System identification and simulation
Walter J. Karplus · 1972
Recent years have seen continuing and increasingly-intensive attempts to extend the art of simulation to areas which heretofore were considered too complex and too difficult to lend themselves to conventional modelling and simulation techniques. These include such environment-oriented fields as air-pollution, water conservation, thermal pollution, etc., as well as systems belonging to the biological, the medical, the economic, and sociological areas. For example, in 1970 the Office of Water Resources Research catalogued over 600 on-going research projects concerned with the modelling of water resource systems. The extension of simulation techniques developed in application areas such as control system design, electro-mechanical systems, etc., to these new areas has often been disappointing, if not completely unsuccessful. This is due to the difficulty in constructing a sufficiently-valid mathematical model---a model which can be used for prediction with a reasonable amount of confidence. It is well-known, of course, that even under the best conditions, inverse problems such as system identification problems, do not have unique solutions. That is, inevitably an infinite number of possible models will satisfy a specified set of excitation/response relationships. Where the identification process is further handicapped by uncertainties as to system structure and inadequate experimental data, the pertinent question is often not: "How good is the model?" but rather: "Is there any point to modelling at all?"