Balanced Smart Predict‐Then‐Optimize Framework for Container Yard Intelligent Retrofit Decision‐Making

Intelligent transformation of container yards is essential for increasing terminal capacity. Demand uncertainty may lead to the risk of delays in port container operations. Traditional “predict‐then‐optimize” (PTO) frameworks often yield suboptimal results because forecasting goals are isolated from the actual decision objectives. To bridge this gap, we use a balanced smart predict‐then‐optimize (Bal‐SPO) framework for multiperiod retrofit scheduling. This approach integrates an long short‐term memory (LSTM) network with a mixed‐integer programming (MIP) model through decision‐focused learning. A key innovation is the inclusion of a “balance” term in the loss function. This feature stops too many zones from being closed at once for retrofitting. Experimental results based on historical data demonstrate that this framework reduces operational costs by 17.1% compared to traditional PTO frameworks. It also reduces delay penalty costs by 16.9%. Ultimately, this research provides port managers with a robust and practical decision‐support tool for implementing intelligent yard retrofits in uncertain environments.