采用水稻无土栽培方法人为控制含氮率,在水稻某特定生长期,同时测量水稻冠层反射率和叶片含氮率,建立了基于冠层反射率的水稻叶片氮含率的回归预测模型.通过分析不同氮环境下各冠层反射率光谱图,确定了与水稻含氮率相关性高的特征波段....采用水稻无土栽培方法人为控制含氮率,在水稻某特定生长期,同时测量水稻冠层反射率和叶片含氮率,建立了基于冠层反射率的水稻叶片氮含率的回归预测模型.通过分析不同氮环境下各冠层反射率光谱图,确定了与水稻含氮率相关性高的特征波段.针对最小二乘支持向量机(leastsquares support vector machines,LS-SVM)参数难定问题,采用遗传算法对LS-SVM参数进行优化.试验结果表明,传统人为选定参数的LS-SVM算法模型的平均回判精确率达到97.21%,预测平均误差率达到5.70%,遗传算法最小二乘支持向量机(genetic algorithm least squares support vec-tor machines,GA-LS-SVM)算法模型的平均回判精确率达到99.60%,预测平均误差率达到2.72%.GA-LS-SVM算法模型的回判及预测效果均明显优于人为选定参数的LS-SVM算法.展开更多
Support vector machine(SVM)has been widely applied to classification and regression problems, but it suf-fers from some important limitations, one of the most significant being that it makes point predictions rather t...Support vector machine(SVM)has been widely applied to classification and regression problems, but it suf-fers from some important limitations, one of the most significant being that it makes point predictions rather thangenerating probability output. A notion of predicting credibility is proposed in support vector regression machinebased on the problem, which can make predicting value have a definite measure, and then relationship between pre-dicting credibility and noise is discussed. Finally, an example of predicting chaotic time series shows the rationality ofthe definition.展开更多
针对电力系统年用电量增长的特点,将最小二乘支持向量机LS-SVM(least squares support vector m a-ch ine)回归模型引入年电力需求预测领域,并给出了相应的过程和算法。与常规基于人工神经网络ANN(ar-tific ial neural networks)的智能...针对电力系统年用电量增长的特点,将最小二乘支持向量机LS-SVM(least squares support vector m a-ch ine)回归模型引入年电力需求预测领域,并给出了相应的过程和算法。与常规基于人工神经网络ANN(ar-tific ial neural networks)的智能预测方法比较,该模型优点是明显的:1)将神经网络迭代学习问题转化为直接求解多元线性方程;2)整个训练过程中有且仅有一个全局极值点,确定了预测的稳定性;3)将年电力需求预测的外插回归问题转换为内插问题,提高了预测精度。应用实例表明:该模型实现容易、预测精度高,更适合年电力需求预测。展开更多
确定性的点预测在精度上无法满足大规模光伏并网的调度需求,基于此,提出一种光伏出力区间预测方法。针对光伏功率原始数据的强波动特性,采用变分模态分解(variational model decomposition,VMD)方法将其分解为若干个子序列,并依据样本...确定性的点预测在精度上无法满足大规模光伏并网的调度需求,基于此,提出一种光伏出力区间预测方法。针对光伏功率原始数据的强波动特性,采用变分模态分解(variational model decomposition,VMD)方法将其分解为若干个子序列,并依据样本熵理论,将复杂度较高的子序列重组为波动分量S,采用高斯过程回归(Gaussian process regression,GPR)法对分量S进行预测,得到其波动区间。考虑到GPR本身固有的缺陷,采用纵横交叉(crisscross optimization,CSO)算法对它的超参数寻优过程进行改进,而复杂度相对较低的其他VMD子序列代表光伏出力稳定分量,因此,采用支持向量机(support vector machine,SVM)法直接对它们进行确定性预测,最后通过重组各分量的预测值,得出光伏出力的区间预测结果。展开更多
文摘采用水稻无土栽培方法人为控制含氮率,在水稻某特定生长期,同时测量水稻冠层反射率和叶片含氮率,建立了基于冠层反射率的水稻叶片氮含率的回归预测模型.通过分析不同氮环境下各冠层反射率光谱图,确定了与水稻含氮率相关性高的特征波段.针对最小二乘支持向量机(leastsquares support vector machines,LS-SVM)参数难定问题,采用遗传算法对LS-SVM参数进行优化.试验结果表明,传统人为选定参数的LS-SVM算法模型的平均回判精确率达到97.21%,预测平均误差率达到5.70%,遗传算法最小二乘支持向量机(genetic algorithm least squares support vec-tor machines,GA-LS-SVM)算法模型的平均回判精确率达到99.60%,预测平均误差率达到2.72%.GA-LS-SVM算法模型的回判及预测效果均明显优于人为选定参数的LS-SVM算法.
文摘Support vector machine(SVM)has been widely applied to classification and regression problems, but it suf-fers from some important limitations, one of the most significant being that it makes point predictions rather thangenerating probability output. A notion of predicting credibility is proposed in support vector regression machinebased on the problem, which can make predicting value have a definite measure, and then relationship between pre-dicting credibility and noise is discussed. Finally, an example of predicting chaotic time series shows the rationality ofthe definition.
文摘针对电力系统年用电量增长的特点,将最小二乘支持向量机LS-SVM(least squares support vector m a-ch ine)回归模型引入年电力需求预测领域,并给出了相应的过程和算法。与常规基于人工神经网络ANN(ar-tific ial neural networks)的智能预测方法比较,该模型优点是明显的:1)将神经网络迭代学习问题转化为直接求解多元线性方程;2)整个训练过程中有且仅有一个全局极值点,确定了预测的稳定性;3)将年电力需求预测的外插回归问题转换为内插问题,提高了预测精度。应用实例表明:该模型实现容易、预测精度高,更适合年电力需求预测。
文摘确定性的点预测在精度上无法满足大规模光伏并网的调度需求,基于此,提出一种光伏出力区间预测方法。针对光伏功率原始数据的强波动特性,采用变分模态分解(variational model decomposition,VMD)方法将其分解为若干个子序列,并依据样本熵理论,将复杂度较高的子序列重组为波动分量S,采用高斯过程回归(Gaussian process regression,GPR)法对分量S进行预测,得到其波动区间。考虑到GPR本身固有的缺陷,采用纵横交叉(crisscross optimization,CSO)算法对它的超参数寻优过程进行改进,而复杂度相对较低的其他VMD子序列代表光伏出力稳定分量,因此,采用支持向量机(support vector machine,SVM)法直接对它们进行确定性预测,最后通过重组各分量的预测值,得出光伏出力的区间预测结果。
