This study focuses on detecting Power Quality Disturbance Events (PQDE) in microgrids integrated with a Solar Energy Conversion System (SECS). The research proposes a novel signal reduction technique called Clark Transformed Modal, which reduces three-phase voltage to a single unit signal, optimizing memory utilization and reducing computational load during feature extraction. A total of 16 features are extracted from the proposed modal signal by performing multi-resolution analysis through Maximum Overlap Discrete Wavelet Transform (MODWT). Various disturbances, including sag, swell, transients, notches, and flicker, are intentionally simulated in a PV-grid tied MATLAB/Simulink model to obtain a dataset of 10800 samples. Further, the dataset is randomly divided into training-testing subsets to verify the classification ability of a novel ensemble classifier called subspace weighted k-nearest Neighbor (SWKNN). In addition to that the optimum mother wavelet (dmay) is identified to even further boost the classifier performance. The results demonstrate the superior classification capabilities of the proposed MODWT-SWKNN classifier in terms of various performance metrics like precision, recall and F1-score. It also outperformed when compared with several competitive PQ classification models based on PV-integrated systems both under ideal and noisy conditions. Additionally, the disturbance detection system is validated in an OPAL-RT real-time environment to demonstrate its efficiency in terms of detection time. The accuracy of detection is found to be 99.74% in ideal case and fall back to no more than 3% regulation i.e., 97.28% even in dense noise of 20dB with as low as 8 WKNN subspaces. Further, average detection time with 500 trails is found to be 0.0285 seconds. The efficacy of the proposed PQ detection algorithm is also tested in a large PV integrated IEEE 13-bus system.