A Comparative Analysis of Multi-Step Predictors of Nonlinear Systems

This thesis presents an approach to multi-step prediction for model predictive control (MPC) of nonlinear systems based on artificial neural networks (ANNs). A comparative review of predictor performances on a benchmark problem is first given, to assess advantages and disadvantages of the method we introduce. Subsequently, we verify our approach on the case of blood glucose regulation in people with type 1 diabetes (T1D). By capturing the complex dynamics of glucose-insulin interactions, the ANN improves the predictive accuracy across the MPC horizon, resulting in improved overall glucose control with the MPC. Simulation results demonstrate that this approach outperforms other similar techniques.