Container ships are the high-value-added transportation main body of the Xijiang
Golden Waterway, and the prediction of their lock waiting time is of great
significance for ship scheduling and transportation efficiency optimization.
However, the actual waiting time of container ships passing through locks is
affected by multiple factors such as navigation direction, seasonal water level,
and queuing length, showing a highly right-skewed distribution with strong
volatility, which brings great challenges to accurate prediction. At present,
most lock scheduling systems only provide queuing quantity but cannot give
estimated waiting time, and traditional prediction methods have problems such as
low accuracy and poor robustness. To solve these problems, this study analyzed
the influencing factors of the lock waiting time of container ships in the
Xijiang Golden Waterway, constructed a feature engineering system including
queuing intensity, seasonal characteristics and direction interaction, and used
the Gradient Boosting Regression model to predict the lock waiting time. The
results show that the lock waiting time is significantly affected by navigation
direction, seasonal changes, queuing length and other factors, presenting a
highly right-skewed distribution law. The mean absolute error (MAE) of the
Gradient Boosting Regression model is 8.43 hours, which is 57.7% better than
that of the Long Short-Term Memory (LSTM) model, 65.3% better than that of the
historical average benchmark method, and the coefficient of determination
is increased from
-0.0151 to 0.7605. The above research results can provide accurate lock waiting
time prediction services for shipping enterprises and improve shipping
efficiency.