Research on Optimization Algorithm of Data Flow Forecasting Analysis Based on ARIMA Model

Model parameter estimation is the most important part of system modeling. Improving the parameter estimation accuracy and calculating and computation speed is always the important content of times series applied research. Domestic and foreign scholars have done a lot of work and got certain achievements in applying time series analysis to develop the research of system modeling, however the traditional parameter estimation method with the defects of low precision, poor convergence and parameter estimation white noises coupling is still used for model parameter estimation. This paper has presented the new type optimization algorithm of data flow model, avoiding the traditional algorithm’s defects of pointwise optimization, long time optimization, and linear model of relying on simplification reducing the accuracy. The combination of the advantages of fast convergence near the minimum value and being able to converge for any initial value not only ensure the convergence of iteration planning but improve the convergent speed. In the process of calculation we only need to get the first derivative without getting the inverse matrix. This algorithm has the advantages of fewer iterations and storage unit, fast convergence speed, not requiring to get the inverse matrix and superlinear convergence. It breakthroughs the linear planning method of traditional solving nonlinear unconstrained problem and provides the new thinking of solving unconstrained time series problems. It can be applied in the parameter optimization analysis of network flow and other forecasting model which make the network flow forecasting control and network management obtain the powerful guarantee.