Many goods, including pharmaceuticals, require close temperature monitoring. This is important not only for complying with regulations but also for guaranteeing safety of use. A particular challenge in controlling a product's temperature arises during transportation. In cold supply chains (SCs), temperature is maintained by refrigerated containers. However, many situations, e.g. cooling system failure, lead to ambient temperature changes, and this needs to be detected as early as possible to prevent product damage. Existing approaches to temperature prediction are confined to long-term forecasts with relatively stable ambient temperatures and/or rely on multiple sensors in the known fixed positions. Since interventions in a SC are required immediately, there is a need for methods that provide real-time predictions regarding regular ambient temperature instability, i.e. when the ambient temperature changes unexpectedly in the short term. We propose a novel method that extends the applicability of Newton's law of cooling (NLC) to changeable ambient temperatures based on a set of temperature stability conditions and a sensor measurement error. In the method, an optimal number of measurements that characterize stable ambient temperatures and improve prediction reliability are selected. We compare the adapted NLC with artificial neural networks and autoregressive moving average models with respect to deviation prediction, prediction error, and execution time. Our evaluation based on real-world data shows that the adapted NLC outperforms existing baseline methods. In contrast to existing solutions, our method does not require any knowledge about the positioning of products within the container, further increasing its practical value.

许多商品，包括药品，都需要严密的温度监控。这不仅对于遵守法规很重要，而且对于保证使用的安全性也很重要。在运输过程中出现了控制产品温度的特殊挑战。在冷供应链（SC）中，温度通过冷藏容器保持。但是，许多情况（例如冷却系统故障）会导致环境温度变化，因此需要尽早检测到这种情况，以防止产品损坏。现有的温度预测方法仅限于具有相对稳定的环境温度和/或依赖于已知固定位置的多个传感器的长期预测。由于必须立即采取干预措施，因此，需要一种提供关于规则的环境温度不稳定性的实时预测的方法，即当环境温度在短期内意外变化时。我们提出了一种新颖的方法，该方法根据一组温度稳定条件和传感器测量误差将牛顿冷却定律（NLC）的适用性扩展到可变的环境温度。在该方法中，选择了表征稳定的环境温度并提高预测可靠性的最佳测量次数。我们将改进的NLC与人工神经网络和自回归移动平均模型进行比较，以进行偏差预测，预测误差和执行时间。我们基于实际数据的评估表明，改编的NLC优于现有的基线方法。