基于支持向量机回归的营养液调控模型研究

Regulation Model Research of Nutrient Solution Based on Support Vector Machine Regression

针对目前设施栽培中营养液动态调配精确度低的问题,提出一种基于支持向量机回归(Support vector machine regression,SVR)的营养液调控模型。首先,通过设计嵌套试验采集了13个温度、50组不同Knop营养液(A:99%Ca(NO3)2·4H2O、B:98%KNO3、C:99%KH2PO4、D:98%MgSO4·7H2O、E:99%EDTANaFe 5种化合物)配比下的营养液pH值、EC、K+质量浓度、Ca2+质量浓度和NO-3质量浓度等检测指标值,并基于SVR构建营养液检测指标预测模型;然后,采用离散斜率法计算营养液检测指标值与5种化合物含量的响应曲线离散斜率,并利用人工鱼群算法获取离散斜率最大突变点;最后,以该突变点对应的5种化合物含量作为最优调控目标值,基于SVR构建营养液调控模型,并进行验证试验。结果表明:基于SVR的营养液调控模型中对应5种化合物含量的决定系数分别为0.99、0.98、0.99、0.96、0.99,均方根误差分别为4.29、7.39、5.02、2.85、3.96 mg,拟合效果良好。对比逐步拟合响应模型获取目标值的结果发现,基于SVR的营养液调控模型5种化合物含量的平均相对误差分别降低了37.65%、49.94%、40.53%、50.58%、42.84%;在验证试验中,对比逐步拟合响应模型发现,基于SVR的营养液调控模型5种化合物使用量的相对误差平均值分别降低了46.42%、52.08%、54.03%、53.59%、54.54%,调控过程中5种化合物使用量的平均降低率分别为1.69%、5.81%、5.85%、3.65%、7.08%。本文基于SVR构建的营养液调控模型具有高效、节能特点,可为设施作物栽培的实际生产应用提供参考。

Aiming to struggle with the problem of low precision of nutrient solution dynamic deployment in protected cultivation.Based on support vector machine regression(SVR),a model for regulating nutrient solution was established.Firstly,the pH value,EC,K+concentration,Ca2+concentration and NO-3 concentration of nutrient solution were collected under 13 temperatures and 50 groups of Knop nutrient solution ratio(A:99%Ca(NO3)2·4H2O,B:98%KNO3,C:99%KH2PO4,D:98%MgSO4·7H2O,E:99%EDTANaFe),and SVR was used to construct the index value prediction model.Then,the discrete slope method was used to calculate the discrete slope of the content response curve for nutrient solution detection index value and five compounds,and artificial fish swarm algorithm was used to obtain the maximum mutation point of discrete slope.Finally,the optimal regulation model of nutrient solution was constructed based on SVR with the amount of five compounds corresponding to the largest mutation feature site as the optimal control target value.The determination coefficients of the five compounds in the nutrient solution regulation model were 0.99,0.98,0.99,0.96 and 0.99;the root mean square errors were 4.29 mg,7.39 mg,5.02 mg,2.85 mg and 3.96 mg.These results showed that the fitting effect was good.Compared with the control effect of stepwise regression method to obtain the target value,the average relative errors of the five compounds were reduced by 37.65%,49.94%,40.53%,50.58%and 42.84%.In the validation test,compared with the stepwise regression method,the relative average errors of five compounds in the nutrient solution regulation model was reduced by 46.42%,52.08%,54.03%,53.59%and 54.54%.The average reduction rates of the five compounds were 1.69%,5.81%,5.85%,3.65%and 7.08%,respectively.The nutrient solution regulation model based on SVR had the characteristics of high efficiency and energy saving,which may provide a reference for the practical production and application of protected crop cultivation.