Dynamic prediction model for ship cabling
Sainio, Miikka (2022)
2022
Sähkötekniikan DI-ohjelma - Master's Programme in Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2022-05-31
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202204284064
https://urn.fi/URN:NBN:fi:tuni-202204284064
Tiivistelmä
Resource planning is an essential part of successful enterprise management. Data-based solutions are increasingly used in industry to provide more accurate predictions about needed resources. In this thesis, different machine learning models were applied and compared to provide data-based tools for two tasks: predicting the cable quantity needed in the ship, and forecasting the progress of the cabling process. The research aimed to answer questions of how well the needed cable quantity can be predicted from the size of the ship, whether dynamic models can be used to improve the cable quantity prediction, and how accurately the cabling progress can be forecasted based on historical project data.
The research was carried out with data from the Meyer Turku shipyard from their archive of delivered ship projects. The data used to train and evaluate the models consist of 11 ships. The data was diverse in terms of ship sizes and types. The prediction capabilities of the models were evaluated using MAPE and RMSPE measures with cross-validation. The CV method was leave-p-out where two ships were left out from the training set per cycle.
For the first task, two predictors, gross tonnage and ship area in square meters, were used for predicting the cable quantity based on ship size, were used to test the relation between the ship size and cable quantities. The models used were linear regression and K-Nearest Neighbor Regression. Additionally, ensemble learning was utilized for dynamic quantity predictions by combining the linear regression model with dynamic cabling data. The Linear regression and KNN achieved 7.1% and 6.7% mean errors, respectively. The KNN model had over 100% error at the worst CV cycle. Compared to the maximum 20% linear regression error, the KNN model was considered unreliable for production use. The ensemble model achieved 3.1% mean error. Results indicate strong linear correlation between the ship size and needed cable quantity.
The forecasting models utilized two types of data, progress data of completed projects and progress data of observed part of the process that is being forecasted. S-curve fitting and Gaussian Process Regression were used for forecasting. Compared to the S-curve model, GPR model performed better at the beginning of the process, which indicates that old project data helps the forecasting. At the second half of the processes, there is no significant difference in performance between the two forecasting models.
The research was carried out with data from the Meyer Turku shipyard from their archive of delivered ship projects. The data used to train and evaluate the models consist of 11 ships. The data was diverse in terms of ship sizes and types. The prediction capabilities of the models were evaluated using MAPE and RMSPE measures with cross-validation. The CV method was leave-p-out where two ships were left out from the training set per cycle.
For the first task, two predictors, gross tonnage and ship area in square meters, were used for predicting the cable quantity based on ship size, were used to test the relation between the ship size and cable quantities. The models used were linear regression and K-Nearest Neighbor Regression. Additionally, ensemble learning was utilized for dynamic quantity predictions by combining the linear regression model with dynamic cabling data. The Linear regression and KNN achieved 7.1% and 6.7% mean errors, respectively. The KNN model had over 100% error at the worst CV cycle. Compared to the maximum 20% linear regression error, the KNN model was considered unreliable for production use. The ensemble model achieved 3.1% mean error. Results indicate strong linear correlation between the ship size and needed cable quantity.
The forecasting models utilized two types of data, progress data of completed projects and progress data of observed part of the process that is being forecasted. S-curve fitting and Gaussian Process Regression were used for forecasting. Compared to the S-curve model, GPR model performed better at the beginning of the process, which indicates that old project data helps the forecasting. At the second half of the processes, there is no significant difference in performance between the two forecasting models.