An improved artificial neural network using multi-source data to estimate food temperature during multi-temperature delivery

Product temperature deviation is an important concern in the cold chain management and monitoring of food. Existing “rule-based” monitoring solutions are limited to the direct use of air temperature data of the vehicle used for transport, which can differ significantly from the real temperature of the food being assessed. Thus, this study focuses on developing a new artificial neural network model to precisely estimate the temperature of food products that are stored in multi-temperature refrigerated transport vehicles with minimum sensors. In addition to identifying the temperature in the car, the model also receives input from a multi-source dataset that includes various information such as the outside temperature, initial food temperature, door status, loading and unloading times, etc. The result of the study suggests that the proposed model could substantially enhance estimation accuracy and reliability with fewer temperature sensors in the transport vehicle. It was found that the root mean square error of food temperature estimation based on this model could be decreased by 77% and 79% for chilled and frozen zones, respectively. Moreover, long short-term memory and deep neural networks could avoid overfitting and reduce their estimation errors by about 55% and 48%, when compared to a back propagation neural network. Based on sensitivity analysis, food temperature estimation is significantly influenced by the product's initial temperature and the cumulative time that a door is open. The proposed model could precisely track the real-time food temperature even with sudden ambient changes, thus enabling precautions to take place when required.