Time-Series Prediction Of Voyage Data By Machine Learning Methods

Accurately predicting voyage speed in different atmospheric and oceanic conditions is crucial for optimizing shipping routes,reducing fuel consumption,and ensuring safety at sea.This study explores the utilization of a random forest model(RF)to predict voyage speed on the China-Brazil route,considering various environmental factors like atmospheric and oceanic conditions.To enhance the accuracy and performance of the RF model,the study incorporates a sophisticated approach called Time-Series Imputation with Seasonal Decomposition for RF(TISD-RF).This method combines time-series imputation and seasonal decomposition,effectively addressing missing data points and capturing complex seasonal patterns.By leveraging these techniques,TISD-RF optimizes the RF modelling process,resulting in improved accuracy and overall performance,making it a valuable tool for predicting voyage speed in time-series scenarios.In this thesis,we compare various data-driven models and select the RF framework as the foundation,considering the challenges of utilizing real-time monitoring data directly.The study utilizes historical data on sea wave conditions and voyage speed to train the model,which is then employed to predict voyage speed considering various influencing factors.The findings demonstrate a significant improvement in accuracy by employing the TISDRF method over 20 epochs,achieving an impressive accuracy rate of 98.22%.These results highlight the effectiveness of the TISD-RF method and its potential to enhance predictive accuracy within this specific context.Additionally,the TISD-RF model demonstrates superior performance compared to both deep learning-based solutions,such as transformer models,and traditional machine learning algorithms like k-nearest neighbors and support vector machines.It achieves this by showcasing a lower mean squared error(0.00006),a lower mean absolute error(0.0013),and a higher coefficient of determination(0.9999).However,the study acknowledges limitations in the data and assumptions made in the model,potentially restricting its applicability to other routes or situations.Future research could expand the model to include additional variables impacting voyage speed and explore alternative modeling techniques to compare their effectiveness.Accurately estimating voyage speed under diverse oceanic conditions is vital for ensuring marine safety.This study confirms the success of the TISD-RF model in achieving this objective,which holds significant practical implications for the shipping industry.The following are the innovative research achievements for this study:(1)TISD-RF Approach: A significant innovation lies in the introduction of the TISD-RF approach.TISD-RF effectively addresses the issue of missing data by filling in gaps within the RF framework.Furthermore,it integrates insights from seasonal decomposition,particularly valuable in industries sensitive to changing weather patterns.TISD-RF automates data imputation and streamlines preprocessing,enhancing forecast reliability while reducing resource demands.Its adaptability makes it a versatile tool for time series-based voyage speed prediction.(2)Utilizing TISD-RF for Voyage Speed Prediction: The study pioneers the application of a TISD-RF model for predicting voyage speed along the China-Brazil route,incorporating a diverse set of environmental factors including atmospheric and oceanic conditions.This innovative approach leverages the TISD-RF’s capacity to handle complex relationships within the data,extending its utility beyond traditional applications.(3)Historical Data-Based Training: In addressing the challenge of utilizing real-time monitoring data,the research employs historical data on sea wave conditions and voyage speed for model training.This shift from solely relying on real-time data to using historical data provides a practical and reliable solution for predicting voyage speed under various influencing factors.