Research On Power System Transient Stability Evaluation Method Based On Deep Learning

With the rapid development of the social economy,the demand for electricity in daily life and production of the people has increased sharply.These demands have led to the continuous expansion of China’s power system,which also puts a severe test on the safe and stable operation of the power system.The transient stability of the power system is essential for the safe and stable operation of the power system.The serious consequences of several blackouts related to transient instability that occurred in history are vivid.The traditional time domain simulation method and the direct method cannot meet the requirements of real-time and accuracy of the power system transient stability assessment method.The existing machine learning-based evaluation method is lacking in accuracy due to weak feature extraction ability.Deep learning has powerful feature extraction capabilities and also provides new methods for transient stability assessment.Firstly,this thesis analyzes the transient stability process of the power system and extracts the parameters of the power system that have an important influence on the transient stability process.Based on these parameters,the original feature set is constructed to ensure the good reflection for the transient stability process,which can improve the accuracy of the transient stability assessment method.The parameters selected during the construction process can be obtained from the wide area measurement system,avoiding the need for complex mathematical solution processes in the acquisition process,and does not affect the real-time nature of the transient stability assessment method.Then construct a primitive feature set that can effectively reflect the transient stability process,and the original sample set is simulated on the New England 10-machine 39-node system.Secondly,the idea of convolutional neural network is applied to the autoencoder,and the convolutional layer,the pooling layer and the anti-pooling layer are used instead of the traditional fully connected layer to construct a deep convolutional autoencoder network.It is used to perform feature screening on the original sample set.The filtered data can effectively save computing power during calculation and improve the accuracy of transient stability assessment.Adding noise reduction processing to the network,that is,using the noise-added samples for training,the network has anti-interference ability and better robustness.Then construct a deep denoising convolution autoencoding network.The gap between the network reconstructed data and the original data is used as the loss value to reflect the training effect of the network.The original sample set is used for training,and perform feature screening on the original sample set after training is complete.Finally,the final transient stability assessment is performed using a capsule network with stronger feature extraction capabilities.The convolutional layer is used to extract low-level features and reduce the size of the input data.The output of the convolutional layer is converted into a capsule form to contain more implicit information and improve the evaluation accuracy of the network,not only can it effectively prevent over-fitting,but also achieve higher evaluation accuracy.The capsule network was trained using the featured screened sample set and compared with several common deep neural networks.The comparison results demonstrate the superiority of the proposed method in real-time and accuracy.Finally,this paper summarizes the transient stability assessment method in this paper in the construction of feature sets,the screening of feature data,and the assessment of transient stability,and look forward to the future.