基于梯度提升算法的温室黄瓜株高生长模拟

Simulation of greenhouse cucumber plant height growth based on gradient boosting algorithm

为解决温室黄瓜株高生长模拟模型经验因子较多和实用性不强的问题,选取3个玻璃温室进行数据收集,以环境因子和时间因子为输入量,每日株高生长量为输出量,采用XGBoost模型建立黄瓜5个生育期的株高生长量模拟模型,并与LASSO模型进行对比分析,采用Pearson相关分析和XGBoost模型的特征重要性得分确定各生育期影响株高生长的重点因子。结果表明:1)XGBoost模型在黄瓜不同生育期的模拟性能均优于LASSO模型,苗期、伸蔓期、结果前期和结果末期表现出了较好的拟合效果,结果中期的拟合效果一般,苗期的模拟效果最好,决定系数为0.821;2)苗期、伸蔓期和结果期影响株高生长的重点因子分别是当前生育期生长天数、日平均湿度和日平均温度。本研究所建立的温室黄瓜株高生长模拟模型可为温室黄瓜生产环境调控优化提供决策支持。

In order to solve the problems of having many empirical elements and the weak generalization ability of plant height growth simulation model in greenhouse cucumber production,three glass greenhouses were selected to collect the data.Taking the environmental element and time element as input and the daily plant height growth as output,the simulation model of plant height growth in five growth stages of cucumber was established by XGBoost model and compared with LASSO model.Pearson correlation analysis and feature importance score of XGBoost model were used to determine the key elements affecting plant height growth in each growth stage.The results showed that:1)The simulation performance of XGBoost fitting effect in full fruiting stage was general.The simulation effect in the seeding stage was the best,and the determination coefficient was 0.821.2)The key elements affecting plant height growth in seedling stage,tendril elongation stage and fruiting stage were growth days in the current growth stage,daily average humidity and daily average temperature,respectively.In conclusion,the plant height growth simulation model of greenhouse cucumber established in this study can provide decision support for the regulation and optimization of greenhouse cucumber production environment.