Blending Simulation and Machine Learning Models to Advance

Energy Management in Large Ships

Eirini Barri

, Christos Bouras

, Apostolos Gkamas

, Nikos Karacapilidis

Dimitris Karadimas

, Georgios Kournetas

and Yiannis Panaretou

Department of Computer Engineering and Informatics, University of Patras, 26504 Rio Patras, Greece

University Ecclesiastical, Academy of Vella, Ioannina, Greece

Industrial Management and Information Systems Lab, MEAD, University of Patras, 26504 Rio Patras, Greece

OptionsNet S.A, Patras, Greece

Keywords: Agent-based Simulation, Energy Consumption, Cruise Ships, Machine Learning Algorithms.

Abstract: The prediction of energy consumption in large passenger and cruise ships is certainly a complex and

challenging issue. Towards addressing it, this paper reports on the development of a novel approach that

builds on a sophisticated agent-based simulation model, which takes into account diverse parameters such as

the size, type and behavior of the different categories of passengers onboard, the energy consuming facilities

and devices of a ship, spatial data concerning the layout of a ship’s decks, and alternative ship operation

modes. Outputs obtained from multiple simulation runs are then exploited by prominent Machine Learning

algorithms to extract meaningful patterns between the composition of passengers and the corresponding

energy demands in a ship. In this way, our approach is able to predict alternative energy consumption

scenarios and trigger meaningful insights concerning the overall energy management in a ship. Overall, the

proposed approach may handle the underlying uncertainty by blending the process-centric character of a

simulation model and the data-centric character of Machine Learning algorithms.

1 INTRODUCTION

It is broadly known that shipping contributes

significantly to environmental pollution. Obviously,

energy saving has many benefits both for the

environmental protection and the reduction of a ship’s

operating costs. In this direction, the International

Maritime Organization aims to reduce ship emissions

by at least 50% by 2050, while ships to be built by

2025 are expected to be a massive 30% more energy

efficient than those built some years ago (IMO,

2018). A particular ship category is that of large

passenger and cruise ships, which reportedly

consume a large amount of energy and thus constitute

https://orcid.org/0000-0001-8733-983X

https://orcid.org/0000-0001-9160-2274

https://orcid.org/0000-0003-0966-5140

https://orcid.org/0000-0002-6581-6831

https://orcid.org/0000-0002-8816-862X

https://orcid.org/0000-0001-8668-296X

https://orcid.org/0000-0003-0609-1604

an interesting area for investigating diverse energy

consumption and energy saving solutions.

Interestingly enough, while such solutions have been

thoroughly investigated in the case of buildings, very

limited research has been conducted so far for the

abovementioned ship category.

Aiming to contribute to this research gap, this

paper reports on the development of a novel

approach that builds on a sophisticated agent-based

simulation model. The model takes into account the

size, characteristics (e.g. age, special needs etc.) and

behavior of the different categories of passengers

onboard, as well as the energy consuming facilities

and devices of a ship. In addition, the simulation

Barri, E., Bouras, C., Gkamas, A., Karacapilidis, N., Karadimas, D., Kournetas, G. and Panaretou, Y.

Blending Simulation and Machine Learning Models to Advance Energy Management in Large Ships.

DOI: 10.5220/0009876601010109

In Proceedings of the 10th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2020), pages 101-109

ISBN: 978-989-758-444-2

101

model exploits spatial data corresponding to a

detailed layout of the decks of a specific ship, thus

offering customized visualizations. Finally, the

model caters for alternative ship operation modes,

corresponding to cases where the ship cruises during

the day or night, or is anchored at a port. The

proposed agent-based simulation model has been

implemented with the use of the AnyLogic

simulation software (https://www.anylogic.com/),

which provides a nice graphical interface for

modeling complex environments and allows the

extension of its simulation models through Java

code.

A novelty of our approach concerns the

exploitation of the outputs obtained from multiple

simulation runs by prominent Machine Learning

(ML) algorithms to extract meaningful patterns

between the composition of passengers and the

corresponding energy demands in a ship. In this

way, our approach is able to predict alternative

energy consumption scenarios and trigger insights

concerning the overall energy management in a ship.

In addition, it handles the underlying uncertainty

and offers highly informative visualizations of the

energy consumption.

