The work aims to achieve research and development of an application terminal for customized cold chain logistics packaging solutions based on the feedback artificial neural network algorithm. Through experiments, a dataset of insulation performance of refrigerated logistics requirements under various influencing factors was established. The inherent limitations of classic feedback neural network algorithms were summarized. On this basis, a more robust and efficient algorithm - SAHId-Elman was proposed. Based on this algorithm, a human-computer interaction interface for customized cold chain logistics packaging solutions was designed. The proposed SAHId-Elman model outperformed the other three models in terms of evaluation indicators such as coefficient of determination R2, mean square error MSE, and mean absolute error MAE, reaching 0.999 88, 0.006 38, and 0.056 31 respectively; The application terminal of the designed human-computer interaction interface could be operated without a professional technical background, with a single run time of only 26 seconds, and the predicted insulation time was 12.86 hours, with an error of only 0.06 hours compared with the 12.92 hours in the physical experiment, and the prediction accuracy rate was 99.53%. In conclusion, the application terminal designed based on the optimized SAHId-Elman algorithm can accurately predict insulation time, indicating its feasibility and reliability in the application of cold chain logistics packaging solution formulation.