Recognition and Prediction of Multi-Level Handling Complexity at Automated Terminals Based on ARIMA

Abstract

Accurate recognition and prediction of the multi-level handling complexity in automated container terminals (referred to as “automated terminals”) is a prerequisite for improving the effectiveness of scheduling and realizing intelligent operation and maintenance. According to the operating characteristics of the automated terminal equipment, the operating network is constructed of automated terminals that describe the characteristics of operating complexity. We use K-medoids and a light gradient boosting machine (LightGBM) to construct a K-LightGBM model that recognizes multi-level handling complexity. The key lies in the reasonable construction of prediction models. A hyper-heuristic autoregressive integrated moving average (ARIMA) model is proposed to address the problem that the ARIMA is ineffective in predicting nonlinear data. We combine ARIMA and the LightGBM model to establish an ARIMA-LightGBM model to predict multi-level handling and residuals. To improve accuracy, we propose the two residual prediction strategies of direct prediction and limited residual boundary prediction based on the residuals generated by ARIMA. We propose a hyper-heuristic algorithm based on a gradient descent-trust region (GD-TR) to compute the weights of predicted values under the two strategies, which improves the global search capability by GD and TR. The particle swarm optimization algorithm, simulated annealing algorithm, and ant colony optimization algorithm are low-level heuristics. Simulation results show that the proposed model possesses the lowest root mean square error on all characteristics compared to ARIMA, E-ARIMA, and ARIMA-LSTM. Therefore, the proposed model is very effective in improving the accuracy of predicting the multi-level handling complexity at automated terminals.

Keywords

• automated container terminal
• handling complexity
• complex network
• machine learning
• hyper-heuristic algorithm