The work reported in this paper is carried out in

the context of the ECLiPSe project

(http://www.eclipse-project.upatras.gr), which aims

at leveraging existing technological solutions to

develop an integrated energy consumption and

energy saving management system for the needs of

large passenger and cruise ships. A major task of the

project concerns the development of efficient

algorithms for the analysis and synthesis of the

associated multifaceted data, which may

considerably enhance the quality of the related

decision-making issues during the operation of a

vessel. These algorithms will trigger

recommendations about the management of energy

consumption, enabling stakeholders to gain energy

saving insights.

The remainder of this paper is organized as

follows: Section 2 outlines a literature review of

related work. Section 3 describes the proposed

approach that builds on the strengths of both

simulation and machine learning. Sections 4 and 5

present indicative experiments and corresponding

results from the application of the proposed

approach, and the analysis of the associated data

through appropriate ML algorithms, respectively.

Finally, Section 6 discusses concluding remarks and

briefly reports on future work directions.

2 RELATED WORK

As mentioned above, while considerable research has

been conducted so far on the optimization of various

energy consumption issues in buildings (being they

smart or not), very limited work has been reported so

far in the case of large ships. For instance, an agent-

based model for office energy consumption is

described in (Zhang et al., 2010). This work

elaborates the elements that are responsible for

energy consumption and presents a mathematical

model to explain the energy consumption inside an

office. The proposed model is validated through three

sets of experiments giving promising results.

Adopting another perspective, a review of

Machine Learning (ML) models for energy

consumption and performance in buildings is

presented in (Seyedzadeh et al., 2018); the motivation

of this work was the exploitation of contemporary

technologies, including network communication,

smart devices and sensors, towards enhancing the

accuracy of prediction in the above energy

management issues. On a similar research direction,

a combination of mathematical statistics and neural

network algorithms to solve diverse energy

consumption problems is proposed in (Guzhov and

Krolin, 2018); this work analyzes the associated big

data aiming to facilitate energy consumption

predictions for various types of buildings.

A comparative analysis of energy saving solutions

in buildings appears in (Chebotarova et al., 2019); the

proposed tool for assessing the effectiveness of

energy saving technologies implementation allows

not only to evaluate individual decisions, but also to

compare and rank them according to the breakeven

rate for the efficiency implementation decline. A

combination of Nearest Neighbors and Markov Chain

algorithms for the implementation of a system that is

able to support decision making about whether to turn

on or off a device in a smart home setting, thus

handling the related energy management issues, is

described in (Rajasekaran et al., 2017).

Research on the energy consumption of ships

during four different transatlantic cruises over the

period of one month is reported in (Marty et al.,

2012), through the elaboration of 250 samples of ship

data concerning ship speed, wind speed, ship draft,

latitude and longitude, etc. Data considered also

concern devices that produce power, such as the

ship’s oil and heat recovery boilers. Based on all these

data, a huge database containing thousands of files

has been built, which in turn feeds a simulation

environment that enables a ship operator to estimate

the energy consumption of cruise ships.

SIMULTECH 2020 - 10th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

102

A new method to model the ship energy flow and

thus understand the dynamic energy distribution of

the marine energy systems is introduced in

(Guangrong et al., 2013); using the

Matlab/Simscape environment, a multi-domain

simulation method is employed. As reported, the

proposed method can help people better monitor the

ship energy flow and give valuable insights about

how to efficiently operate a vessel. In a similar

research line, aiming to provide a better

understanding of the use of energy, of the purpose it

serves, and of the efficiency of its conversion on

board, an analysis of the energy system of a cruise

ship operating in the Baltic Sea is provided in (Baldi

et al., 2018); being based on a combination of direct

measurements and computational models of the

energy system of the ship, the proposed approach

ensures to provide a close representation of the real

behavior of the system.

3 THE PROPOSED APPROACH

Our approach adopts the Action Research paradigm

(Checkland and Holwell, 1998), which aims to

contribute to the practical concerns of people in a

problematic situation; it concerns the improvement

of practices and strategies in the complex setting

under consideration, as well as the acquisition of

additional knowledge to improve the way shipping

stakeholders address issues and solve problems.

