利用北京市1960-2004年的月平均气温数据,建立最小二乘支持向量机(LSSVM)与Elman 神经网络模型,分别运用粒子群算法(PSO)与试凑法对这2种模型进行优化,并对2005-2009年的月平均气温进行预测估计,比较2种模型的预测结果,以便找出更准确的气温预测模型.结果表明,2种模型总体上均能较好地拟合气温序列(R2均大于0.985),但是对于低温预测效果均相对欠佳;PSO_LSSVM预测误差(RMSE=1.380 6)明显小于Elman神经网络(RMSE=1.732 5),拟合精度更高,能更好地对短期气温变化进行模拟.因此,可用PSO_LSSVM模型进行气温预测,指导当地的农业生产与工业开发.%Based on the data analysis of monthly average temperature from 1960 to 2004 in Beijing,a least squares support vector machine (LSSVM) and an Elman neural network model were established,and optimized by particle swarm optimization (PSO), trial and error method respectively, and then applied to predict the data of monthly average temperature from 2005 to 2009. Finally a comparative analysis of those two models was made and a more reasonable model was selected. The results showed that two models were all relatively feasible for simulation of the temperature series as a whole (R2 was higher than 0. 985). However,it did not work well when they were used in low temperature prediction. Moreover, compared with Elman neural network model (RMSE=l. 732 5),PSO_LSSVM had less prediction errors(RMSE=1.380 6),and the simulation result was better. Accordingly,PSO_LSSVM model could be applied to temperature prediction, which can serve as a guidance for local agriculture production and industrial development.
展开▼
