A Feature-Extraction-Based Lightweight Convolutional and Recurrent Neural Networks Adaptive Computing Model for Container Terminal Liner Handling Volume Forecasting

The synergy of computational logistics and deep learning provides a new methodology solution to the operational decisions container terminal handling systems (CTHS) at strategic, tactical, executive levels. Above all, tactical complexity is discussed by logistics, liner volume (LHV) has important influences on series scheduling decision problems. Subsequently, feature-extraction-based lightweight convolutional recurrent neural network adaptive computing model (FEB-LCR-ACM) presented initially predict LHV fusion multiple algorithms mechanisms, especially for specific feature extraction package tsfresh. Consequently, container-terminal-oriented service support design paradigm put forward tentatively FEB-LCR-ACM. Finally, typical large-scale China chosen implement, execute, evaluate FEB-LCR-ACM based running log around indicator LHV. In case severe vibration between 2 twenty-foot equivalent units (TEUs) 4215 TEUs, while forecasting 300 liners five years, error within 100 TEUs almost accounts 80%. When predicting operation 350 ships six deviation reaches up nearly 90%. abovementioned two experimental performances with are so far ahead results classical machine algorithm that similar Gaussian vector machine. achieves sufficiently good performance prediction architecture small datasets, then it supposed overcome nonlinearity, dynamics, coupling, CTHS partially.