Exact solution of the estimation problem for one class of nonlinear systems (Q1921561)

The author studies a class of nonlinear systems driven by the equation \[ \dot x = f(t,x), \quad t \in T = [t_0,\theta] , \] where \(x\) is the (unobservable) state vector of the system and \(f\) belongs to a subclass of the continuous functions. Observations are taken in noise, \(\xi(t)\), according to the equation \[ y = G(t)x + \xi(t) . \] The only information for the initial state \(x(t_0)\) and the error \(\xi(t)\) is that they belong to compact sets \(X^0\) and \(\Theta(t)\), respectively, where \(\Theta(t)\) is further supposed to be convex. The problem is to estimate \(x(\theta)\), the state of the system at time \(\theta\), from observation, \(y^*\), of \(y\) up to time \(\theta\). An information set of states of the system is defined to be the set of those states which are consistent with \(y^*\) and the above equations for some admissible \(x(t_0)\) and \(\xi(t)\), \(t\in T\). The problem then is to find the information set since it contains all the available information about the state of the system and this information cannot be reduced. The author shows that the information set can be obtained as the intersection of the regions of attainability of a set of differential inclusions. An example is elaborated in the last section of the paper.