Study on Fault Prediction Methods for Short-wave Transmitting System Based on SSA and CKF-LSSVR

On the basis of support vector machine (SVM) regression prediction theory, and in view of the nonlinear and non-stationary characteristics of RF power monitoring data and the limitation of single kernel function in LSSVR prediction algorithm, a fault prediction method for short-wave communication system of least square SVM regression based on singular spectrum analysis and combined kernel function was studied. Meanwhile, singular spectrum analysis was used to filter out the noise components of the power monitoring data of the short-wave transmission system, and the main components and long-term trend information of the signal were extracted to obtain the reconstructed power monitoring data as the training set. A new kernel function was constructed by combining RBF kernel function with polynomial kernel function, and the self-adaptive Fruit Fly Algorithm was adopted to carry out parameter optimization. Then the fault prediction model of the short-wave transmitting system based on SSA and CKF-LSSVR was constructed, and the simulation experiment was carried out to verify, analyze and compare the prediction model. The results indicated that this method could effectively improve the prediction accuracy of the model.