FOINIKAS and ANEK are ship-owning companies, operating respectively tankers and passenger vessels. FOINIKAS serves customers around the globe all year in non-ordinary trips. ANEK, on the other hand, runs voyages within Greece and between Greece and Italy. Both companies operate in accordance with the Energy Efficiency Design Index (EEDI), that was made mandatory for new ships, and the Ship Energy Efficiency Management Plan (SEEMP), made mandatory for all ships at MEPC 62 (July 2011) with the adoption of amendments to MARPOL Annex VI (resolution MEPC.203(62)) by Parties to MARPOL Annex VI.
An important factor contributing to the financial viability of the two companies is the reduction of fuel consumption. Fuel consumption is not constant, but on the contrary, varies significantly depending upon a plethora of parameters, which include both environmental conditions (e.g. wind direction, wind speed, wave height and many more), as well as operational choices by the ship’s captain (e.g. average speed). During operation, multiple sensors are currently used for measuring operational and performance parameters that impact fuel consumption. Examples of such monitoring activities are Global Positioning System (GPS) (to measure velocity, direction, route etc.), Voyage Data Recorder (VDR), RMP in main engine (M/E), M/E power, pressures (water, lubes etc.), temperatures (exhaust gases, cooling waters etc.), and of course consumptions of liquids including fuel, water and lubricants.
Towards this end, both ANEK and FOINIKAS are eager to investigate how the environmental conditions and the operational decisions taken (including but not limited to the aforementioned examples) influence fuel consumption. Within the context of this scenario, both companies wish to investigate the relationship between ship displacement and trim, as well as meteorological and environmental conditions (e.g. sea condition, wind force, wind direction etc.), and fuel consumption. The companies wish to investigate the relationship and the potential impact of sea conditions, the wind force, as well as other environmental indicators, made available through datasets not provided by the company (but for example from open source sources, such as Copernicus), on the ships’ average speeds and engine RPMs. These factors are directly associated with the engine fuel consumption, and the goal is to estimate whether, based upon the weather forecast updates, the vessel could initiate the trip with reduced initial fuel load, and whether it could reach the destination on time with reduced speed, thus not jeopardizing the enterprise’s reputation.
The methodology of the investigation of the fuel consumption reduction consists of the following steps: i) a data preparation step, ii) data analysis and machine learning step and iii) the execution step. The data preparatory step involves data aggregation and preprocessing executed by the BigDataOcean administrator in cooperation with the interested stakeholder. The data analysis and machine learning step aims to train the algorithms that will later facilitate the extraction of insights and is executed by a corporate data analyst. The last phase of the fuel consumption reduction methodology consists of a set of steps that execute the workflow designed by the analyst and visualize the results through tailored reports on behalf of the interested stakeholder.
Data Aggregation & Pre-Processing.
This process consists of a set of steps facilitating the data acquisition from the interested stakeholder, as well as its cleaning, curation and storage. More specifically, the steps (of which the stakeholder may execute a subset) foreseen by the BigDataOcean methodology include:
- Data Acquisition: The data acquisition step consists of two sub steps: one regarding internal data assets made available from the shipping companies, and one regarding external data made available from the open data sources (or other sources).
- The data acquisition sub-step regarding internal data assets made available from the shipping companies involves harvesting dynamic data regarding the vessels’ trip characteristics (including the latitude and longitude of the ship’s position, the speed and the direction), which change during the trip, as well as additional, more static data which are not monitored per regular intervals such as liquids (including fuel, water and lubricants) consumptions, displacement of the vessel and and number of passengers, cars and trucks loaded on the vessel. Real-time information regarding the vessel trip characteristics is gathered by the ship’s sensors and on board-real time systems. As the internet connectivity is limited during a ship’s trip (especially on long trips around the globe), the data is sent at specific intervals, or when arriving at a port, in batches. The data is then gathered centrally from the IT systems and databases of each company. In the context of this scenario these datasets were exported from the IT systems and databases of the companies and were made available as CSV files to be uploaded to the BigDataOcean platform.
