Optimal neurocontrollers for discretized distributed parameter systems

We propose to use the framework of backpropagation through time (BPTT) [1], [2], [3] to create optimal feedback neurocontrollers for distributed parameter systems (DPS). DPS are systems distributed in space while evolving in time. Unlike the lumped parameter systems, DPS are represented by a set of partial differential equations in the state space. Our neurocontrollers obtained for discretized DPS in the infinite-horizon regulator setting are applicable to a broad set of initial states (an envelope of initial state profiles). We compare our technique and results with another approach to synthesizing optimal DPS neurocontrollers introduced in [4], [5].