| Transient stability analysis is an important assignment of power system security analysis.It plays an important role and value for the safe and stable operation of power system and providing the reference for the control strategy of power system operation scheduling.How to accurately and quickly assessment the transient stability status and margin after the system failure is an urgent problem to be solved in the power system security analysis.The traditional analytical methods include time-domain simulation and direct method.The former method required a large amount of computation and time-consuming process,and the latter method can hardly to satisfy conditions to construct the energy functions in complex systems,which is difficult to meet the real-time requirements of the security assessment of large power grids.In recent years,with the advantages of high evaluation accuracy and short time-consuming,artificial intelligence methods have gradually become one of the main tools for fast transient stability assessment.However,the ability of this kind of algorithms to express the characteristics of the input data is limited by their shallow structure,which is result in poor generalization ability when solving high-dimensional data classification problems.Most of them are binary classification problems for determining transient stability or not,and there is a lack of quantitative evaluation.Moreover,this kind of algorithms are not used to assessment transient stability margin of power system,most of them assessment the transient stability status.Aiming at the above-mentioned problems,a deep learning method with self-learning and express abstract features ability is introduced into the transient stability evaluation of power system.A method for assessing power system transient stability based on short-term disturbance trajectories and convolutional neural network is proposed in this paper.The main work of this paper is as follows:(1)The convolutional neural network is constructed to establish the mapping relationship between short-term disturbance trajectories of generator terminal and power system transient stability.Furthermore,a method for constructing the samples matrix considered the disturbed degree of generator,which can reflect the transient stability information,is introduced in this model.It leads to the great robustness of features extracted by the model,the less false positive and negative samples,the superior generalization ability and evaluation performance of the model.(2)The optimal parameters are selected by the comprehensive evaluation index of the network according to the principle of dimension calculation of convolutional neural network.In order to further reduce the number of false positive and negative samples and improved theevaluation accuracy of the network,the mapping relationship between the samples matrix and transient stability is constructed on the basis of considering the disturbed degree of generator after a failure.(3)A improved network is proposed to establish the mapping relationship between short-term disturbance trajectories and power system transient stability margin.It combines convolutional neural network with back propagation neural network.And the improved network is combined with transient stability classification model to form a composite network.Firstly,using convolutional neural network to classify the input samples.Then,using the improved networks to predict the transient stability margin of different kinds of samples to realize the proposed method.The IEEE 39-bus system is used to verify the effectiveness of the proposed method that can realize the transient stability status and margin evaluation of power system based on short-term disturbed trajectories and improved convolutional neural network,which can provide reference for control strategy of power system operation scheduling. |