- In terms of external datasets, the data acquisition consists of the identification and retrieval of historical meteorological and environmental data concerning the same time period and geographical coverage as the internal data assets, from Copernicus. These datasets include environmental and meteorological information that could have an impact on the fuel consumption, in conjunction with the operational choices by the ship’s captain (e.g. average speed).
Pre-processing and cleaning. Pre-processing of the data involves selecting and cleansing the data in order to avoid out-of-range values or invalid data, impossible data combinations and missing values. Analysing data that has not been carefully pre-processed can produce misleading results. Moreover, pre-processing involves the enhancement of the representation and quality of data. In the context of this scenario, the internal data assets that are provided by the shipping companies are pre-processed, ensuring that errors associated with conformance to specific constraints are identified and resolved, including for example ensuring conformance to specific data types, conformance to value representation, conformance to uniformity, conformance to range constraints, conformance to pre-defined values and more. During this step, irrelevant and redundant information present or noisy and unreliable data is omitted in order to improve the quality of the further analysis.
Curation. In the context of this scenario, this step includes the harmonisation and semantic uplifting (when appropriate and applicable) of the raw input both from the internal as well as from the external data sources, according to the BigDataOcean common Context Model and vocabularies utilised. The semantic enrichment of the raw input, aligning it with the BigDataOcean common Context Model and vocabularies, allows for further processing capabilities by the consumption applications, including queries, visualisation, and analytics.
Storage. At the final phase, semantically enriched datasets are stored within the main storage database of the BigDataOcean Platform.
This analytics and machine learning step is associated with training the algorithms that will later facilitate the extraction of insights, executed by the corporate data analyst. Within the context of this step, and in the context of the investigation of the fuel consumption reduction scenario, the consortium proceeded with experimenting with a variety of algorithms (including, for example, linear regression, decision tree regression, and random forest regression), as well as with algorithmic parameters (including, for example, maximum number of iterations to allow for the model optimization, regularization parameter to control overfitting, and maximum depth of the tree) and parameters associated with the datasets at hand (including, for example, fuel_oil_initial_supply, water_initial_supply, number_of_cars, number_of_trucks, and draft), in order to examine what the optimal algorithmic configuration that could facilitate the most proper execution of the scenario and extract the most accurate results is.
The execution of the scenario can be illustrated with the use of the following schematics:
Table 1. Fuel Consumption Investigation Execution
Select parameters affecting fuel consumption: The user selects a set of parameters to be related to the vessel’s fuel consumption, and the respective time frame. These are either provided by the company (e.g. ship’s average speeds, engine RPMs, wave heights, wind force) or by external entities (e.g. specific environmental indicators).
Select outputs and visualisation options: The time frame selected in the previous step will be considered for assessing the desired output, e.g. the increased fuel consumption in relation to the selected parameters. Different visualisation options will be available, namely (i) minimum and/or maximum values to be shown; and (ii) type of graph (time series, histogram, matrix).
Obtain report: After setting the desired graph, the user can generate a pdf report that describes all the relevant data (e.g. the considered dataset). This report can be used as the basis for the fuel consumption reduction strategy to be designed by the company.
Initial Insights & Results.
The first version of the scenario demonstrated the potential of associating environmental conditions (e.g. SeaState representing the mean value of the wind force during the trip), as well as operational choices by the ship’s captain (e.g. average speed), with the fuel consumed during the trips. A set of reports were produced demonstrating to the stakeholders how the draft of the vessel (which is directly associated with the vessel’s cargo/load and ballast) is directly associated with the fuel consumption, and similarly how the fuel consumed is impacted by the wind force and the speed during the trip.
As previously mentioned, the first version of the scenario demonstrated the potential of associating environmental conditions (e.g. SeaState) made available through the internal datasets, as well as operational choices by the ship’s captain, with the fuel consumed during the trips. Future work includes correlating the fuel consumed, not only with the variables from the internal datasets provided by the company, but also (as indicated in the Methodology section) with external datasets containing meteorological and environmental variables of the same time period and geographical coverage. These external datasets could come from a variety of sources, including Copernicus, so as to produce more accurate insights, which in turn will lead to better estimations of whether, based upon the weather forecast updates, the vessel could initiate the trip with reduced initial fuel load, and whether it could reach the destination on time with reduced speed.