光温因子驱动的园艺作物叶龄模型模拟精度比较

Comparison of simulation accuracy of leaf age models for horticultural crops driven by light and temperature factors

为了提高光温因子驱动的园艺作物通用性叶龄模拟模型的模拟精度,以黄瓜、芹菜、菠菜、小香芹、郁金香和茶叶为供试材料,进行了7年(2016—2022年)的分期播种试验,依据作物生长发育与关键气象因子(辐射和温度)的关系,采用4类建模方法(温差法、积温法、生理发育时间法和辐热积法)构建了园艺作物叶龄模拟模型,并以6种方式(平均值、最值均值、中值、逐步回归、BP神经网络和Elman神经网络)和2种集成逻辑(直接和分步)集成模拟结果,最终优化模型模拟精度。结果表明:1)2种集成逻辑下模型模拟精度均较高,且分步集成逻辑优于直接集成逻辑,平均绝对误差(mean absolute error,MAE)差值为0.31 d,平均相对误差(mean relative error,MRE)差值为0.33%,均方根误差(root mean square error,RMSE)差值为0.40 d,归一化均方根误差(normalized root mean square error,NRMSE)差值为0.46%;2)2种集成逻辑下模型最优时间尺度为逐时尺度,最优作物类型为茶叶,最优建模方法为Elman神经网络集成模拟模型。研究结果可为园艺作物智慧生产管理和可视化提供理论依据和技术支撑。

The purpose of this study was to improve the simulation accuracy of a universal leaf age model driven by light-temperature factors for horticultural crops.In order to achieve this,cucumber,celery,spinach,coriander,tulip,and tea were selected as experimental materials and 7 years(2016-2022)staged sowing experiment was conducted.Based on the relationship between crop growth and key weather factors(radiation and temperature),4 modeling methods(accumulated temperature difference method,accumulated temperature method,physiological development time method,and accumulated product of thermal effectiveness and photosynthetically active radiation method),6 approaches(mean value,mean of extreme values,median,stepwise regression,BP neural network,and Elman neural network)and 2 integration logics(direct and stepwise)were employed to integrate the simulation results,aiming to optimize the accuracy of the model used to construct the leaf age simulation model for horticultural crops.Results showed that:1)The models under both integration logics exhibited high simulation accuracy,with the stepwise integration logic performing better than the direct integration logic.The differences in mean absolute error(MAE),mean relative error(MRE),root mean square error(RMSE),and normalized root mean square error(NRMSE)were 0.31 d,0.33%,0.40 d,and 0.46%respectively.2)The optimal time scale for the models under both integration logics was hourly,while tea was the optimal crop type,and the Elman neural network integration simulation model was the optimal modeling method.The findings of this study can provide theoretical basis and technical support for intelligent production management and visualization of horticultural crops.