Support Vector Regression Model Assisted Robust Frequency Control for Marine Microgrids Under Stochastic Conditions With Real-Time Validation

Highly variable loads, integration of renewable energy sources (RESs) and stochastic environmental conditions encourage precise and effective frequency regulation in marine microgrid (M$\boldsymbol{\mu}$G). Considering this as motivation, the proposed work suggests predictive and hybrid control framework by integrating traditional rule-based tuning with supervised learning-driven optimization through support vector regression (SVR). For simplifying the analysis, M$\boldsymbol{\mu}$G model is approximated to its first-order-plus-dead-time (FOPDT) model. FOPDT model provides values of gain, delay time and time constant, which are further utilized in control strategy. The control strategy utilizes proportional–integral–derivative (PID) controller and performs its initial tuning using rule-based methods, i.e., Ziegler–Nichols (ZN) and Cohen Coon (CC) methods to obtain baseline controller parameters. In the later stage, these parameters are further finely tuned using SVR-based framework, i.e., SVR-ZN and SVR-CC, for obtaining more adaptive and precise tuning. For accurate mapping of nonlinear relationship between input features and target, radial basis function (RBF)-based kernel trick is used. To further elaborate on overall performance and control accuracy of M$\boldsymbol{\mu}$G, performance analyses based on time domain parameters and error indices are performed. Stability analyses are performed by plotting Bode and root locus plots. To evaluate the robustness, disturbances are injected in RESs generation and load, and the outcomes are analyzed. For sensitivity analysis, frequency deviation plots are obtained for different values of inertia constant and damping coefficient. All above-mentioned analyses are validated in real time on OPAL-RT platform.