基于支持向量回归的低轨卫星通信系统信道预测方法

Channel Prediction Method of LEO Satellite Communication System Based on Support Vector Regression

针对低轨卫星通信系统,研究快时变背景下的信道预测问题。通过引入强化学习的训练模式,将支持向量机(Support Vector Machine,SVM)模型改进成支持向量回归(Support Vector Regression,SVR)模型,提出一种循环迭代实现低轨卫星通信系统信道预测的算法。采用Gaussian核函数,通过遗传算法(GA)寻求最佳惩罚系数D和不敏感损失函数μ,最终得到最优超平面,实现多步预测,并实时更新训练集数据提高预测准确度。仿真结果表明,与传统ARMA以及改进后的ARIMA预测模型相比,VR模型收敛速度快、预测误差小、性能表现更好。同时,VR模型对训练数据样本数要求更少,更适用于低轨卫星通信系统背景下快时变信道的信噪比预测。

Aiming at LEO satellite communication system,the problem of channel prediction under the background of fast time-varying is studied.By introducing the training mode of reinforcement learning,the SVM(support vector machine)model is improved into Support Vector Machine(SVR)model,and a cyclic iterative algorithm for channel prediction of LEO satellite communication system is proposed.The Gaussian kernel function is used to find the best penalty coefficient D and insensitive loss function μ through Genetic Algorithm(GA).Finally,the optimal hyperplane is obtained,the multi-step prediction is realized,and the training set data is updated in real time to improve the prediction accuracy.The simulation results show that compared with the traditional ARMA and the improved ARIMA prediction model,the SVR model has faster convergence speed,less prediction error and better performance.At the same time,SVR model requires less training data samples,which is more suitable for signal-to-noise ratio prediction of fast time-varying channel under the background ofLEO satellite communication system.