Temperature Prediction of a Temperature-Controlled Container with Cold Energy Storage System Based on Long Short-Term Memory Neural Network

Abstract

Temperature prediction is important for controlling the environment in the preservation of fresh products. The phase change materials for cold storage make the heat transfer process complex, and the use of physical models for characterization and temperature prediction can be challenging. In order to predict the variation of the thermal environment in a temperature-controlled container with a cold energy storage system, we propose an LSTM model based on historical temperature data in which the trends of temperature variations of the fresh-keeping area, the phase change material (PCM), and the fresh products can be predicted immediately without considering the complex heat transfer process. An experimental platform of a temperature-controlled container with a cold energy storage system is built to obtain the experimental data for the prediction model’s construction and validation. The prediction results based on the LSTM model are compared to the results of a physical model. In order to optimize the input data for better prediction performance, the proportion of input samples from the dataset is set to 80%, 50%, 20%, and 10%. The prediction results from different input groups are compared and analyzed. The results show that the LSTM model is able to accurately predict temperature variations of the fresh-keeping area and products, and the predicted values are in agreement with the actual values. The LSTM-based prediction model has a higher accuracy compared to the physical-based prediction model; the RMSE, MAE, and MAPE are 0.105, 0.103, and 0.010, respectively, and the relative error for the prediction of effective control hours of environmental temperature is 0.92%. It is suggested to use the initial 20% of the historical temperature data as the input to predict the future temperature variation for better prediction performance. The results of this paper offer valuable insights for accurate temperature prediction in the fresh-keeping environment with a cold energy storage system.