A Transient Frequency Acquisition Method Based on Stacked Denoising Autoencoder

Abstract

The invention discloses a transient frequency acquisition method based on a stack noise-reduction autoencoder, which can be used to comprehensively judge the transient frequency stability of a power system by predicting the multi-dimensional frequency index value after a disturbance accident. An important basis for emergency control strategies such as machine and low-frequency load shedding. The present invention first adopts the random forest algorithm to screen out important characteristic variables as input data to realize dimensionality reduction of the input data; then, stacks multiple noise-reduction autoencoders to build a deep learning network structure; adopts the method of "pre-training-parameter fine-tuning" for training Network parameters. In the pre-training process, dropout technology is introduced to improve the generalization ability of the algorithm and prevent overfitting. The network parameters are fine-tuned based on the RMSProp optimization method to speed up the algorithm speed and reduce the probability of falling into local optimum. The stack noise reduction is automatic Encoders can effectively represent complex functions, improve prediction accuracy and generalization ability.

Description

technical field [0001] The invention belongs to the technical field of electric power, and in particular relates to a transient frequency acquisition method based on a stacked noise-reduction autoencoder. Background technique [0002] With the large-scale grid connection of wind power, photovoltaic and other renewable energy units, the proportion of traditional synchronous generators has gradually decreased. However, renewable energy units usually use power electronic converters as grid-connected interfaces, which have almost no moment of inertia, and it is difficult to provide the necessary inertia and active backup support for the grid. Therefore, the moment of inertia of the power system with a high proportion of renewable energy is greatly reduced, the primary frequency regulation capability is weakened, and the risk of frequency instability increases sharply under large-capacity active power disturbance events. [0003] At present, the time-domain simulation method is ...

AI Technical Summary

Problems solved by technology

The offline training speed of this method is fast, but due to the random generation of the connection weight between the input layer and the hidden layer and the threshold of the hidden layer, it is easy to cause the output weight of some neurons to be too small and become invalid neurons, which makes the neurons Decreased network stability

[0005] In summary, the network structure of the above method is usually a shallow network with only one hidden layer, which has limited ability to process typical characteristics of input data, and it is difficult to map the input-output relationship in complex situations, overfitting or underfitting The combination problem is prominent and the generalization ability is poor; and only a single frequency index after the disturbance fault is evaluated, it is difficult to comprehensively judge the transient frequency stability

Images