Building on the strengths of existing related work,

as reported in the previous section, the proposed

approach comprises two main phases: (i) agent-

based simulation of the energy consumption in

various sites of a ship, and (ii) utilization of

prominent ML algorithms on the outputs of multiple

simulation runs to extract meaningful insights about

the relation between the passenger composition and

corresponding energy demands. Through these

phases, our approach is able to gather, aggregate and

analyze heterogeneous data representing both the

energy consumption in diverse devices and facilities

and the concentration of passengers in different

areas of a ship.

To fine tune our approach, a series of meetings

with shipping companies were conducted; through

them, we identified the types of devices and facilities

that mainly affect energy consumption in the ship

categories under consideration, and obtained valuable

information concerning the parameters to be taken

into account in energy consumption models (such as

that energy supply in a ship is provided by a number

of electric power generators, which are often of

different capacity and do not work in parallel;

estimations of energy demands according to the

number of passengers were also obtained through

such meetings). In addition, information collected

concerned the layout of ship decks and its relation to

the energy management issues investigated. Finally,

we clarified issues related to the alternative types of

passengers and how these may influence alternative

energy consumption and energy saving scenarios

(Barri et al., 2020).

3.1 Agent-based Simulation

Our approach aims to enable stakeholders predict the

energy needs of a ship (e.g. to recommend the

appropriate number of power generators to operate

each time), facilitate predictive maintenance issues

(affecting the related equipment), and hopefully

reduce the energy related operating costs. To fulfil

these aims, our simulation model takes into account

the passengers’ behavior and its dependencies with a

ship’s facilities, devices and resources.

A basic assumption of our approach is that the

energy demands in many sites of a ship (such as the

restaurant, the nightclub, the kindergarten etc.)

depend on the number of passengers who gather at

these sites at a given time, as well as their

composition in terms of type (customer or crew

member), age, gender etc. We consider that different

age groups have different paths and habits

(differences among passenger groups may even affect

the speed of a moving agent). To estimate the

populations gathered in these sites, we relied on the

behavioral preferences that large subgroups of

passengers have. For instance, we assume that young

passengers prefer to spend their time at nightclub

from 10pm to 3am, while elderly passengers prefer to

eat dinner at a fancy restaurant. Our model may also

simulate the behavior of persons with special needs

(PWSN); in particular, we assume that these people

move at a slower pace and are in most cases

accompanied by another person. Such assumptions

enable us to predict the gathered populations and,

accordingly, the energy demands during day and

night. This approach facilitates the modeling of

energy consumption, especially for ships that do not

have sophisticated energy consumption monitoring

and control systems.

In addition, according to our approach, the

passengers’ behavior is being considered and

modelled through three basic scenarios

corresponding to the ship (i) being moved during the

day, (ii) being moved during the night, and (iii) being

anchored at a destination or port. In the above

Blending Simulation and Machine Learning Models to Advance Energy Management in Large Ships

103

scenarios, we assume different behaviors from

passengers, which may result to different energy

demands. Finally, to accommodate the spatial

particularities of each ship, our approach pays much

attention to the layout of each deck. These layouts

provide us with the spatial data that are needed to

calculate the movement of passengers inside the ship.

AnyLogic offers a user-friendly import of sectional

plans (views), thus enabling the production of a more

realistic model of the distribution of ship passengers,

facilities and devices. Taking into account what our

models predict in terms of energy needs, we suggest

different policies of energy management, aiming to

reduce energy consumption.

3.2 ML Algorithms

Having thoroughly assessed the palette of broadly

used ML algorithms for the needs of our approach,

we decided to utilize two classification algorithms,

namely the Decision Trees (DT) and the K-Nearest

Neighbors (Κ-NN) algorithms. This is due to the fact

that these algorithms provide high interpretability of

their results, they have low computational cost, and

they fit well to our data structure.

Decision Trees is one of the simplest and widely

used classifiers in the field of Data Mining. They

constitute a non-parametric supervised learning

method, aiming to create a model that predicts the

value of a target variable by learning simple decision

rules inferred from the data features. DT

demonstrates excellent applicability in datasets with

either categorical or continuous variables. In addition,

it requires little data preparation and it is able to

process large amounts of data (Rokach and Maimon,

2008).

