<p>Power quality (PQ) disturbances in photovoltaic (PV) combined direct current (DC) microgrids are typically caused by fluctuations in solar irradiance, load changes, and switching events, leading to voltage sags, swells, and harmonic distortions. These disturbances negatively impact the stability and efficiency of the entire microgrid system. However, traditional methods for detecting and classifying these disturbances often struggle with high variability and noise in the data, making accurate identification challenging. Deep learning (DL) provides a robust solution, leveraging advanced algorithms to efficiently analyse complex patterns and improve the accuracy of disturbance detection and classification in power systems. In this manuscript, the Hamiltonian Deep Neural Network Technique Optimized With Lyrebird Optimization Algorithm for Detecting and Classifying Power Quality Disturbances in PV Combined DC Microgrids System (HDNN-LOA-DCPQD-PVMG). Initially, the input signal is gathered from the Power Quality Classification Dataset—2. The Vold Kalman Filter (VKF) is used to filter the actual signal from the noise signals. The pre-processed signals are fed to the feature extraction phase using General Synchro extracting Chirplet Transform (GSCT). Then, extracted features are fed to Hamiltonian Deep Neural Networks (HDNN) for effectively detecting and classifying the Power quality (PQ) disturbances as normal, voltage sag, voltage swell, transients and harmonics. Typically, HDNN does not provide ways for adapting optimization to find the ideal parameters for precise PQ disturbance detection and classification. Hence, the Lyrebird Optimization Algorithm (LOA) is proposed to optimize Hamiltonian Deep Neural Networks (HDNN) which accurately detect and classify PQ disturbances. The proposed technique is implemented in MATLAB, and the efficiency of the process is determined using a variety of performance metrics, such as F-measure, accuracy, recall, precision, and Kappa Statistics (KSs), mean absolute error (MAE), receiver operating characteristics (ROC), and root mean square error (RMSE). The proposed HDNN-LOA-DCPQD-PVMG method achieves 32.03%, 24.14% and 21.21% higher accuracy, 21.03%, 24.34% and 26.8% higher precision and 23.36%, 28.05% and 20.41% lower MAE when analyzed with existing methods such as a new deep learning method for the classification of power quality disturbances in hybrid power system (BOA-CNN-CPQD) and power quality detection and classification algorithm based on FDST and hyper-parameter tuned light-GBM using memetic firefly algorithm (LGBM-PQDC) and Classification of Power Quality Disturbances in Solar PV Integrated Power System Based on a Hybrid Deep Learning Approach (SVM-CPQD-PVPS)methods respectively.</p>