Deep learning-based forecasting of port cargo throughput using PCA and error correction multivariate LSTM

In this study, we explore deep learning's role in predicting port cargo throughput, focusing on Ningbo Zhoushan Port within the context of smart cities. Using principal component analysis (PCA), cargo throughput is meticulously analysed. Through correlation analysis, gross domestic product (GDP) and container throughput are identified and incorporated as external inputs into a multivariate long short-term memory (LSTM) model, enhancing prediction accuracy. Errors from the initial predictions are utilised to generate error sequences for correction, further improving accuracy. The model is benchmarked against the vector autoregressive (VAR), Holt-Winters, and Grey models. Notably, our PCA-based LSTM model exhibits an average absolute percentage error of 0.31%, an average absolute error of 48.7, and a Root Mean Square Error (RMSE) of 56.15. These findings underscore deep learning's instrumental role in advancing smart city development and optimising port operations, showcasing artificial intelligence's (AI) transformative impact on urban infrastructure and planning.