What question did this study set out to answer?

The research aims to enhance frequency stability in microgrids by improving wind speed forecasting.

May 9, 2026Open Access

Resilient frequency stabilization of renewable-penetrated microgrids via deep learning and machine learning

JHJasmine HansdaOdisha University of Agriculture and Technology PRPrakash K. RayDepartment of Health & Family Welfare AMAsit MohantyEnergy Research Institute

Key Points

The research aims to enhance frequency stability in microgrids by improving wind speed forecasting.
Utilized Support Vector Regression (SVR) and Deep Neural Networks (DNN) for wind speed prediction.
Implemented a PID controller optimized through a hybrid Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) algorithm.
Conducted simulation and real-time testing on the OPAL-RT Hardware-in-the-Loop platform.
SVR model reduced prediction errors by up to 52.94% in MAE and 44.59% in MAPE.
The hybrid PSO-GWO tuned PID controller improved maximum overshoot by 50% and settling time by 16.13% compared to PSO.
For DNN predictions, the hybrid controller achieved 92.5% improvement in maximum overshoot and 57.14% in integral square error.

Abstract

Abstract The intrinsic variability of wind generation necessitates accurate wind speed forecasting to minimize active power imbalance between supply and demand and thereby enhance frequency stability in microgrids (MGs). In this context, this paper investigates wind speed prediction at four geographically distinct sites using Support Vector Regression (SVR) and Deep Neural Networks (DNN) and analyses their impact on microgrid (MG) frequency regulation. The predicted wind data are incorporated as disturbances in a MG system comprising Wind Turbine generator (WTG), Diesel Engine Generator (DEG), Aqua-Electrolyser (AE), Fuel Cell (FC), Battery Energy Storage System (BESS), Flywheel Energy Storage System (FESS), and Ultra Capacitor (UC) units. To mitigate frequency deviations caused by load variations and wind power uncertainty, a PID controller optimized using a hybrid Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) algorithm is proposed. The forecasting analysis indicates that the SVR model outperforms the DNN model, reducing prediction errors by up to 52.94% in MAE, 44.59% in MAPE, 83.54% in MSE, and 59.55% in RMSE. Simulation results further demonstrate that the hybrid PSO–GWO tuned PID controller significantly improves dynamic performance compared with PSO-PID and GWO-PID controllers. For the SVR-based prediction case, the proposed controller reduces maximum overshoot (Mp) by 50%, settling time (Ts) by 16.13%, and integral square error (ISE) by 77.78% compared with PSO, while achieving 33.33%, 10.86%, and 50% improvements over GWO, respectively. Similarly, for the DNN-based case, the hybrid controller achieves further improvements of 92.5% in Mp, 21.21% in Ts, and 57.14% in ISE compared with PSO. The proposed approach is validated through time- and frequency-domain analyses and real-time implementation using the OPAL-RT Hardware-in-the-Loop (HIL) platform, confirming its effectiveness in enhancing the stability and reliability of renewable-integrated MGs.

AI에게 질문

Bookmark

View Full Paper