基于BP神经网络算法的温室番茄CO_2增施策略优化

Optimization of CO_2 Enrichment Strategy Based on BPNN for Tomato Plants in Greenhouse

CO2浓度是植物光合作用的主要原料之一，确定植株生长阶段的最适CO2浓度需求量，对日光温室内CO2浓度调控具有重要意义。以开花期番茄植株为研究对象，将定植后的番茄分为4个CO2浓度梯度处理组，其中，C1、C2、C3处理组CO2增施摩尔比分别为（700±50）、（1000±50）、（1300±50）txmol／mol，CK处理组为温室内自然状态下C0：摩尔比（约450μmol／mol）。实验利用无线传感器网络节点实时监测温室环境因子，包括空气温湿度、光照强度和c0：浓度；利用LI-6400XT型便携式光合速率仪进行光合日动态和环境因子交互影响实验测定。光合日动态组间差异性研究表明，对开花期番茄增施1000—1300Ixmol／mol的CO2时，可使番茄单叶净光合速率提高约37．13％-40，42％。以环境因子为输人参数，建立基于BP神经网络的光合速率预测模型，用于不同CO2浓度梯度下的光合日动态预测。结果表明，模型训练集和测试集的相关系数分别为0．98和0．93，预测精度较高；C1、C2、C3和CK处理组的日动态预测相关系数分别为0．96、0．94、0．78和0．96，与实测结果吻合度较高且相对误差较小，因此该模型可以为可变环境下的番茄光合日变化动态预测提供依据。

Carbon dioxide （CO2） as the important raw materials of plant growth in greenhouse, it is also one of the main factors that affects the plant photosynthesis. Adding CO2 gas fertilizer has been one of the important techniques for increasing production of tomatoes in greenhouse. In order to determine the proper amount of CO2 based on the plant demands, the impact of different CO2 concentrations on net photosynthesis rate （P.） of tomato plants was studied. Tomatoes after planting were treated under four different CO2 concentration levels, including elevated CO2 concentrations of （700 ± 50 ） μmol/mol （ C1 ）, （ 1 000± 50） μmol/mol （ C2）, （ 1 300 ±50） μmol/mol （ C3 ）, and ambient CO2 concentration in greenhouse （about 450 μmol/mol, CK）. The above-mentioned CO2 enrichment was taken in the sunny daytime （09：00-12：00）. During the experiment, firstly, the sensor nodes based on WSN were used to monitor greenhouse environmental parameters, including air temperature, air humidity, light intensity and CO2 concentration. Secondly, the diurnal dynamics of photosynthesis rate of tomato plants were achieved by LI -6400XT portable photosynthesis analyzer at the flowering stage. The parameters were measured by hourly from 08：00 to 18：00. In the environmental factors nested test of photosynthesis, the CO2 concentration gradients were set to 400, 600, 800, 1 000, 1 300 and 1 500 μmol/mol, respectively, the PAR gradients were set to 300, 600, 900 and 1 200 μmol/（ m ＂s）, respectively, and the temperature gradients were set to 28~C and 35℃, respectively. The air humidity came from the ambient environment （23.16% -46.07% ）. Then, in order to better understand the characteristics of tomato growth and achieve the purpose of the regulation of CO2 concentration in greenhouse, BP neural network was used to create photosynthesis prediction model according to the environmental factors nested test of photosynthesis. The diurnal dynamics of photosynthesis rate from different groups were simulated based on established model in the natural environment （except CO2 concentration）, from which the CO2 concentration saturation point was obtained. The results indicated that CO2 enrichment raised Pn of tomato significantly, and the value was 37. 13% and 40.42% higher in C2 and C3 than that in CK, respectively. Furthermore, the photosynthesis prediction model created by training group was accurate with average relative error of 3.91% , mean absolute error of 0. 51 Ixmol/（ m2＂ s） , root mean square error of 0. 79 μmol/（m2-s） and correlation coefficient of 0. 98. The corresponding values of testing group were 10. 08%, 1.36μmol/（m2 -s）, 1.80 i＆mo]./（m2 -s） and 0. 93, respectively. The prediction effect of diurnal dynamics of photosynthesis revealed that the correlation coefficient between the measured and calculated values was 0. 96 when CO2 concentration was set to 700μmol/mol, 0. 94 when CO2 concentration was set to 1 000μmol/mol, 0. 78 when CO2 concentration was set to 1 300 μmol/mol, 0. 96 in the 450 μmol/mol treatment. Therefore, the prediction model had high accuracy and certain universality, which could provide a theoretical basis for optimal regulation of photosynthetic rate dynamically and precise control of CO2 gas fertilizer in greenhouse.