Study on LS-SVM method for multi-step forecasting of mobile communication traffic

Aiming at the actual requirements of high precision, high efficiency and multi-step forecasting of mobile communication traffic, this paper proposes a forecasting method based on LS-SVM. Self correlation analysis is adopted to determine the embedding dimension and delay time of the input vectors of LS-SVM, which maximally preserves historic information and reduces sample dimension. Besides, the input vectors are constructed with least forecasted values that substitute for real values, and multi-step forecasting is realized with high precision. The developed forecasting system software was applied in the network management system in Heilongjiang Co. Ltd., China Mobile Communications Corporation (CMCC). Test results with real communication traffic data indicate that the proposed method can realize real-time forecasting of mobile communication traffic with high precision and high efficiency, which is valuable for improving call quality.