基于双重预测模型的非线性时间序列预测

摘要

在支持向量机( SVM)预测问题中,为了减小错误参数选取对预测结果的影响,提出了1种基于双重预测模型的非线性时间序列预测算法.该算法在充分考虑支持向量机参数对推广能力影响的基础上,分别利用自回归预测模型(AR)、自回归滑动平均模型( ARMA)、线性回归和决策树模型对SVM参数进行预测,将预测参数运用到SVM预测模型中.最后,利用该算法对中国火灾发生起数进行了预测,实验表明,该方法在预测准确率上明显优于基于遗传算法选参和粒子群算法选参的SVM预测方法.%A new nonlinear time series prediction algorithm based on dual-forecasting model is proposed for support vector machine (SVM) to minimize the influence of error parameters. Fully utilizing the effect of the SVM parameter on generalization ability, we use Autoregressive (AR), Autoregressive Moving Average ( ARM A), linear regression, and decision tree model to predict SVM parameter, followed by applying the predicted parameter to the SVM model. At last, fire occurred in China is forecasted by the algorithm, and numerical simulation demonstrates that the proposed method surpasses the Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) based SVM in terms of prediction accuracy.