K-NN is a simple supervised ML algorithm that

can be used for both classification and regression

problems, and has been extensively applied in diverse

disciplines, such as Economics and Health

(Cunningham & Delany, 2007). It relies on labeled

input data to learn a function that produces an

appropriate output when given new unlabeled data. In

most cases, K-NN yields competitive results and has

significant advantages over other data mining

methods. It differs from other classifiers in that it does

not build a generic classification model; instead,

whenever a new record is being inserted in the

system, it tries to find similar records (nearest

neighbors) from past data stored in its memory and

assigns it the value of the dependent variable that its

neighbors have.

4 EXPERIMENTS

To demonstrate the applicability and potential of the

proposed approach, this section presents a particular

set of experiments carried out for a specific vessel. In

particular, we elaborate energy demands that are

associated with four popular facilities of a ship,

namely (i) the night club, (ii) the kindergarten, (iii)

the casino, and (iv) the restaurant. For the case under

consideration, we consider and import in the

simulation software the original deck layouts, where

all ship facilities and passenger cabins are mapped.

Moreover, we assume a total population of 3100

passengers onboard, belonging to four distinct age

groups (i.e. 1-14, 15-34, 35-54, ≥55 years old). Table

I summarizes sample data concerning the populations

of each age group in the facilities considered. For

each individual group of passengers, we create a

simple linear behavioral model in which each

individual group remains in a specific facility for

some time. We do this for every group of passengers

and every time period to create a comprehensive

routine for all passengers throughout the day. In this

way, we are able to simulate diverse scenarios, which

may be easily aggregated to create an illustrative

energy consumption map for the whole vessel.

Table 1: Distribution of age groups in various ship facilities.

Ship’s Cite Age Group 1-14 15-34 35-54

൒55

Nightclub

Percentage

0% 60% 30% 10%

Population

0 300 150 50

Kindergarten

Percentage

35% 10% 55% 0%

Population

53 15 82

Restaurant

Percentage

12% 8% 35%

45%

Population

46 30 134

172

Casino

Percentage

0% 0% 35%

65%

Population

0 0 112

208

4.1 Night Club

For the case elaborated in this paper, we generated

random samples of 500 passengers, assuming that the

percentage of passengers visiting this facility is

between 15% and 17%. This facility operates from

11pm to 5am. The conditional probability of someone

visiting the night club is shown in Table 1. We also

set the time spent there (from passengers of all age

SIMULTECH 2020 - 10th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

104

Figure 1: An instance of a simulated energy consumption scenario in the nightclub.

Figure 3: An instance of a simulated energy consumption scenario in the kindergarten.

groups) to follow a triangular distribution with a

lower limit equal to 50 minutes, mode equal to 95

minutes, and upper limit equal to 110 minutes.

Finally, we imported the layout of a specific deck,

where detailed spatial data about the cabins and the

possible pathways leading to the night club area are

described. By running the corresponding simulations,

we are able to visualize the possible concentration of

passengers during the night at this area of the ship (see

Figure 1). Consequently, by estimating the energy

requirements of the night club with respect to the

number of passengers hosted, we can calculate the

possible energy needs for the particular time period

and facility (see Figure 2). Such estimations can be

used for future predictions of energy consumption in

cases where passengers are distributed in a similar

way. Furthermore, the derived data can be statistically

analyzed to reveal the data patterns and mechanisms

that may cause the particular energy demands.

Figure 2: Energy demands corresponding to passengers’

concentration in the nightclub.

Blending Simulation and Machine Learning Models to Advance Energy Management in Large Ships

105

4.2 Kindergarten

For this facility (see Figure 3), we considered that the

passengers who visit it are mainly children (1-14

years old) and their parents (who may belong into the

age groups of 15-34 and 35-54 years old). The

opening hours of this facility are from 11am to 2pm.

We assumed that the kindergarten is not the only

choice that the above groups have for entertainment

purposes. Also, compared to other areas on the ship,

the kindergarten is not large enough to accommodate

all parents with their children. We have therefore

assumed that the proportion of passengers visiting it

daily ranges from 4% to 5.5%, i.e. from 120 up to 176

persons. The time people spend while visiting this

facility is described by a triangular distribution with a

minimum time of 50 minutes, a maximum time of 110

minutes, and a dominant value of 80 minutes. The

experiments carried out gave the concentration of

passengers shown in Figure 4.

Figure 4: Passengers’ concentration in the kindergarten.

