基于时频图和时序特征组合的电能质量复合扰动识别OA

To address the challenge of identifying power quality disturbances(PQDs),this paper proposes a lightweight two-branch multimodal fusion recognition model,LIRC-BiLSTM.The model first applies an S transform to the raw PQD signals to produce time-frequency images that are fed to a convolutional block attention module(CBAM)branch,while the raw one-dimensional PQD time series vectors are sent to a bidirectional long short-term memory network(BiLSTM)branch.In the CBAM branch,a multi-scale feature-extraction module captures image features at different resolutions,and a CBAM is introduced to adaptively enhance channel and spatial attention,focusing on key patterns and overall trends in the time-frequency images.In the BiLSTM branch,the time-series matrix undergoes lightweight convolutional preprocessing before being input to a BiLSTM,and a self-attention mechanism is applied to strengthen the temporal features.Finally,the outputs of both branches are fused to combine time-frequency and temporal features for PQDs type classification.Simulation results show that the proposed LIRC-BiLSTM model effectively integrates time-frequency images with temporal detail,significantly improving classification accuracy and noise robustness for multiple classes of power quality disturbances.