| Abstract [eng] |
Time series analysis purpose is to establish a model that describes the dynamics of the time series and then use this model for time series forecasting. Although there are a lot of effective operating time series forecasting methods, but not the best forecasting method. All methods have their own advantages and disadvantages. One of the most commonly used time-series forecasting methods are autoregressive integrated moving average model (ARIMA(p,d,q)). This method will be used in our time series to compare with our proposed method. The aim of this work is to create the time series dynamic model, which is based on non-uniform embedding in the phase-space. False nearest neighbor method is used for determine the embedding dimension. Ant colony optimization algorithm is used for non-uniform time delay search. Ants’ pheromone is used to calculate the distribution of the roulette wheel selection algorithm. Optimization objective function is the average area of all attractors. The model is applied for forecasting with the best the time delay values. Finally, radial basis function neural network is used to forecast values. In the beginning half of the time series is used for training the neural network, and the other half - prognosis and the prediction error evaluation. Results of forecast are compared with other time series methods using the prediction error metrics. The paper examined the hypothesis that our proposed model is better compared with other methods, and the other authors of the studies. The study is presented in Mackey-Glass chaotic time series prediction, and other real-world time series. Mackey-Glass time series were predicted with small tolerances. Most other models predicted worse than our proposed model. However, there are methods with much smaller prediction tolerances. Real-world time series prediction error was compared with the ARIMA model. Our proposed model everywhere predicted more accurately than ARIMA model. Studies have shown that our proposed model is able to predict far into the future with small tolerances. |