4.3 Casino

The samples of passengers used in the particular set

of experiments concerned 320 people (i.e. 10% of

average passengers’ population). We assumed that

this facility operates from 7pm to 7am and mainly

attracts passengers that are older than 35 years old

(65% of them belonging to the ≥55 age group and the

remaining 35% to the 35-54 age group). Moreover,

passengers that visit the casino are divided into two

categories, those who choose to waste their time

exclusively in the casino during the night (20%) and

those who visit the casino for a certain time period

(they may leave and re-enter the casino during the

night). The first category concerns the 20% of the

casino visitors (their stay follows a triangular

distribution with a minimum time of 250 minutes, a

maximum of 300 minutes and a dominant value of

270 minutes). Similarly, for the rest 80% of casino

visitors we considered that their time spent follows a

triangular distribution with a minimum time of 20

minutes, a maximum time of 80 minutes and a

dominant value of 35 minutes).

4.4 Restaurant

We considered one of the available ship restaurants

(offering an “à la carte” menu, thus not being an

economic one), operating from 7pm to 11pm. This

facility concerns all passengers, regardless of age

group. We assumed that 10%-12% of passengers

(320-380 people) choose this particular restaurant;

their stay is described by a triangular distribution with

a minimum time of 75 minutes, a maximum time of

150 minutes and a dominant value of 120 minutes.

5 DATA ANALYSIS AND

SYNTHESIS

The experiments described above demonstrate

diverse features and options offered by the proposed

simulation model. To predict energy consumption in

large passenger and cruise ships, our approach

aggregates results obtained from each particular

facility of a ship and produces a corresponding time

series diagram, in which the dependent variable is the

energy consumption measured in energy units per

hour and the time interval is 10 minutes. Figure 5

illustrates the overall energy demands with regards to

the estimated gathering of passengers in the facilities

discussed in the previous section throughout the day.

Obviously, our experiments have not considered the

entirety of facilities and energy consumers available

on a ship (such as air condition, lighting, heating etc.);

however, all of them can be easily aggregated to our

model and thus provide a detailed mapping of the

overall energy consumption.

Figure 5: Cumulative concentration of passengers in four

major ship’s facilities and corresponding energy demand.

Building on the proposed agent-based simulation

model that facilitates the creation of alternative

energy consumption scenarios, we can produce

SIMULTECH 2020 - 10th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

106

realistic data that can be further elaborated by

prominent machine learning algorithms to provide

meaningful insights for managing diverse energy

consumption patterns (Deist et al., 2018). Parameters

taken into account by the proposed machine learning

algorithms also include the number of ship generators

(categorical variable), the alternative age groups and

their populations (as defined for each ship), and the

time slots considered each time (the ones adopted in

our approach are shown in Table 2).

Table 2: Time slots considered in our approach.

Time interval Time slot

7:00am – 11:59am Morning

12:00pm – 4:59pm Midday

5:00pm – 9:59pm Evening

10:00pm – 6:59am Night

Table 3: Sample of our dataset.

Compo

sition

Age Groups

PWSN

Time

slot

Number

of gen. in

simulta-

neous

operation

1-14 15-34 35-54

൒55

290 535 945 1432 97

Morn.

Mid.

Even.

Night

200 750 1200 1100 75

Morn.

Mid.

Even.

Night

175 700 1150 1150 20

Morn.

Mid.

Even.

Night

48 885 1890 550 100

Morn.

Mid.

Even.

Night

Ιn our experiments, we generated a large dataset

of 919 different passenger compositions for each time

slot. A small sample of this dataset, concerning only

four of these compositions for the time slots defined,

is presented in Table 3 (the number of generators that

operate for each data combination is calculated upon

the definition of a set of energy unit intervals and their

association with the energy produced by the

simultaneous operation of a certain number of

generators). A big part of this dataset (70%) was used

as the training set of the two ML algorithms

incorporated in our approach. Through the utilization

of these algorithms, one may predict the required

number of generators per time slot for a specific

passenger composition.

Focusing on the ‘morning’ time slot, Figure 6

illustrates the output of the Decision Tree algorithm,

which classifies alternative passenger compositions

into different numbers of power generators required.

As it can be observed, the energy consumption of the

ship in this time slot is being affected by (i.e.

positively correlated to) the ratio of passengers that

are older than 55 to those that are younger than 35

years old. The interpretation of this may be that older

people use to be more active in the morning

(compared to young populations). Results shown in

Figure 7 provide additional evidence in favor of the

above insight; as depicted, the correlation between

the number of generators being used in the morning

and the number of elderly passengers is positive.

