IAMU Research Project FY2021 Final Report
Data fusion and machine learning for ship fuel efficiency analysis: a small but essential step towards green shipping through data analytics
| Publication Year | 2022 |
| Research Coordinator | Yuquan (Bill) Du |
| Research Partners | Shu-Ling Chen; Nataliya Nikolova; Prashant Bhaskar; Jiangang Fei; Alessandro Schonborn; Zhuo Sun |
| Publisher | International Association of Maritime Universities (IAMU) |
| DOI | https://doi.org/10.68436/r20210205 |
Abstract
The shipping industry is concerned about ship fuel/energy efficiency due to the motivation of saving bunker fuel cost and mitigating ship emissions. A foundation for various energy/emission-efficient measures is the accurate quantification of bunker fuel consumption of a ship in one day or hour given its sailing speed, draft/displacement, trim, weather conditions, and sea conditions. This study takes advantage of four industry data sources including voyage report data, AIS data, sensor data, and meteorological data, and fuses these data sources to find the best datasets for ship fuel efficiency analysis. Based on fused datasets, we experimented with state-of-the-art machine learning models to quantify a ship’s daily/hourly bunker fuel consumption, over eight 8,100-TEU to 14,000-TEU containerships of a global shipping company. When voyage report data is used as the basis for ship fuel/emission analysis, meteorological data and AIS data can be combined into voyage report data to improve the data quality. The fit errors of best machine learning models over the recommended datasets are normally within 5 ton/day, and can be as low as less than 1 ton/day. When sensor data is considered, combining meteorological data (waves, sea currents, sea water temperature) into sensor data will significantly improve the modeling accuracy. The best machine learning models achieve their R2 at 0.999 or 1.000 on the training sets, and their R2 values over the test sets are also all above 0.966. Their fit errors are below 0.75 ton/day (RMSE), or below 0.52 ton/day (MAE). The proposed datasets and models would be useful for sailing speed optimization, trim optimization, weather routing, voyage planning, and virtual (just-in-time) arrivals. We also published our computer code in Python and trained machine learning models in GitHub which is accessible to the public.
Executive summary
With promotions of the International Maritime Organization (IMO) and governmental organizations, the shipping industry has been implementing operational measures to save bunker fuel and mitigate emissions from ships, including sailing speed optimization, trim optimization, weather routing, and the virtual arrival policy. Many frustrations have been emerging during the process of implementation of these measures. These frustrations are boiled down, if not fully, to how we can quantify the synergetic contributions of many factors (speed, draft/displacement, trim, weather conditions, sea conditions) on a ship’s bunker fuel consumption rate (ton/day, ton/hour). A latest review paper points out that the basis of all operational measures for ship bunker fuel savings and emission mitigation is quantitatively modeling the relationship between fuel consumption rate and many determinants, including sailing speed, draft/displacement, trim, weather conditions, and sea conditions. This project addresses this theoretical challenge that restricts the implementation of energy-efficient operational measures by investigating the complementary roles of different data sources available to a shipping company, fusing these data sources, and employing state-of-the-art machine learning techniques.
We collected voyage report data and sensor data of eight 8,100-TEU to 14,000-TEU containerships from a global shipping company, purchased the AIS data of these ships from MarineTraffic with the financial support of the International Association of Maritime Universities (IAMU), and downloaded meteorological data from European Centre for Medium-range Weather Forecasts (ECMWF) and Copernicus Marine Service (CMEMS). Based on the information contained in these four data sources, we designated three data fusion solutions (DFSs): DFS1 fuses voyage report data and meteorological data, by considering the inaccurate information of weather and sea conditions recorded by voyage report; DFS2 further fuses AIS data into voyage report data and meteorological data because AIS data helps find the actual sailing trajectory of the ship and thus helps retrieve more accurate information of weather and sea conditions from meteorological data; DFS3 approaches sensor data as the main data source of a ship’s fuel consumption rate, and overcomes the limitation of sensor data by taking advantage of the complete information of weather and sea conditions contained in meteorological data. For each of the data fusion solutions, eight to nine datasets are constructed.
