On identification of nonobservable contamination inputs

The problem considered is that of estimating time-varying pollution inputs to a environmental system when no direct measurements are possible. Assuming that the rate of input of pollutants can be observed indirectly and that its relation to the observations is known, the problem is a model-inversion problem. However, the relation is generally not unique, i.e. more than one set of input behaviour may be consistent with it and the observations. The paper considers two approaches to obtaining information on the input: the reconstructibility approach and the guaranteed interval approach. The reconstructibility approach leads to reliable but restricted information. The result of the reconstructibility approach is an input performance index (a function). The guaranteed interval approach is an extension of the reconstructibility approach and considers unimprovable lower and upper bounds on an input performance index. The paper presents an algorithm for the guaranteed interval approach.