A New Method of Optical Power Trend Forecasting Using ARIMA-SVM Combination Model
Wang Li · Electric Power Information and Communication Technology · 2015
Aiming at the problem of optical fi ber line condition trend forecasting in electric power communication system, this paper proposed an optical power trend forecasting method using ARIMA-SVM combination forecasting model. Firstly, according to the nonlinear and time-varying characteristics of optical power data, the wavelet transform was used to decompose and reconstruct the optical power data. Then the SVM forecasting model using hybrid kernel function was designed, and the ARIMA model and SVM model using hybrid kernel function were constructed with the reconstructed data and utilized to forecast optical fiber line condition respectively. Finally, the forecast results of above models were combined to complete the optical power forecasting. This paper conducted forecasting experiment using ARIMA, RBF, SVM, ARIMA-RBF and the proposed model respectively, and compared the forecasting effect. The results show that the ARIMA-SVM combination model based on hybrid kernel function has the most accurate forecasting effect in optical power trend with a signifi cant improvement of the forecasting precision and calculation speed. Thus a valid and feasible forecasting method is provided for future trend of optical power.