Livestock Research, Wageningen University and Research Centre, Lelystad, the Netherlands.
J Dairy Sci. 2011 Sep;94(9):4502-13. doi: 10.3168/jds.2010-4139.
Automation and use of robots are increasingly being used within dairy farming and result in large amounts of real time data. This information provides a base for the new management concept of precision livestock farming. From 2003 to 2006, time series of herd mean daily milk yield were collected on 6 experimental research farms in the Netherlands. These time series were analyzed with an adaptive dynamic model following a Bayesian method to quantify the effect of heat stress. The effect of heat stress was quantified in terms of critical temperature above which heat stress occurred, duration of heat stress periods, and resulting loss in milk yield. In addition, dynamic changes in level and trend were monitored, including the estimation of a weekly pattern. Monitoring comprised detection of potential outliers and other deteriorations. The adaptive dynamic model fitted the data well; the root mean squared error of the forecasts ranged from 0.55 to 0.99 kg of milk/d. The percentages of potential outliers and signals for deteriorations ranged from 5.5 to 9.7%. The Bayesian procedure for time series analysis and monitoring provided a useful tool for process control. Online estimates (based on past and present only) and retrospective estimates (determined afterward from all data) of level and trend in daily milk yield showed an almost yearly cycle that was in agreement with the calving pattern: most cows calved in winter and early spring versus summer and autumn. Estimated weekly patterns in terms of weekday effects could be related to specific management actions, such as change of pasture during grazing. For the effect of heat stress, the mean estimated critical temperature above which heat stress was expected was 17.8±0.56°C. The estimated duration of the heat stress periods was 5.5±1.03 d, and the estimated loss was 31.4±12.2 kg of milk/cow per year. Farm-specific estimates are helpful to identify management factors like grazing, housing and feeding, that affect the impact of heat stress. The effect of heat stress can be decreased by modifying these factors.
自动化和机器人的使用在奶牛养殖中越来越普遍，产生了大量的实时数据。这些信息为精准养殖的新管理概念提供了基础。2003 年至 2006 年，在荷兰的 6 个实验研究农场收集了奶牛平均日产量的时间序列。这些时间序列通过贝叶斯方法下的自适应动态模型进行分析，以量化热应激的影响。热应激的影响通过以下几个方面来量化：发生热应激的临界温度以上、热应激期的持续时间以及由此导致的产奶量损失。此外，还监测了水平和趋势的动态变化，包括每周模式的估计。监测包括检测潜在的异常值和其他恶化情况。自适应动态模型很好地拟合了数据；预测的均方根误差范围为 0.55 至 0.99 千克/天。潜在异常值和恶化信号的百分比范围为 5.5%至 9.7%。时间序列分析和监测的贝叶斯程序为过程控制提供了有用的工具。基于过去和现在的在线估计（仅基于过去和现在）和回溯估计（从所有数据中确定）显示，每日产奶量的水平和趋势几乎呈现出每年一个周期，与产犊模式一致：大多数奶牛在冬季和早春产犊，而不是在夏季和秋季。以工作日效应为基础的估计周模式可以与特定的管理措施相关，例如放牧期间牧场的变化。对于热应激的影响，预计热应激发生的平均临界温度估计值为 17.8±0.56°C。估计的热应激期持续时间为 5.5±1.03 天，估计的损失为每头牛每年 31.4±12.2 千克牛奶。农场特定的估计有助于识别影响热应激影响的管理因素，如放牧、饲养和喂养。通过修改这些因素，可以降低热应激的影响。
