基于回归分析的奶牛热应激-喷淋时长控制模型的构建

Construction of control model for heat stress in cows-spray duration based on regression analysis

为了实现根据环境温湿度智能调控喷淋时长以保证奶牛降温效果的同时控制用水量,试验以西北农林科技大学畜牧教学试验基地的24头泌乳期荷斯坦奶牛为研究对象,于2022年5月13日—8月12日采集奶牛自愿喷淋时长(表征奶牛降温需求的喷淋时长)和牛场环境温湿度数据,共采集92 d,前87 d的数据用于奶牛热应激-喷淋时长控制模型构建,后15 d的数据用于模型验证;2022年5月20日—7月4日,每天早中晚采集3次奶牛呼吸频率(表征奶牛热应激严重程度)数据,共采集46 d;利用采集的牛场环境温湿度数据,通过6种公式计算出环境温湿指数(temperature-humidity index,THI)数据(THI1~THI6);分析THI1~THI6与奶牛平均呼吸频率相关性以筛选适合陕西关中地区的THI;以筛选THI的不同历史时长的平均THI[包括当前喷淋时刻THI(THI_(cur))和当前喷淋时刻前1,2,3,4,5天平均THI(THI_(1d-avg)、THI_(2d-avg)、THI_(3d-avg)、THI_(4d-avg)、THI_(5d-avg))]作为自变量,以喷淋时长的对数值ln(t_(spray))作为因变量分别进行回归分析,以决定系数(R~2,表示模型的拟合度)、残差标准差(residual standard error,RSE)和P值为指标评价回归模型,筛选最优时段THI与喷淋时长的一元线性回归模型;并对筛选的回归模型以均方误差(mean squared error,MSE)、平均相对误差(mean relative error,MRE)、相关系数(r)和P值为指标进行性能验证。结果表明:THI1~THI6与奶牛平均呼吸频率均呈极显著相关(P<0.01),且相关系数接近,在0.88~0.90之间;其中THI2与奶牛平均呼吸频率相关性最高,相关系数为0.90,选取THI1(研究中较常用)和THI2作为奶牛热应激-喷淋时长控制模型的输入参数;不同时段THI1、THI2与ln(t_(spray))的回归模型的R~2、RSE变化趋势相似,即THI1中THI1_(1d-avg)、THI1_(2d-avg)、THI1_(3d-avg)、THI1_(4d-avg)、THI1_(5d-avg)与ln(t_(spray))的回归模型的R~2、RSE均优于THI1_(cur),THI2中THI2_(1d-avg)、THI2_(2d-avg)、THI2_(3d-avg)、THI2_(4d-avg)、THI2_(5d-avg)与ln(t_(spray))的回归模型的R~2、RSE均优于THI2_(cur),其中THI1中的THI1_(1d-avg)和THI2中的THI2_(1d-avg)与ln(t_(spray))的回归模型最佳;THI1_(1d-avg)与ln(t_(spray))的回归模型的预测值与实际值的r为0.76,MSE为0.83,MRE为13.77%;THI2_(1d-avg)与ln(t_(spray))的回归模型的预测值与实际值的r为0.77,MSE为0.82,MRE为13.55%。说明可以采用喷淋前1天的平均THI来预测奶牛平均喷淋时长。

In order to realize the intelligent control of spraying duration according to the environmental temperature and humidity to ensure the cooling effect of cows and control the water consumption,twenty-four lactating Holstein cows in the animal husbandry teaching experimental base of the Northwest A & F University were selected as research objects,and the data on the voluntary spraying duration of the cows(the spraying duration of the cows' demand for cooling) and the environmental temperature and humidity of the farm were collected from May 13 to August 12,2022.The collection time lasted for 92 d.The data of the first 87 d were used to construct the heat stress in cows-spray duration control model,and the date of the last 15 d were used to validate the model.From May 20 to July 4,2022,respiratory frequency(characterizing the severity of heat stress in cows) data were collected three times a day(in the morning,in the middle of the day and in the evening).Frequency(characterizing the severity of heat stress in cows) data were collected.The collection time lasted for 46 d.Based on the collected environmental temperature and humidity data of the cattle farm,the THI data of the environment(THI1 to THI6) were calculated by six temperature-humidity index(THI) formulas.The correlation between THI1 to THI6 and the average respiration frequency of cows was analyzed to screen THIs suitable for the Guanzhong region of Shaanxi.The average THIs of the screened THIs with different historical time lengths(including the THI at the current spraying moment[THI_(cur)],the average THI 1,2,3,4,5 d before the current spraying moment [THI_(1d-avg),THI_(2d-avg),THI_(3d-avg),THI_(4d-avg),THI_(5d-avg)]) were used as independent variable,and the ln(t_(spray)) of spray duration was used as the dependent variable for the regression analyses,respectively.The regression model was evaluated using the coefficient of determination(R~2,which indicates the fit of the model),residual standard error(RSE) and P value as the indicators,and the univariate linear regression model was screened for the optimal time period between THI and spray duration.The performance of the screened regression model was validated by mean squared error(MSE),mean relative error(MRE),correlation coefficient(r) and P value as the indicators.The results showed that THI1-THI6 were significantly correlated with the average respiratory frequency of cows(P<0.01),among them,the correlation coefficient between THI2 and the average respiratory rate of cows was the highest,and the correlation coefficient was 0.90.THI1(commonly used in the study) and THI2 were selected as inputs for the cow heat stress in cows-spray duration control model.The change trends of R~2 and RSE of the regression model with THI1 or THI2 corresponding to ln(t_(spray)) at different time periods were similar,the R~2 and RSE of the regression models of THI1_(1d-avg),THI1_(2d-avg),THI1_(3d-avg),THI1_(4d-avg),and THI1_(5d-avg) on ln(t_(spray)) in THI1 were better than those of THI1_(cur).The R~2 and RSE of the regression models of THI2_(1d-avg),THI2_(2d-avg),THI2_(3d-avg),THI2_(4d-avg),and THI2_(5d-avg) corresponding to ln(t_(spray)) in THI2 were all better than those of THI2_(cur).The regression models of the THI1_(1d-avg) in the THI1 and the THI2_(1d-avg )in the THI2 corresponding to ln(t_(spray)) were the best.The r of predicted value and actual value of the regression model of THI1_(1d-avg) corresponding to ln(t_(spray)) was 0.76;MSE was 0.83,and MRE was 13.77%.The r of predicted value and actual value of the regression model of THI2_(1d-avg) corresponding to ln(t_(spray)) was 0.77;MSE was 0.82,and MRE was 13.55%.This indicated that the average THI of the day before spraying could be used to predict the spraying duration of cow.