The Research Of Time Series Prediction Method Based On Swarm Intelligence Algorithm

Time series forecasting has the widespread application value and is widely used in social science, natural science, human mind exploring, and other fields. The optimal estimation method of model parameters is the research key of time series prediction. Only a set of parameters is used by classical parameter estimation method.Its estimated results are not evaluated or optimized. So the forecasting precision often cannot meet the requirements. In this paper,swarm intelligence optimization mechanism is analyzed and applicated into univariate and multivariate time series.Least square estimation method has some shortage, such as, the larger calculation quantity, difficult to solve inverse matrix, unable to satisfy dynamic system, and etc. Although recursive least squares estimation is improved, accumulated error leads to divergence of results in the recurrence process. According to the recursive estimation and Swarm optimization Model, swarm intelligence time series forecasting model (STFM) is proposed in this paper. Swarm intelligence optimization algorithms don’t rely on strict mathematical Model and randorn searching mode is used in the algorithms. Because of the strong global optimization searching performance, it is used to solve the particle degradation problems of particle filter and improve the prediction accuracy.According to the swarm intelligence time series forecasting model, the Swarm Optimization algorithms of Memetic and Glowworm are adopted to estimate parameters in univariate time series prediction. The local-searching operator and dynamic moving step is used to improve the Glowworm Swarm Optimization algorithm. At the same time, the GSO algorithm is adopted in particle filter to improve the importance sampling process. Simulation results show that STFM, compared with the traditional algorithm, has better accuracy of prediction.The multivariate time series is complex and different facts have great influence on each other. The multivariate time series prediction is researched in this paper. In order to improving the prediction accuracy, Swarm intelligence time series prediction model is adopted in multivariate time series prediction.Simulation results show that multivariate analysis method improved prediction accuracy compared to univariate analysis method.