Ivashchenko H. Hybrid models of short-term time series forecasting based on artificial immune systems

Dissertation is devoted to development of hybrid models of short-term time series prediction based on a combination of the principles of artificial immune systems and existing forecasting methods that can improve the forecast accuracy and resistance to distortion of the original data. The hybrid model of time series forecasting is proposed. This model performs fragmentation of the original time series and selection for each fragment of his prediction method, which reduces of learning sample and increases the accuracy of the forecast. For the time series segmentation the model uses a combination of the case based reasoning and the model of clonal selection. Considered using of multiantibodies as precedents in the artificial immune network model. The resulting model can change the parameters of the work and structure of the immune network in the learning process. The hybrid model based on the model of clonal selection is proposed. This model is used to predict of the distorted time series, takes into account the impact of external factors and allows increasing the accuracy of the forecast, and provides a possibility of forecast in real time. To ensure data processing in real time, these approaches use parallel systems based on common and individual memory. Criterion is proposed for determining the applicability of the immune approach of short-term time series forecasting based on the model of clonal selection and a CBR, which allows configuring parameters of hybrid model, that reduces the time for training of AIS and reduces the size of the antibody population. It was developed instrumental environment for the analysis of the immune algorithm and solving the problem of short-term time series forecasting, which allows to perform a comparative analysis for evaluating effectiveness of the proposed approaches.