Study On Modeling And Forecasting In The Mobile Network By Time Series Analysis Technology

With the rapid development of wireless communication industry, the network scale is enlarging. As a result, it's diffcult to balance the limit network resource and growing requirements of wireless channel and hardware processing capacity. One of efficient way to relieve the diffcult is putting limited resources on hot spots through analyzing performance data generated by network. But current methods are diffcult to do so due to most of these methods based on experience of engineers.To solve the diffcult mentioned above, time series modeling technology be used to analyze and forcast network performance.There are 3 types of items in network performance data should be concerned by maintenance technician: items for network resource utilization, hardware loading and traffic, so 3 characteristic be choosed items as our modeling objects, they are PDCH channel utilization factor, PRP loading of PCU and Gb traffic.ARIMA model is a good traffic model capable of capturing the behavior of a network traffic stream. Seasonal ARIMA model is used when there are one or more periods in series, but the modeling and forecast process of ARIMA is very complicated, especially for seasonal model. In this paper, we give a practical method which is easy to programming: First,spectra analysis be done to find out the periodicity of the training data. Based on it, seasonal and non-seasonal difference was conducted on training data, and then the autocorrelation function of the new series be analyzed, find out each nonzero place. All nonzero places are considered to be permutation and combination of each rank of Autoregression and Moving Average, and then polynomial coefficients are estimated by CLS. The most appropriate model will be selected by AIC rule.Then the performance data of every hour of last year be got from live network of Changde MCC, these data is the training data for model construction by using method we give.Future data be predicted by using the model just be constructed, after comparing the predicted result with actual one; it's proved that seasonal ARIMA models could be used to model and to predict actual wireless data traffic.By the way, due to so many cells for a network, it's unavailable to build model for each cell because the quantity of model computing is very large. Thus, in this paper,A solution be given: Firstly, cells are classified by using clustering algorithms, then model be built for center cell of each class, and the model is applied to all cells of this class.Finally, a general ARMA algorithm software was wrote as a plug-in integrated with analysis service of SQL server 2005.