Over these datasets from three data fusion solutions, a large range of widely adopted machine learning models were experimented with, including decision tree-based models, artificial neural network (ANN), support vector machine (SVM), ridge regression (Ridge), and LASSO. Tree-based models include the basic decision tree (DT) model and models produced by two ensemble strategies: Extremely randomized trees (ET) and random forest (RF) from the bagging ensemble strategy; AdaBoost (AB), gradient tree boosting (GB), XGBoost (XG), and LightGBM (LB) from the boosting ensemble strategy. During the experiments with these machine learning models, the impacts of data normalization, hyperparameter optimization, and the randomness in splitting training sets and test sets are well addressed.
Extensive experiments were conducted to answer three research questions regarding the choice of datasets from three data fusion solutions and the choice of machine learning models. A voting scheme is developed to break down the impacts of dataset choice and model choice. When dataset choice is considered, the original voyage report dataset Set1 has a decent quality for ship fuel efficiency modeling; if more effort is paid to fuse voyage report data and meteorological data, data quality improves slightly and Set3precise can be adopted. When AIS data is available, further including AIS data might also be beneficial, which suggests the adoption of the dataset AIS5precise. Overall, the best datasets found with DFS1 and DFS2, including Set1, Set3precise, and AIS5precise, ensure accurate fit performances of best ML models: R2 on the training set is above 0.96 and even reaches 0.99 to 1.00, and R2 on the test set is between 0.74 and 0.90; the fit errors measured by RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) are between 0.5 and 4.5 ton/day. When sensor data, rather than voyage report data, is used as the main data source of ship bunker fuel consumption analysis, it elevates the modeling accuracy to a higher level, possibly the highest level if meteorological data is fused in. With DFS3, given the best dataset Sensor2, the best ML models achieve their R2 values over the training set at 0.999 or 1.000, and their R2 values over the test set are all above 0.966. Their fit errors with RMSE values are below 0.75 ton/day, and with MAE below 0.52 ton/day.
As far as ML model choice is concerned, we recommend the installation of four decision-tree based models including ET, AB, GB, and XG because they usually possess the highest fit performance and good generalization performance. Their performances are also quite robust against random splits of a dataset into training and test sets. Our experiments with DFS1, DFS2, and DFS2 reach consistent findings about the performances of ML models and rank their performances into four tiers.
Tier 1: ET, AB, GB, and XG.
Tier 2: RF, LB
Tier 3: DT, SVM, ANN
Tier 4: Ridge, LASSO.
Apart from this technical report, as the research outcomes, this project produces three research papers that have been submitted to a peer-review journal. We have also developed course material (teaching slides) for a three-hour teaching module for IMO’s TTT course on Energy Efficient Ship Operations, titled “Understanding ship fuel efficiency with real data”. To broadcast our research findings to the maritime industry, we delivered three industry presentations to industry professionals in May 2022 in Europe, Australia, and Asia, respectively. Computer code in Python in this study is published in GitHub as a software infrastructure to reduce the exploration efforts of industry professionals. Best trained machine learning models are also published in GitHub, which enables maritime researchers to estimate the bunker fuel consumption rates of different sizes of mega containerships in different sailing speed, draft, trim and weather/sea conditions. Reader can find our Python code and trained machine learning models in the URL below: https://github.com/yuquandu/Data-driven-Ship-Fuel-Efficiency-Modeling
Our recommendations for industry application are summarized in the following table.
[Table: see the full report (PDF)]
Keywords
Ship fuel efficiency, Fuel consumption rate, Voyage report, Sensor data, AIS data, Meteorological data, Data fusion, Machine learning
References
- Adland, R., Cariou, P., Jia, H., & Wolff, F. C. (2018). The energy efficiency effects of periodic ship hull cleaning. Journal of Cleaner Production, 178, 1-13.
- Adland, R., Cariou, P., & Wolff, F.C. (2019). When energy efficiency is secondary: The case of Offshore Support Vessels. Transportation Research Part D: Transport and Environment, 72, 114-126.
- Adland, R., Cariou, P., & Wolff, F. C. (2020). Optimal ship speed and the cubic law revisited: Empirical evidence from an oil tanker fleet. Transportation Research Part E: Logistics and Transportation Review, 140, 101972.
- Beúikoi, E. B., Arslan, O., Turan, O., & gloer, A.I. (2016). An artificial neural network based decision support system for energy efficient ship operations. Computers & Operations Research, 66, 393-401.
