A novel data condition and performance hybrid imputation method for energy effcient operations of marine systems

Cheliotis, Michail and Gkerekos, Christos and Lazakis, Iraklis and Theotokatos, Gerasimos (2019) A novel data condition and performance hybrid imputation method for energy effcient operations of marine systems. Ocean Engineering. pp. 1-48. ISSN 0029-8018 (In Press)

[thumbnail of Cheliotis-etal-OE-2019-A-novel-data-condition-and-performance-hybrid-imputation-method-for-energy-effcient]
Text (Cheliotis-etal-OE-2019-A-novel-data-condition-and-performance-hybrid-imputation-method-for-energy-effcient)
Accepted Author Manuscript
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo

Download (885kB)| Preview


    Datasets with missing values can adversely affect the accuracy of any subsequent decision making, for instance in condition- and performance-monitoring for energy efficient operations of ship systems. Missing data imputation is therefore, a necessary step as it ensures that the data can reach their full knowledge extracting potential. This paper aims at developing a novel hybrid imputation method, which can be employed to condition data acquired from marine machinery systems, thus increasing the quality of the original dataset and improving the decision making for ship efficient operations. The paper includes of all necessary imputation preparatory steps and further post-imputation processes. The developed method employs a hybrid k-NN and MICE imputation algorithm which combines data mining with first-principle knowledge. The proposed hybrid approach is compared with the individual performance of k-NN and MICE algorithms and is implemented in a dataset acquired from the main engine system of an oceangoing vessel. It is shown that the hybrid approach performs best, exhibiting an average error of 2.2% compared to the k-NN and MICE algorithms with errors 5.6% and 3.3%, respectively, highlighting that the small error of the proposed novel method improves the quality of data used in condition- and performance-monitoring.

    ORCID iDs

    Cheliotis, Michail, Gkerekos, Christos ORCID logoORCID: https://orcid.org/0000-0002-3278-9806, Lazakis, Iraklis ORCID logoORCID: https://orcid.org/0000-0002-6130-9410 and Theotokatos, Gerasimos ORCID logoORCID: https://orcid.org/0000-0003-3547-8867;