38 / 2025-03-28 15:33:58

Dominant Instability Mode Identification of Power System under Multi-Feature Fusion

feature fusion, dominant instability mode, deep residual network, bidirectional gated recirculating unit, cross-attention mechanisms

In order to solve the problem of system failure and large-scale power outage caused by transient instability of power system, this paper proposes a dominant instability mode (DIM) identification model of power system based on Multi-feature Fusion (MFF). Firstly, the Laida criterion was used to preprocess the raw data to effectively identify and eliminate noise interference. Secondly, the residual bottleneck structure is introduced into the Deep Residual Memory Network (DRMN) to greatly reduce the computational complexity, and the bidirectional dependencies in the power system time series data are fully captured by combining bidirectional learning. In addition, the Tanh-Sigmoid hybrid activation function is introduced, and its nonlinear characteristics significantly enhance the expressive ability of the model. Finally, the cross-attention mechanism is used to dynamically allocate the weights of amplitude and phase angle features, which realizes the efficient fusion of multi-source features and the accurate capture of interactive information. Experimental results demonstrate that the proposed model achieves an accuracy of 99.18% in conventional tests and 95.50% under 5 dB noise conditions.. Compared with other models, the proposed model shows stronger effectiveness and anti-noise ability in the recognition of DIM in power system.