2026· Energy Engineering· pp. 1-10· 0 citations· 32 references
Abstract
: Sub-synchronous oscillation (SSO) is a critical stability threat in wind farms connected to power grids through series-compensated transmission lines, where delayed or inaccurate recognition may lead to converter overcurrent, turbine disconnection, shaft torsional vibration, and large-scale power fluctuations. Existing model-based approaches depend heavily on complete system parameters, while conventional signal-processing methods often require long observation windows and are less suitable for rapid online warning. To address these limitations, this paper proposes a physics-aware data-to-image recognition framework and a transfer-learning-based prior VGG model for identifying SSO hazard levels in series-compensated doubly fed induction generator wind farms. First, multi-source power-quality variables, including voltage, current, frequency, active and reactive power, voltage total harmonic distortion, current total harmonic distortion, and rotor-speed-related information, are organized into a two-dimensional feature matrix according to the actual topology of the DFIG grid-connected system. This matrix preserves the electrical relationships defined by the DFIG grid structure and effectively enhances discriminative feature representation by aligning heterogeneous measurements with grid-side and wind-farm-cluster connectivity, enabling more structured feature extraction for subsequent learning models. The resulting matrix is then normalized and mapped into RGB images, allowing neural networks to learn both local physical consistency and cross-cluster spatial coupling patterns. Second, feedforward neural network, capsule neural network, and VGG-style convolutional models are developed and compared for three SSO states: attenuating oscillation, constant-amplitude oscillation, and diverging oscillation. To alleviate limited SSO samples and deep network convergence difficulty, simplified prior SSO images are generated from physically interpretable class prototypes and used to pre-train the VGG feature extractor before transfer to simulated SSO images. The proposed framework achieves strong recognition performance under the random sample-level protocol and shows the best independent case-level performance among the compared methods, providing a useful reference for SSO analysis of wind turbines.
The rise of inverter-based renewable energy in weak-grid systems heightens sub-synchronous resonance (SSR), sub-synchronous control interaction (SSCI), PLL-induced instability, and torsional oscillations, jeopardizing secure system operation. This paper presents a NAS-MOEA-based adaptive damping framework to reduce multi-modal oscillations in a DFIG-integrated modified IEEE 39-bus system under weak-grid conditions. The framework combines NAS-enabled Temporal Convolutional Network (TCN) learning, wide-area PMU feedback, modal stabilization, and many-objective optimization for adaptive oscillation suppression across different operating conditions. Synchronized measurements of rotor speed deviation, PCC voltage oscillations, line current dynamics, PLL angle variation, and DC-link voltage fluctuations are used to create additional damping signals for RSC/GSC and STATCOM control. A small-signal and EMT-based validation framework is developed, accounting for varying short-circuit ratios, renewable penetration levels, wind-speed variations, series compensation levels, and severe transient disturbances. Simulation results show that the NAS-MOEA framework greatly surpasses traditional PI, lead-lag, PSO-based, and DRL-based damping methods. The method boosts the damping ratio from 0.178 to 0.304, a 71.1% increase, and cuts settling time from 8.74 s to 1.86 s, achieving about 78.7% faster stabilization. Significant reductions in oscillation amplitude, ITAE, and control energy occur under severe weak-grid conditions with an SCR as low as 1.5 and series compensation levels reaching 70%. The NAS-TCN architecture boosts oscillation prediction accuracy to 98.2% and cuts inference time by about 31.6%. The results confirm the effectiveness, robustness, and scalability of the NAS-MOEA framework for future inverter-dominated renewable power systems.
A. Katkar· 2026 International Conferenc...· 0 citations
Rotor electrical unbalance (REU) in doubly-fed induction generators (DFIG) used in wind energy conversion systems is a progressive fault whose early detection remains challenging under variable-speed operating conditions. This study presents a comprehensive hybrid framework integrating classical signal processing with advanced deep learning for robust, multi-severity REU detection. The framework operates in three stages. In the first stage, spectral analysis via the Fast Fourier Transform (FFT) extracts fault-indicative sideband amplitudes at (1±2ks)f1; Hilbert-transform envelope analysis captures amplitude modulation patterns; and four statistical indicators (kurtosis, crest factor, skewness, RMS) quantify signal impulsiveness. In the second stage, the Robust Local Mean Decomposition (RLMD) decomposes the stator current into Product Functions (PFs), from which Fault Energy Ratios (FER) are computed as dimensionless, load-invariant descriptors. These stages form a compact eleven-dimensional feature vector. The third stage introduces temporal modelling via LSTM networks and a hybrid CNN-LSTM. Experimental validation on a 30 kW laboratory DFIG test bench with four REU severity levels (Healthy, 150%, 225%, 300%) under 5-fold cross-validation yields 98.0% accuracy and 97.7% macro-F1 for the CNN-LSTM. Inference time is 4.2 ms/sample on CPU.
