Two-parameter Kautz network-based LTV-MPC for non-linear standalone micro-grid control

This study proposes a novel linear time-variant model predictive controller (LTV-MPC) for the centralised control of non-linear standalone micro-grids. At each sample, within the prediction horizon, LTV-MPC linearises the non-linear micro-grid model around the state and input reference trajectories resulting in a linear time-variant (LTV) model. The LTV model is used for predicting the forced response of the micro-grid. The natural response is predicted by solving the non-linear model along the state and input reference trajectories. An optimal control problem for the LTV-MPC is formulated using the complete predicted response, which is a quadratic programming problem instead of a non-convex non-linear programming problem. The quadratic programming problem is solved online at each sample to generate the optimal control trajectories within the control horizon. The study recommends the use of two-parameter orthonormal Kautz networks in the LTV-MPC design for the control trajectories approximation. The approximation drastically reduces the number of optimising variables in the optimal control problem without compromising LTV-MPC performance. A standalone eight bus micro-grid with one synchronous distributed generator (DG) and one photovoltaic-DG is considered for the analysis. The LTV-MPC performance is assessed for the different load disturbance and source intermittency scenarios. The results are compared with the existing MPC designs.