I. Introduction
Correct-by-construction control synthesis has seen recent success in safety-critical applications such as vehicle control [1], [2] and robot navigation [3]. This approach bases the controller on concepts such as reachable set and control invariant sets to synthesize a controller that is capable of enforcing safety. However, reachability analysis and invariant set computation rely on computational tools such as Hamilton Jacobi [4], Linear Matrix Inequality (LMI) [5] and sum of squares (SOS) programming [6], [7] –these methods scale poorly with the dimension of the system. Because of this limitation, sometimes referred to as “the curse of dimensionality,” the applications of the correct-by-construction control synthesis have been limited to systems with low state dimension. There has been effort to break “the curse of dimensionality,” which typically utilizes either the compositional analysis or system symmetry [8]–[11]. For example, in [9], the weakly coupled longitudinal and lateral dynamics of the vehicle are treated independently by finding a bound on the coupling effect. In [10], when a large network system consists of small subsystems that are identical, the symmetry is utilized to compute invariant sets for a large number of subsystems. However, correct-by-construction synthesis for network systems with heterogeneous subsystems and strong coupling between them remains an open problem. One example is the power grid, which consists of various types of generation buses and load buses, as shown in Fig. 1.
Power grid with generator buses and load buses