High-Precision Intelligent Identification of Complex Power Quality Disturbances Based on Improved KST and CNNs

With the widespread usage of power electronics and other non-linear loads, the power quality (PQ) issue in the power grid is becoming increasingly prominent, threatening the power system’s stability. As a result, accurate identification and categorization of complicated power quality disturbances (PQD) is a necessity and critical to mitigating grid pollution. In this paper, a new approach for high-precision intelligent recognition of complex PQDs based on improved Kaiser window s-transform (IKST) and convolutional neural network (CNN) (KSTCNN) is proposed. Firstly, KST is applied for PQD time-frequency (TF) signal detection, and then the control function is modified to adjust the shape of the KS-window, and its parameters can be automatically optimized according to the maximum energy concentration to satisfy the various TF signal detection requirements. Then, the CNN architecture is improved using a SimAM attention mechanism that assigns higher weights to the neurons conveying useful PQD feature information and suppresses the surrounding neurons with irrelevant information. Then, an improved hierarchical 2D dense network structure is proposed to achieve the feature extraction and high-precision recognition of PQDs. Then, the results of simulation experiments demonstrate that the classification accuracy of the proposed KSTCNN is better than other compared methods under different noise levels. Finally, in the practical hardware platform, the recognition accuracy of PQDs reaches 97.93, and the real-time detection time is 0.11s, which further verifies the practicality of the KSTCNN and meets the requirement of real-time recognition of complex PQDs.