradient boosting decision tree is studied in this thesis.To choose characteristic analysis of power system simulation data set,to outliers and standardized methods for data preprocessing operations,as the input of power system transient stability assessment model,build a model for transient stability assessment based on Gradient Boosting Decision Tree,and compared with the traditional machine learning decision tree,k nearest neighbor algorithm,support vector machine,multilayer perceptron.According to the comparison results of evaluation indexes,the gradient boosting decision tree model is more suitable for the transient stability assessment of this thesis,but there are problems in the difficulty of classification of data of different categories.Aiming at the classification difficulty difference problem of unbalanced data sets,a gradient boosting decision tree assessment model based on the improved focal loss function was constructed.The focal loss function was used to adjust the weight of different classes of data,and the classification difficulty difference caused by unbalanced data sets was effectively improved.Then,a transient stability assessment method of power system based on light gradient boosting machine-logistics regression algorithm is studied in this thesis.In view of the decrease of model classification performance caused by unbalanced data set,use the combination of synthetic minority oversampling technique and adaptive synthetic sampling approach to balance the original data set.Then,aiming at the problem of large number of features and high dimension of power system data,the light gradient boosting machine with improved frame of gradient boosting decision tree was used to extract important features from the original data set,construct new feature combinations as the input of logistics regression model,and assess the transient stability combining with the classification advantages of logistics regression model.This method improves the low training efficiency caused by high data dimension and further improves the accuracy.Finally,a transient stability assessment method of power system based on stacking ensemble learning is proposed.On the basis of the first two classification models,five models,random forest(RF),multilayer perceptron(MLP),support vector machine(SVM),focal loss-gradient boosting decision tree(FL-GBDT)and light gradient boosting machine-logistics regression(LGBM-LR),are used as basic models,and the logistics regression model was used as the meta-model to construct the stacking ensemble learning assessment model.Compared with different basic models on the test set,according to the comparison of evaluation indicators,stacking ensemble learning model can effectively combine the classification advantages of different basic models,while reducing the risk of model overfitting,which has a good effect on transient stability assessment of power system in this thesis.In summary,this thesis uses three improved machine learning algorithms to build a series of transient stability assessment models for quasi-unbalanced power system transient simulation data set,achieving more accurate transient stability assessment effect,which has certain reference value for modern power system transient stability analysis and control research.