CNN–LSTM-Based Prediction of Marine Engine CO₂ Emissions under Real Operating Conditions
AUTHORS
Ariana Delgado,Department of Mechanical Engineering, Aalto University, 02150 Espoo, Finland
Mikael Virtanen,Department of Mechanical Engineering, Aalto University, 02150 Espoo, Finland
Juho Laaksonen,Maritime Technology Research Group, University of Turku, 20014 Turku, Finland
ABSTRACT
Reducing Greenhouse Gas (GHG) emissions from maritime transport remains a critical engineering challenge, particularly in regions such as Finland, where shipping activity and environmental regulations are both highly intensive. Accurate quantification of Carbon Dioxide (CO₂) emissions from marine engines is essential for supporting decarbonization strategies; however, direct onboard measurement is often constrained by technical, spatial, and economic limitations. This study proposes a data-driven approach for predicting CO₂ emissions using deep learning techniques under real operating conditions. Engine operational data were collected from a medium-sized passenger vessel equipped with a slow-speed diesel engine, including exhaust gas temperature, combustion pressure, compression pressure, and cooling water temperature. Two predictive models—Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTMs)—were developed and trained to estimate CO₂ emissions using these inputs. Model performance was evaluated using mean absolute error (MAE), Root Mean Square Error (RMSE), and correlation coefficients, and predictions were validated against measurements from a portable emission measurement system. The results demonstrate that both models achieve high predictive accuracy, with the CNN model outperforming the LSTM architecture across all evaluation metrics. Specifically, the CNN model yielded lower MAE and RMSE values and exhibited stronger correlation with observed emission data, indicating superior ability to capture nonlinear relationships among engine performance parameters. These findings highlight the effectiveness of deep learning approaches for emission prediction in complex maritime environments. The proposed methodology provides a scalable, practical solution for estimating CO₂ emissions when direct measurement is infeasible, offering significant potential for integration into Finland's maritime engineering practices and broader emission-monitoring frameworks. This work contributes to advancing intelligent, data-driven tools to support sustainable shipping and compliance with evolving environmental regulations.
KEYWORDS
Carbon dioxide emissions, Marine engines, Deep learning, Convolutional neural networks (CNN), Long short-term memory (LSTM), Emission prediction modeling
REFERENCES
[1] P. T. Aakko-Saksa, K. Lehtoranta, N. Kuittinen, A. Järvinen, J. Jalkanen, K. Johnson, H. Jung, L. Ntziachristos, S. Gagné, C. Takahashi, P. Karjalainen, T. Rönkkö, and H. Timonen, “Reduction in greenhouse gas and other emissions from ship engines: Current trends and future options,” Progress in Energy and Combustion Science, vol. 94, pp. 101055, (2022) DOI:10.1016/j.pecs.2022.101055(CrossRef)(Google Scholar)
[2] Finnish Ministry of Economic Affairs and Employment, “Finland’s Climate and Energy Strategy 2035,” Helsinki, Finland, (2022)
[3] E. Majamäki, L. Johansson, T. Grönholm, and J. Jalkanen, “Improving the global prediction of shipping emissions by modeling the effects of ambient conditions,” Science of The Total Environment, vol. 997, pp. 180126, (2025) DOI:10.1016/j.scitotenv.2025.180126(CrossRef)(Google Scholar)
[4] International Maritime Organization (IMO), “Initial IMO Strategy on Reduction of GHG Emissions from Ships,” London, UK, (2018)
[5] A. Bogdanowicz and T. Kniaziewicz, "Marine diesel engine exhaust emissions measured in a ship's dynamic operating conditions," Sensors, vol. 20, no. 22, pp. 6589, (2020) DOI:10.3390/s20226589(CrossRef)(Google Scholar)
[6] H. Winnes and E. Fridell, “Particle emissions from ships: Dependence on fuel type,” Journal of the Air & Waste Management Association, vol. 59, no. 12, pp. 1391–1398, (2009) DOI:10.3155/1047-3289.59.12.1391(CrossRef)(Google Scholar)
[7] C. J. Huang and P. H. Kuo, “A deep CNN-LSTM model for particulate matter (PM2.5) forecasting in smart cities,” Sensors, vol. 18, no. 7, pp. 2220, (2018) DOI:10.3390/s18072220(CrossRef)(Google Scholar)
[8] C. Tian, J. Ma, C. Zhang, and P. Zhan, “A deep neural network model for short-term load forecast based on long short-term memory network and convolutional neural network,” Energies, vol. 11, no. 12, pp. 3493, (2018) DOI:10.3390/en11123493(CrossRef)(Google Scholar)
[9] J. J. Corbett, J. J. Winebrake, E. H. Green, P. Kasibhatla, V. Eyring, and A. Lauer, “Mortality from ship emissions: A global assessment,” Environmental Science & Technology, vol. 41, no. 24, pp. 8512–8518, (2007) DOI:10.1021/es071686z(CrossRef)(Google Scholar)
[10] Fan, A., Li, B., Yan, J., Yang, L., Shu, Y., Xiong, Y., & Zhang, M. (2024). Analyzing ship emissions under complex operating conditions: Insights from onboard measurement data. Marine Pollution Bulletin, 209, 117280. DOI: 10.1016/j.marpolbul.2024.117280(CrossRef)(Google Scholar)
[11] S. Chen, Q. Meng, P. Jia, and H. Kuang, “An operational-mode-based method for estimating ship emissions in port waters,” Transportation Research Part D: Transport and Environment, vol. 101, pp. 103080, (2021) DOI:10.1016/j.trd.2021.103080(CrossRef)(Google Scholar)
[12] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, pp. 436–444, (2015) DOI:10.1038/nature14539(CrossRef)(Google Scholar)
[13] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, (2019)
[14] M. Abumohsen, A. Y. Owda, M. Owda, and A. Abumihsan, “Hybrid machine learning model combining CNN-LSTM-RF for time series forecasting of solar power generation,” E-Prime - Advances in Electrical Engineering, Electronics and Energy, vol. 9, pp. 100636, (2024) DOI:10.1016/j.prime.2024.100636(CrossRef)(Google Scholar)
[15] T. Chen and M. Wang, “Deep learning-based carbon emission forecasting and peak carbon pathways in China’s logistics industry,” Sustainability, vol. 16, no. 5, pp. 1826, (2023) DOI:10.3390/su16051826(CrossRef)(Google Scholar)
[16] S. Mahjoub, L. Chrifi-Alaoui, B. Marhic, and L. Delahoche, “Predicting energy consumption using LSTM, multi-layer GRU and Drop-GRU neural networks,” Sensors, vol. 22, no. 11, pp. 4062, (2021) DOI:10.3390/s22114062(CrossRef)(Google Scholar)
[17] L. Johansson, J. Jalkanen, and J. Kukkonen, “Global assessment of shipping emissions in 2015 on a high spatial and temporal resolution,” Atmospheric Environment, vol. 167, pp. 403–415, (2017) DOI:10.1016/j.atmosenv.2017.08.042(CrossRef)(Google Scholar)
[18] H. Xing, S. Spence, and H. Chen, “A comprehensive review on countermeasures for CO₂ emissions from ships,” Renewable and Sustainable Energy Reviews, vol. 134, pp. 110222, (2020) DOI:10.1016/j.rser.2020.110222(CrossRef)(Google Scholar)