Unleashing the Power of TQWT and Adaptive CNNs for Accurate PQ Disturbance Detection

Power quality (PQ) disturbances significantly impact the reliability of modern electrical systems, necessitating robust and precise detection methods. This study introduces a novel framework that integrates the Tunable-Q Wavelet Transform (TQWT) and an Adaptive Convolutional Neural Network (Adaptive CNN) for detecting and classifying PQ disturbances with high accuracy. TQWT is employed for multi-resolution signal decomposition, extracting rich features across varying frequency ranges to capture complex disturbances such as interharmonics, transients, sags, and swells. The decomposition parameters, including the Q-factor, redundancy, and number of levels, are dynamically optimized using a Signal-Adaptive Parameterization Algorithm (SAPA). This optimization ensures that the decomposition aligns with the unique characteristics of each PQ disturbance, maximizing feature sparsity and noise resilience. The extracted features are processed through an Adaptive CNN, equipped with a Dynamic Kernel Pooling (DKP) mechanism that adjusts kernel shapes and sizes based on the spectral characteristics of input features. Additionally, a hybrid loss function integrates classification accuracy, feature sparsity, and noise suppression, ensuring robust model training. Comprehensive evaluations on synthetic and real-world datasets demonstrate the framework's superior performance, achieving high classification accuracy and robustness under noisy conditions.