- Bennett, G.G. (1996). Practical rhumb line calculations on the spheroid. Journal of Navigation, 49(1), 112-119.
- Bergstra, J., Yamins, D., & Cox, D.D (2013). Making a Science of Model Search: Hyperparameter Optimization in Hundreds of Dimensions for Vision Architectures. In Proceedings of the 30th International Conference on Machine Learning (ICML 2013) (pp. 115-123).
- Bocchetti, D., Lepore, A., Palumbo, B., & Vitiello, L. (2015). A statistical approach to ship fuel consumption monitoring. Journal of Ship Research, 59(03), 162-171.
- Boser B.E., Guyon I.M., & Vapnik V.N. (1992). A training algorithm for optimal margin classifiers. In: Haussler D. (Ed.), In Proceedings of the Annual Conference on Computational Learning Theory (pp. 144-152). ACM Press, Pittsburgh, PA.
- Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.
- Breiman, L., Friedman, J., Olshen, R., & Stone, C. (1984). Classification and Regression Trees. Wadsworth, Belmont, CA.
- Cai, B., Xu, L.F., & Fu, F. (2019). Shear Resistance Prediction of post-fire reinforced concrete beams using artificial neural network. International Journal of Concrete Structures and Materials, 13 (1), 1-13.
- Chen, T., & Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785-794).
- C-MAP. (2022). VVOS: Vessel and Voyage Optimization Solution. https://www.oceanweather.com/forecast/VVOS/index.html. Access on 19 March 2022.
- Coraddu, A., Oneto, L., Baldi, F., & Anguita, D. (2017). Vessels fuel consumption forecast and trim optimisation: A data analytics perspective. Ocean Engineering, 130, 351-370.
- Drucker, H. (1997). Improving regressors using boosting techniques. In Proceedings of the Fourteenth International Conference on Machine Learning (pp. 107-115).
- Du, Y., Meng, Q., Wang, S., & Kuang, H. (2019). Two-phase optimal solutions for ship speed and trim optimization over a voyage using voyage report data. Transportation Research Part B: Methodological, 122, 88-114. European Union (EU). (2021). Monitoring, reporting and verification of EU ETS emissions. https://ec.europa.eu/clima/eu-action/eu-emissions-trading-system-eu-ets/monitoring-reporting-and-verification-eu-ets-emissions_en. Accessed on 19 March 2022.
- Farag, Y. B., & gloer, A. I. (2020). The development of a ship performance model in varying operating conditions based on ANN and regression techniques. Ocean Engineering, 198, 106972.
- Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1), 119-139.
- Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine. Annals of Statistics, 1189-1232.
- Geurts, P., Ernst, D., & Wehenkel, L. (2006). Extremely randomized trees. Machine Learning, 63(1), 3-42.
- Gkerekos, C., Lazakis, I., & Theotokatos, G. (2019). Machine learning models for predicting ship main engine Fuel Oil Consumption: A comparative study. Ocean Engineering, 188, 106282.
- Haykin, S.O. (2008). Neural Networks and Learning Machines. The third edition. Prentice Hall: New Jersey. Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Horinyi, A., Muxoz Sabater, J., Nicolas, J., Peubey,
- C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., & Thppaut, J-N. (2018): ERA5 hourly data on single levels from 1979 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS). (Accessed on 10-Sep-2021), 10.24381/cds.adbb2d47. https://doi.org/10.24381/cds.adbb2d47
- Hoerl, A.E., & Kennard, R.W. (1970). Ridge regression: Biased estimation for nonorthogonal problems. Technometrics, 12 (1), 55-67.
- Hornik, K., Stinchcombe, M., & White, H. (1989). Multilayer feedforward networks are universal approximators. Neural networks, 2(5), 359-366.
- IMO. (2012). Guideline for Development of a Ship Energy Efficiency Management Plan (SEEMP), MEPC, 213 (63) Annex 9.
- IMO. (2020). Fourth IMO GHG study 2020. https://wwwcdn.imo.org/localresources/en/OurWork/Environment/Documents/Fourth%20IMO%20GHG%20Study%202020%20Executive-Summary.pdf. Accessed on 19 March 2022.
