This paper proposes the hybrid method of short-term forecasting of time series with missing values, based on the model of clonal selection which uses heterogeneous antibodies. The method involves time-series segmentation and selection of the most appropriate forecasting method for each section. There is the usage of the case-based reasoning method, where antibodies perform the role of the cases.The antibodies contain samples of known values of the time series, including missing values, and its variant of the forecast for its sample. The creation of the antibodies use one of forecasting method set, depending on the amount of missing values in source sample. Antigen includes the known values of the sample immediately preceding of the predicting values. The problem is to select antibodies having the greatest affinity for the antigen. During the training of the model it is forming of the antibodies that are based on distinctive patterns describing thesetime series. The experimental results illustrate the features of the proposed approach for short-term forecasting of distorted time series. However increase in the number of missing values requires an increase in the training sample and increasing the size of antibodies.