Mohamed Salah Channouf, L. Saidi, K. Bacha· International Conference on...· 0 citations
One of the impacts of increasing the penetration of wind energy into the power systems of the future is the rise in transient stability and power quality problems, especially in the case of squirrel cage induction generator (SCIG) based wind turbines connected to the grid. Since SCIG systems lack autonomous excitation control and connect directly to the grid, they are very susceptible to voltage dips, short, circuit faults, and harmonic disturbances. In this article, an adaptive deep artificial neural network (ANN), based control framework is suggested to improve transient stability and harmonic tolerance of grid, connected SCIG during harsh disturbance situations. A complete dynamic model of the SCIG is made in the synchronous reference frame, where the electromechanical coupling, reactive power dynamics, and harmonic distortion effects are considered. To help controller training, a multi, objective performance index that limits rotor speed variation, voltage sag severity, and total harmonic distortion (THD) is developed. The deep ANN that is proposed here uses multi, layer nonlinear mapping with adaptive online learning to produce in real time the most suitable reactive compensation commands. The simulation results under severe three, phase faults, harmonic injection, and combined disturbance scenarios reveal the ability to achieve the major enhancements in rotor speed recovery, voltage restoration time, damping performance, and THD reduction compared to conventional PI and shallow ANN controllers. Also, robustness testing with varying parameters shows that the proposed method is effective, flexible, and suitable for wind energy integration into a weak grid.
K. Durga, Syam Prasad, N. Reddy et al.· 2026 International Conferenc...· 0 citations
A deep learning framework based on a Mamba-driven state-space model architecture for comprehensive PQ disturbance classification is proposed and results indicate that the proposed method is well-suited for real-time smart grid monitoring and intelligent protection systems.
Pintu Das, Chandan Jana, Sannistha Banarjee et al.· Engineering Research Express· 0 citations
This paper introduces a shunt active power filter (SAPF) with machine learning that is used to provide power quality (PQ) improvement to a PV-Battery hybrid inverter in a grid-connected system. A predictive compensation strategy based on a long short-term memory (LSTM) is used to produce adaptive reference current to reduce harmonic distortion and supply reactive power. The proposed approach takes advantage of time dependence of disturbance profiles, unlike traditional fixed-gain PI-based controllers typical of active filters, which provide proactive compensation to nonstationary conditions generated by renewable variability and nonlinear loads. An in-depth MATLAB/Simulink analysis integrating the hybrid patterns of irradiance with intermittent dips at cloud fronts was used in testing the PQ performance. The suggested controller resulted in a decrease of the current total harmonic distortion (THD) to 28.6% (without compensation) and 7.8% (PI-controlled SAPF), and an increase in power factor to unity. The momentary analyses indicated lower overshoot and quicker convergence in PV and load disturbances. Computational profiling showed an average inference latency of 38 μs, meeting real-time control requirements on embedded ARM-dsp designs. These findings suggest that predictive control with machine learning support can be used to supplement the traditional SAPF capabilities and improve auxiliary PQ services in renewable-based distribution networks.
Nagaraja Bodravara, G. Ezhilarasan· Advances in Data Science and...· 0 citations
Voltage stability in off-grid systems that use self-excited induction generators (SEIGs) is difficult to maintain. The limited generation capacity and the frequent load variations of rural micro-hydro settings are the main reasons. Accurate prediction of the point-of-common-coupling (PCC) voltage is therefore important for effective control and equipment safety. This research introduces a data-driven predictive model for forecasting the PCC voltage of a digitally controlled SEIG-electronic load controller (ELC) system that experiences sudden load changes. An experimental setup based on an STM32F407VG microcontroller-based ELC was built to gather real-time operational data. Three measured quantities-load power, dump power, and switching events-were used to describe the PCC voltage behaviour. A stacking ensemble (SE) learning model was then employed. It uses ridge regression as the meta-learner and integrates extreme gradient boosting, Gaussian process regression, and support vector regression as base learners. The choice of these learners is justified both qualitatively, from the structure of the SEIG steady-state model, and quantitatively, through a paired statistical comparison. The model's performance was compared with the individual learners using standard statistical metrics, and the differences were assessed for statistical significance using paired Wilcoxon signed-rank and Diebold-Mariano tests. A convex combination-based data augmentation method, validated through kernel density estimation, was also examined but reduced predictive accuracy, which confirms the sufficiency of the original dataset. The proposed ensemble achieved superior performance ([Formula: see text]) with minimal prediction error. The associated prediction uncertainty was quantified through the probabilistic Gaussian-process member to provide confidence bounds on the estimated voltage. These results show the potential of ensemble learning to improve voltage prediction and to support advanced control in SEIG-ELC based off-grid micro-hydro systems.