The Prediction Of Telephone Traffic Time Series Based On Chaotic Theory

The telephone traffic is a kind of nonlinear time series affected by various factors; it can provide evidence for mobile operators on network construction, optimization, investment, and marketing strategy by its prediction. So, the telephone traffic prediction is an important basic work for mobile operators.This paper mainly studys on the prediction of telephone traffic of daily rush hour, researches on prediction methods at home and abroad; using phase space reconstruction technology of chaotic theory, using autocorrelation function algorithm and improved false adjoint point algorithm, get the optimal time delay and embedding dimention parameters of the phase space reconstruction, reconstruct the phase space of telephone traffic time series, extract and display the original law of the system from the telephone traffic time series, that is recover the chaotic attractors from high dimensional phase space, and then use trajectory calculation algorithm of maximum Lyapunov exponent during the precursor monitoring time series to calculate the maximum Lyapunov exponent and maximum time-scale forecasting, the research shows that the telephone traffic time series is of chaotic characteristic and predictable.As the time series of the telephone traffic is complicated non-linear chaotic, this paper uses the prediction method of chotic time series, uses adding-weight one-rank local-region method, establishes the prediction model, and predicts the telephone traffic time series, the prediction result is not satisfied. In order to increase the prediction accuracy, in view of that the neural network has powerful non-linear mapping capability, it is an effective tool in non-linear system approximation and modeling, it is widely used in prediction area. So this paper uses it to study the chaotic time series, and then predict. At first, using the static neural network, uses improved BP algorithm to establish the prediction model, and does the prediction research on chaotic time series. The prediction effect is better than that of the adding-weight one-rank local-region method, but the static BP neural network does not has the characteristic of non-linear, it is hard to describe the complicated law that non-linear system has. So this paper uses the dynamic neural network:Elman network to improve the prediction accuracy, and to predict the telephone traffic time series. As the dynamic neural network:Elman network has the inner feedback connection, and this connection make it has sensitivity to historical data. It increases the ability of the net to deal with the dynamic signal. The simulation result shows that the method increases the prediction accuracy.In a word, in describing dynamic characteristic of non-linear system, the dynamic neural network is better than the static neural network, the research of this paper makes a beneficial try to prediction research of the telephone traffic of mobile communication network.