Research on Predicting Container Vessel Handling Time Based on Improved Deep Residual Shrinkage Net

In the era of modern global trade, ports play a pivotal role as comprehensive transportation hubs connecting land and sea routes. Among the limited resources, berths stand out as the most critical, directly influencing the core competitiveness of ports. Efficient utilization of berths is crucial for optimal port performance. Predicting the vessel turnaround time, a key factor in berth scheduling, holds significant practical importance. Presently, ports commonly employ a rough estimation method, dividing the total container number by the product of allocated quay cranes number and a constant efficiency factor. However, this approach lacks consideration for other influencing factors, resulting in inaccurate and inflexible predictions. Addressing the shortcomings of traditional methods, this paper proposes an Adaptive Threshold Function based Deep Residual Shrinkage Network (ADP-DRSN). Built upon the ResNet-18 architecture, this model replaces the fixed soft threshold function with a more flexible adaptive threshold function, enhancing noise reduction capabilities while preserving feature learning capabilities. Finally, the paper compares the performance of various deep learning models under the ResNet-18 architecture using the same dataset. Results indicate that the ADP-DRSN model achieves an R
²
of 0.745 and a PearsonR of 0.879 on the test set. Inclusive of both training and test sets, the model achieves an R
²
of 0.919 and a PearsonR of 0.960, outperforming other models under the same architecture. This suggests its practical applicability and value.