Liliya A. Demidova -Professor, Department of Computational and Applied Mathematics, Faculty of Computer Engineering, Ryazan State Radio Engineering University.
Address: 59/1, Gagarina street, Ryazan, 390005, Russian Federation.
E-mail: liliya.demidova@rambler.ru

      Models of short-term forecasting of short-time series on the base of strictly binary trees and modified clonal selection algorithm are considered. These enable to increase forecast accuracy by selecting analytical dependences to be formed on the antibodies base, coding strictly binary trees and adequately describing known values of time series,.
      The antibody constitutes a symbolical line, which elements are selected from three preset symbolical alphabets: alphabet of arithmetic operations; alphabet of functionalities and alphabet of terminals. When implementing the modified clonal selection algorithm the use of three symbolical alphabets ensures correct transformation to analytical dependences of antibodies formed in a random way, which structure can be described by means of strictly binary trees.
      When antibodies are coded on the base of strictly binary trees all knots of strictly binary tree are consecutively recorded in a symbolical line, beginning from left to right and from bottom to top. When analytical dependences are formed on the base of antibodies the recursive procedure of antibodies interpretation is applied.
      The modified clonal selection algorithm belongs to a group of evolutionary algorithms, which enable to carry out simultaneous search among several decision alternatives to make the best choice. The main distinctive feature of the modified clonal selection algorithm is use of mechanisms of clonal selection, hypermutation and suppression during alternation of generations of antibodies populations, used to form required analytical dependences.
      A new approach to quality estimation of forecasting models on the base of strictly binary trees and modified clonal selection algorithm has been offered and investigated. The paper has shown the expediency of simultaneous accounting of mean relative forecast error rate and tendencies discrepancy indicator in antibodies affinity calculations for the purpose of forecasting models quality estimation to be defined by involving analytical dependences, formed on the base of strictly binary trees. When applying the modified clonal selection algorithm the considered approach to forecasting models quality estimation enables to exclude from further consideration forecasting models, which are characterized by great values of tendencies discrepancy indicator.
      The offered forecasting models enable to reduce significantly time needed to retrieve an analytical dependence, which gives the best description of shorttime series known values, and can be recommended to address short-term forecasting tasks (for 1-3 steps forward).