光合有效辐射预测模型的核函数组合优化

Optimized Photosynthetic Active Radiation Prediction Model Based on Kernel Function Combination

以林下参种植基地光合有效辐射(PAR)、散射辐射(PFDdif)和直射辐射(PFDdir)为研究对象,以支持向量机linear核函数(k1)、polynomial核函数(k2)、radial basis function核函数(k3)为基础,构建新核函数。使用K-fold交叉验证方法,利用粒子群算法(PSO)对惩罚参数c和g值优化。试验结果表明,利用grid search算法设定惩罚参数c为16和g值为1时,通过比较相关系数及符合拟合均衡原则下,选出以0.2k1+0.8k2核函数而构建的光合有效辐射预测模型效果最佳,对由PAR、PFDdir和PFDdif数据组成的预测集1和预测集2拟合程度分别为89.213 2%和81.789 6%。利用粒子群算法对惩罚参数c和g值优化后,预测模型对预测集1拟合程度达到92.156 0%,对预测集2拟合程度达到90.036 0%。可见,采用0.2k1+0.8k2核函数和PSO的支持向量机预测模型对PAR具备预测能力。

Using the photosynthetic active radiation（PAR）,scattering radiation（PFDdif） and direct radiation（PFDdir） from the ginseng base in forest as research object,a support vector machine model about photosynthetic active radiation based on linear function（k1）,polynomial function（k2） and radial basis function（k3） was constituted.By using K-fold cross validation method,penalty parameter c and g numerical value were optimized by particle swarm optimization.Penalty parameter c and g numerical value of photosynthetic active radiation support vector machine model were configured 16 and 1 by grid search algorithm.0.2k1＋0.8k2 kernel function was chosed to construct the predict PAR model by related coefficient and the fitting equilibrium principle.Fitting index of predicting set 1 and 2 was separately 89.213 2% and 81.789 6% based on the predicting model.By using particle swarm algorithm,the two predicting models＇ parameters were optimized.Fitting index of predicting set 1 and 2 was separately 92.156 0% and 90.036 0%.The predicting model based on 0.2k1＋0.8k2 and particle swarm algorithm showed ability to predict PAR variation trend.