论文部分内容阅读
以黑龙江农田黑土为研究对象,利用遗传算法(GA)波长选择结合偏最小二乘法(PLS)回归建立土壤有机碳(SOC)的预测模型。通过设定以下GA参数:波长选择数量上限k、初始种群大小P及迭代次数N,采用单点优化方式逐一确定各参数。结果表明,在主成份数为7的情况下,当GA的参数取N=300、P=300、k=50时,GA模型最优;模型的校正决定系数R2=0.922、校正均方根误差RMSEC=1.74、交叉检验均方根误差RMSECV=1.80;模型的预测决定系数R2=0.931、预测均方根误差RMSEP=1.84、预测相对误差RPD=3.81。与原始光谱的PLS模型相比,R2由0.900提升至0.922,RPD由3.38提升至3.81。结果表明,通过GA进行波长选择能够优化模型,提升模型稳定性以及预测精确性。
Taking farmland black soil in Heilongjiang as an example, a prediction model of soil organic carbon (SOC) was established by genetic algorithm (GA) wavelength selection combined with partial least squares (PLS) regression. By setting the following GA parameters: the upper limit of the number of wavelength selection k, the initial population size P and the number of iterations N, the parameters are determined one by one using the single-point optimization. The results show that the GA model is the best when the parameters of GA are N = 300, P = 300 and k = 50 under the condition that the number of main components is 7. The correction coefficient of model R2 = 0.922, the root mean square error RMSEC = 1.74, root mean square error of cross-validation RMSECV = 1.80; model predictive coefficient R2 = 0.931, root mean square error of prediction RMSEP = 1.84 and relative error RPD = 3.81. Compared with the PLS model of the original spectrum, R2 increased from 0.900 to 0.922, and the RPD increased from 3.38 to 3.81. The results show that the wavelength selection by GA can optimize the model, improve the model stability and prediction accuracy.