Application of neural networks to fluorescent diagnostics of organic pollution in natural waters

A neural net has been used to build a sea water express-diagnostics system that does not require an expert´s participation. The network can classify sea water pollution on the basis of its total luminescent spectrum. A typical minimal detectable PSP (pollutant´s spectrum portion) value is 0.2; this value corresponds to a concentration of 0.5-10 p.p.m. for dissolved organic matter (DOM) in water (10 mg/l of organic carbon), a situation that is most difficult for diagnostics. The net is capable of giving adequate answers for spectra corresponding to a mixture of two pollutants or for spectra of unknown substances. It is possible to detect nonstandard situations, as reflected by the simultaneous activity of a number of output neurons. A linear neural net allows one to define the concentration of a known pollutant with a typical accuracy of 0.1 p.p.m. Variations of the DOM spectrum shape and amplitude within reasonable limits do not affect the neural net´s functioning