Power System Transient Stability Assessment Based On Machine Learning

With the continuous development of modern power system and access of complex modules,new risks are introduced into the power grid,which leads to the complexity of its structure and dynamic characteristics.In the face of failure,the difficulty of stability analysis and control of the system increase.Therefore,it is of great significance to grasp transient situation quickly and realize transient assessment of power system accurately.In the traditional power system transient stability analysis methods,time domain simulation method,with fine model construction and accurate calculation results,takes long time.With the increase of power grid scale,the calculation speed is restricted,and it is difficult to apply online.The direct method is fast to calculate,but the simplified model is used,which makes the result conservative.Machine learning method provides a new solution to the problem.In this paper,machine learning method is introduced into the transient stability assessment of power system,and the main research work and results are as follows:(1)This paper introduces rotor motion and transient stability mechanism of generator in the transient process of power system.PSD-BPA simulation software is used to build the standard system model,simulate phasor measurement unit(PMU)of the power system to collect information and obtain original data set.(2)A power system transient stability assessment method based on MultiGrained Scanning Cascade Light Gradient Boosting Machine(MGS-LGBM)is proposed.MGS-LGBM has the characteristics of simple setting of super parameters,strong generalization ability of model,high classification accuracy and fast training evaluation.Firstly,from the perspective of transient stability mechanism,a three-stage fault information representation method is adopted,which takes into account both timeliness and globality.23 artificial features are extracted to realize the transformation of the original sample set.Then,the artificial feature sample set is input into MGS-LGBM model,and the stability result is taken as the output.The multigrained scanning and cascade structure in the model are used to train efficiently parallel the sample features and results.The simulation results of New England 10 machine 39 bus system show that the proposed algorithm can improve the accuracy of transient assessment while taking into account the rapidity,and can still maintain good assessment performance when it contains irrelevant features and fewer training sets.(3)A power system transient stability assessment method based on Improved Deep Residual Shrinkage Networks(IDRSN)is proposed.Without manual feature extraction and combining with the advantages of deep learning feature extraction,electrical quantity collected by PMU is directly used.The bottom-level measured electrical quantity is constructed as a feature map as the input of model,and the deep structure of model is used to establish the mapping relationship between the input and the stable result.Faced with noise problems of PMU during acquisition and transmission process,the model uses the attention mechanism to automatically learn the noise threshold through a soft threshold function to reduce noise and irrelevant feature interference.The transient stability and instability samples are imbalanced,resulting in the tendency of data-driven transient stability assessment model training and serious misjudgment problems.Through focus loss function(FL),the weight coefficient is introduced to correct the tendency of model training.Modulation factors is used to focus on misclassified samples to improve model training efficiency and evaluation performance.Through the simulation verification,the proposed model can effectively reduce the noise interference of different degrees,correct the bias of the model training on the imbalanced data set,and reduce the misclassified samples.Under different PMU configuration schemes,all are obtained better evaluation effect.