The randomness and volatility of renewable energy pose challenges to the safe and stable operation of power systems.Analyzing the operational modes of power systems under different levels of renewable energy penetration is crucial for their scheduling and operation. Therefore

a comprehensive methodology framework is proposed for extracting typical operational modes of power systems.First

to address the issue of insufficient operational data

a temporal stochastic production simulation is used to quickly generate high-dimensional data. Then

by combining demand response models and extreme scenarios

potential operational risks under high renewable energy penetration are analyzed

and the effectiveness of flexible resources in enhancing system regulation is validated.Subsequently

to solve the high-dimensional and nonlinear issues in temporal data clustering

a deep temporal clustering algorithm is employed.This algorithm integrates a temporal autoencoder

which combines convolutional neural networks(CNN) and bidirectional long short-term memory(Bi-LSTM) networks

with a temporal clustering layer using dynamic time warping as a similarity measure.The clustering centers are derived by jointly optimizing the loss functions of both layers..Finally

based on real and simulated data from a provincial power grid

the clustering results under different penetration levels are analyzed

confirming the effectiveness of the proposed algorithm and demonstrating the role of demand response in regulating extreme scenarios.