These investigations are dedicated to the problems of accuracy improvement 
of short term and middle term temporal series prognosis. The new method 
of data lost location and data recovery is developed in the framework of this research. 
The point neighborhoods and K nearest neighborhoods methods are used for training 
“functional on the tabular functions set” ANN. It allows us to improve accuracy
of the ANN prediction. The Information Analysis System IAS “Prognoz” 
is developed basing on the research results.