Abstract: In the supply chain management of fresh agricultural products, the performance of cold chain logistics supply chain is one of the important factors for the sustainable development of fresh enterprises. However, there are few researches on performance evaluation and prediction at present. Therefore, considering the final output of cold chain logistics operation, combined with the characteristics of recycling in sustainable development and cross-period operation of supply chain, A Dynamic Cyclic Network Data Envelopment Analysis model is designed for performance evaluation. It is found that the changes of expected output and unexpected output have a significant impact on the performance of cold chain logistics supply chain. The model considers the interaction between different time periods, provides a perspective for managers to examine the changes of relevant factors in supply chain performance, and provides help for managers to make resource allocation and decision. Meanwhile, Levenberg-Marquard algorithm is used to im-prove Convolutional Neural Networks to predict performance with minimum absolute error. The results show that LM-CNN effectively improves the accuracy of model prediction results. By using artificial neural network to predict the performance of supply chain in the future period, it helps managers to take active measures before the inefficiency, so as to improve the efficiency of supply chain operation.

Key words: fresh produce supply chain, sustainable development, performance evaluation, performance prediction, DV-NDEA model, LM-CNN predictive model