Data‐driven symbolic model predictive control of unknown nonlinear system via abstract model

Abstract Inspired by the synthesizing Abstraction‐Based Controller Design for reach‐avoid problems, this work proposes a data‐driven MPC approach for complex nonlinear systems, whose dynamics are unknown, and there is no prior knowledge. Firstly, with finite collected input state data from this unknown system, a learning model is trained to capture the system transition relation. Then, an abstract model is constructed based on this learning model. By finitely partitioning the state space and input space, and computing the reachability by the learning model, a reachable table is constructed. This table formalizes the abstract model of the original system. By now, the dynamics of the system are captured and interpreted. Finally, a predictive control approach is presented via the abstract model. This controller design can be seen as a reach‐avoid path planning problem on the system state space. The feasibility of the proposed approach is shown in a real smart water distribution system.