Integrated Digital Signal Processing and Machine Learning Frameworks for Real-Time Power Quality Disturbance Recognition and Embedded Implementation

Power quality (PQ) analysis is critical for maintaining stability, reliability, and operational efficiency in modern smart-grid infrastructures, where nonlinear loads, power electronic converters, and renewable energy integration introduce complex, non-stationary disturbances. Accurate PQ disturbance identification demands advanced digital signal processing (DSP) techniques, adaptive decomposition algorithms, and intelligent classifiers capable of analyzing rapidly varying, multi-event grid conditions.

The proposed tutorial offers a rigorous deep dive into advanced DSP foundations essential for high-fidelity power quality event (PQE) recognition in modern smart-grid systems. Using representative disturbances such as voltage sags, swells, harmonics, flicker, impulsive transients, and inter-harmonics, the session examines state-of-the-art non-stationary signal analysis techniques including Hilbert Transform-based analytic signal construction, Empirical Mode Decomposition (EMD), and Variational Mode Decomposition (VMD). These techniques enable precise extraction of instantaneous frequency, Hilbert spectral features, energy signatures, and mode-specific characteristics, which are crucial for real-time tracking of single as well as combined PQ disturbances.

Building on the extracted DSP feature space, the tutorial systematically integrates classical machine learning and cutting-edge deep learning models—Decision Trees, Extreme Learning Machines (ELM), Random Vector Functional Link Networks (RVFLN), deep convolutional neural networks (DCNN), stacked auto-encoders (SAE), and hybrid RDCSAE-RVFLN architectures. Special emphasis is placed on the emerging role of Transformer-based models, which leverage multi-head self-attention mechanisms to capture long-range temporal dependencies inherent in non-stationary grid signals. The tutorial explores how Transformers operate on raw waveform segments or EMD/VMD-derived feature embeddings, achieving enhanced robustness under noisy, dynamic, and multi-event operating conditions.

The final segment transitions from algorithmic design to real-time hardware validation, showcasing implementations on NI-DAQ (USB-6008), dSPACE MicroLabBox II, TI TMS320C6713 DSP, and Xilinx Virtex-5 FPGA platforms. Participants will be guided through the complete workflow of real-time PQE deployment: signal acquisition, fixed-point conversion, deterministic scheduling, FPGA-level Verilog/VHDL mapping of DCNN components, and DSP-based inference execution. The tutorial highlights how optimized DSP kernels, finite state machine (FSM) architectures, and parallel FPGA pipelines enable sub-millisecond PQE detection suitable for industrial grid environments.