Analiza operativnih podatkov pristanišča in napoved obračalnega časa tovornih ladij

Martinčič, Tomaž (author), Sadikov, Aleksander (mentor), Štepec, Dejan (co-mentor)

Pomorska industrija je bistvenega pomena za svetovno gospodarstvo, saj se več kot 90 % svetovne trgovine prevaža po morju. Kljub razsežnostim pomorske industrije, se ta na informacijskem področju še ni toliko razvila, kot nekatere druge industrije. V diplomskem delu je predstavljena raziskovalna analiza operativnih podatkov iz pristanišča Bordo, ter metodologija in rezultati gradnje modela za napoved obračalnega časa tovornih ladij. Napovedi obračalnih časov se uporabljajo pri planiranju razporeditve resursov in prostora. Model temelji na odprtokodni knjižnici za strojno učenje CatBoost. Napovedni model je validiran z metodami prečnega preverjanja na 11 letih zgodovinskih podatkov ter na dveh mesecih podatkov v živo. Povprečna absolutna napaka MAE na zgodovinskih podatkih je 13,66 ur in 14,12 ur na podatkih v živo. Model dosega bistveno boljše rezultate od trenutno uporabljenega sistema v pristanišču, pri kateremu napaka MAE znaša 41,48 ur

• Language: Slovenian
• Keywords: EDA, strojno učenje, ETD, predviden čas odhoda, napoved obračalnega časa, pomorska industrija, pristanišča
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2020
• PID: 20.500.12556/RUL-119827
• COBISS.SI-ID: 30798595
• Publication date in RUL: 11.09.2020

Secondary language

• Language: English
• Title: Ports operative data analysis and cargo vessels turnaround time prediction

Abstract

The maritime industry is essential to the world economy, as more than 90% of world trade is carried by the sea. Despite the dimensions of the maritime industry, it has not yet developed as much in the field of informatics as some other industries. The diploma thesis presents an exploratory data analysis of operational data from the port of Bordeaux, as well as the methodology and results of building a model for cargo vessels turnaround time predictions. Turnaround times are used to plan the allocation of resources and space. The model is based on the open-source CatBoost machine learning library. The predictive model is validated using a cross-validation method on 11 years of historical data and two months of live data. The mean absolute error MAE on the historical data is 13.66 hours and 14.12 hours on live data. The model outperforms the currently used system in the port, where the MAE is 41.48 hours.

• Keywords: EDA, machine learning, ETD, estimated time of departure, turnaround time prediction, maritime industry, ports