改进随机森林的农作物产量短期最优预测仿真

Simulation of Short-Term Optimal Prediction of Crop Yield Based on Improved Random Forest Algorithm

传统方法在农作物产量长期预测中的预测效果较好,但是短期预测方面还有待加强,为此以随机森林算法为基础,提出一种产量短期预测方法。针对农作物产量及其相关影响因素的非平衡数据集特点,结合合成少数类过采样与相同类别样本重心相同原则,得到非平衡数据抽样方法。将抽样处理后的数据集作为随机森林算法训练数据集合,通过寻得算法中决策树个数与单决策树随机选取特征数这两个关键参数的最优解,基于研究地区的农作物产量相关数据,凭借设置的解释变量与预测变量,完成短期预测模型的构建。仿真环节,采用多个评估指标从多方面检验所提方法的预测性能,由实验结果可知,上述方法单次与多次预测的偏差与变化幅度均较小,具有明显的精准预测优势,且相对稳定。

Traditionally, some methods show satisfactory forecasting ability in long-term crop yield forecast, rather than in short-term forecast. Based on random forest algorithm, a method to forecast short-term yields was proposed. Firstly, according to the characteristics of non-equilibrium data set of crop yield and its related factors, combined with the principle that the center of gravity of over sampling of a few categories is the same as that of samples of the same category, the non-equilibrium data sampling method was obtained. Secondly, we took the sampled data set as a training data set of random forest algorithm. After finding the optimal solution of two key parameters, namely the number of decision trees and the number of features by randomly selected a single decision tree in the algorithm, we used explanatory variables and prediction variables to construct a short-term forecasting model based on the data of crop yield in the study area. In the simulation, multiple evaluation indexes were used to test the prediction performance of the proposed method from multiple perspectives. Experimental results show that the deviation and change range of the single and multiple predictions of the method are both small. In addition, this method has obvious advantages in accurate forecast and stability.