- Johnson, H., & Andersson, K. (2011). The energy efficiency gap in shipping-barriers to improvement. In International Association of Maritime Economists (IAME) Conference.
- Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., & Liu, T. Y. (2017). Lightgbm: A highly efficient gradient boosting decision tree. In Advances in Neural Information Processing Systems, 3146-3154.
- Kolmogorov, A. N. (1957). On the representation of continuous functions of many variables by superposition of continuous functions of one variable and addition. Doklady Akademii Nauk, 114(5), 953-956.
- Lee, H., Aydin, N., Choi, Y., Lekhavat, S., & Irani, Z. (2018). A decision support system for vessel speed decision in maritime logistics using weather archive big data. Computers & Operations Research, 98, 330-342.
- Lepore, A., dos Reis, M.S., Palumbo, B., Rendall, R., & Capezza, C. (2017). A comparison of advanced regression techniques for predicting ship CO2 emissions. Quality and Reliability Engineering International, 33, 1281-1292.
- Li, X., Sun, B., Zhao, Q., Li, Y., Shen, Z., Du, W., & Xu, N. (2018). Model of speed optimization of oil tanker with irregular winds and waves for given route. Ocean Engineering, 164, 628-639.
- Pedersen, B.P., & Larsen, J. (2009). Prediction of full-scale propulsion power using artificial neural networks. In COMPIT’09: 8th International Conference on Computer and IT Applications in the Maritime Industries (pp. 10-12), Budapest.
- Rehmatulla, N., Parker, S., Smith, T., & Stulgis, V. (2017). Wind technologies: Opportunities and barriers to a low carbon shipping industry. Marine Policy, 75, 217-226.
- Rio, M. H., Mulet, S., & Picot, N. (2014). Beyond GOCE for the ocean circulation estimate: Synergetic use of altimetry, gravimetry, and in situ data provides new insight into geostrophic and Ekman currents. Geophysical Research Letters, 41(24), 8918-8925.
- Man, Y., Sturm, T., Lundh, M., & MacKinnon, S. N. (2020). From ethnographic research to big data analytics²a case of maritime energy-efficiency optimization. Applied Sciences, 10(6), 2134.
- Meng, Q., Zhang, Y., & Xu, M. (2017). Viability of transarctic shipping routes: a literature review from the navigational and commercial perspectives. Maritime Policy & Management, 44(1), 16-41.
- Merkel, A., Kalantari, J., & Mubder, A. (2022). Port call optimization and CO2-emissions savings± Estimating feasible potential in tramp shipping. Maritime Transport Research, 3, 100054.
- Smola, A.J., & Sch|lkopf, B. (2004). A tutorial on support vector regression. Statistics and Computing, 14 (3), 199-222.
- Soner, O., Akyuz, E., & Celik, M. (2018). Use of tree based methods in ship performance monitoring under operating conditions. Ocean Engineering, 166, 302-310.
- Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B (Methodological), 58(1), 267-288.
- UyanÕk, T., Karatu÷, d., & Arslano÷lu, Y. (2020). Machine learning approach to ship fuel consumption: A case of container vessel. Transportation Research Part D: Transport and Environment, 84, 102389.
- Wan, Z., El Makhloufi, A., Chen, Y., & Tang, J. (2018). Decarbonizing the international shipping industry: Solutions and policy recommendations. Marine Pollution Bulletin, 126, 428-435.
- Wang, S., Psaraftis, H.N., & Qi, J. (2021). Paradox of international maritime organization's carbon intensity indicator. Communications in Transportation Research, 1, 100005.
- Weintrit, A., & Kopacz, P. (2011). A novel approach to loxodrome (rhumb-line), orthodrome (great circle) and geodesic line in ECDIS and navigation in general. TransNav-International Journal on Marine Navigation and Safety of Sea Transportation, 5(4), 507-517.
- Yan, R., Wang, S., & Psaraftis, H.N. (2021). Data analytics for fuel consumption management in maritime transportation: Status and perspectives. Transportation Research Part E: Logistics and Transportation Review, 155, 102489.
- Yao, Z., Ng, S. H., & Lee, L. H. (2012). A study on bunker fuel management for the shipping liner services. Computers & Operations Research, 39(5), 1160-1172.