The Research Of Cell Traffic Prediction In GPRS Network Optimization

With more and more users surfing the internet by cell phone, bringing great increasing data, it is more important for network operators to do network optimization. It is a key point in the network optimization now that how to discover the potential problem in the network, provide a decision support to optimization engineers to take measures to prevent it.This paper analyses the methods in the network optimization firstly. For the delay in the time of discovering the problems, a method based on prediction is proposed. According to the performance of GPRS network in the past time, predict it next time in the future, and discover the potential problem ahead. Prediction researches are based on the traffic data collected from a cell of China Mobile Chenzhou branch of Hunan Province.First, time series models are used for prediction. According to cell traffic changing in one day cycle, applies multiple seasonal ARIMA model, proposes the traffic prediction model of cells, and a good prediction result is obtained. For a more precise prediction, a preprocess is taken to the raw traffic data, and some classic time series models are applied for prediction for comparison. At last, some other kinds of cells are collected for prediction modeling, and a similarity of these models is found in the experiments.Next, back propagation neural network is used for prediction in experiments. Raw data preprocessing, the size of training data, as well as the size of input vectors have been studied separately, to find out the best parameter set of BP neural network prediction. Experiments show that taking natural logarithm on the raw data, training data of three, and input vector size of three are the best parameters of GPRS cell traffic prediction.KIII model is the olfactory system of the bionic model. Based on the analysis of the prediction method with KIII model, it is applied for traffic prediction. And the affect to prediction from study time, k value of KNN algorithm and input vector size is analyzed, making a foundation for applying KIII model for time series data.Experiments show that traffic predictions of GPRS cells in short time has precise result, and it provides a basis for decision-making to take measures to keep cell performance well.