The large-scale integration of renewable energy has increased the penetration of power electronics in power systems. The weak support and low immunity of power systems have augmented the probability of cascading failures. The rapid identification of power system disturbances based on the phasor measurement unit can offer support for system security and stability control. Nevertheless

the complexity of the dynamic behavior of power systems has resulted in an increase in the intra-class variance and inter-class similarity of disturbances

making the extraction of data features for different disturbances increasingly arduous. To address the abovementioned issues

a power system disturbance identification method based on optimized multi-dimensional feature extraction is proposed in this paper. This method presents an optimized feature extraction approach based on variational mode decomposition

utilizes an improved sparrow search algorithm to optimize the number of modal components and the penalty factor

and extracts the most informative and stable multi-dimensional time-domain features from PMU phasor data

laying the foundation for enhancing classification accuracy. On this basis

a disturbance classification method based on a deep learning and neural network fusion model is proposed. By taking the classification accuracy as the objective function and optimizing hyper parameters such as the learning rate through the transit search algorithm

the optimal fitting effect of data training is achieved. The proposed method has undergone simulation tests on the IEEE-39 node and the western China power grid model and has been verified with actual disturbance data of the power system. The results indicate that the proposed method exhibits good accuracy and generalization.