Figure 6: Decision Tree classification (‘children’,

‘teenAdults’, ‘middleAgeAdults’ and ‘elderlyAdults’

correspond to the 1-14, 15-34, 35-54 and >=55 age groups,

respectively).

For the abovementioned time slot, we also applied

the K-NN algorithm. The confusion matrix produced

(this matrix is actually a technique for summarizing

the performance of a classification algorithm) showed

us limited reliability. In particular, K-NN performed

very well (with more than 95% accuracy) when

classifying compositions of passengers that were

associated with the operation of one or four

generators, while this was not the case for

compositions associated with the operation of two or

three generators (in these cases, the accuracy was

about 45% and 55%, respectively).

Blending Simulation and Machine Learning Models to Advance Energy Management in Large Ships

107

Figure 7: Scatter plot - number of generators vs population

of elderly passengers.

Table 4 summarizes a small set of predictions

produced by the K-NN algorithm for the cases of one

or four generators operating simultaneously. It is

noted that for these cases K-NN produces very similar

results to those obtained by the Decision Tree, i.e. the

energy needs are positively correlated to the ratio of

passengers that are older than 55 to those that are

younger than 35 years old. Such insights, resulting

from multiple simulation runs, were also validated by

shipping stakeholders. According to their validation

feedback, adjustments to the initially set parameters

and energy demand thresholds were performed.

Table 4: Predictions produced by K-NN algorithm.

Age Groups

PWSN

Number o

generators

simultane

ous

operation

1-14 15-34 35-54

൒55

100 755 1100 1300 75

270 668 916 1570 43

174 968 865 1021 40

243 755 1412 656 41

328 686 1450 678 82

410 995 1425 780 10

6 CONCLUSIONS

The prediction of energy consumption in large

passenger and cruise ships is certainly a hard

problem. This is mainly due to the need to

simultaneously consider the interaction between

multiple parameters and agent behaviors. To deal

with this problem, the proposed approach blends the

process-centric character of a simulation model and

the data-centric character of ML algorithms. First, by

building on a comprehensive and informative agent-

based simulation model, it facilitates the generation

and assessment of alternative energy consumption

scenarios that incorporate vast amounts of realistic

data under various conditions. Second, it advocates

the use of prominent machine learning algorithms to

aid the finding, understanding and interpretation of

patterns that are implicit in this data, ultimately

aiming to provide meaningful insights for shaping

energy saving solutions in a ship.

In any case, we need to compare the outputs of the

proposed approach with real data. As far as the

outcomes produced by the agent-based simulation

model are consistent with real data, our machine

learning algorithms will be better trained, which in

turn will enhance the accuracy of the associated

energy consumption predictions. Such reinforcement

learning activities consist one of our future work

directions.

Another research direction concerns the

investigation of alternative modes to combine

simulation and machine learning in our approach.

Specifically, we plan to consider the application of

ML algorithms prior to and within the simulation. In

the former case, we will need real data to develop

rules and heuristics that our agent-based simulation

model can then employ. In the latter, we may reuse

previously trained ML-based models or train the ML

models as the simulation is taking place.

Finally, we plan to expand the proposed agent-

based simulation model with problem-specific

algorithms and interfaces, aiming to enable shipping

stakeholders perform a progressive synthesis and

multiple criteria comparative evaluation of

alternative energy consumption configurations (a

similar approach has been proposed in (Karacapilidis

and Moraitis, 2001)).

ACKNOWLEDGEMENTS

The work presented in this paper has been co-

financed by the European Union and Greek national

funds through the Regional Operational Program

“Western Greece 2014-2020”, under the Call

“Regional research and innovation strategies for

smart specialization (RIS3) in Energy Applications”

(project: 5038607 entitled “ECLiPSe: Energy Saving

0 1000 2000 3000

numberofgenerators

populationofelderly_adults

NumberofGeneratorsvs

Elderly_Adults

SIMULTECH 2020 - 10th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

108

through Smart Devices Control in Large Passenger

and Cruise Ships”.

REFERENCES

Baldi, F., Ahlgren, F., Nguyen, T.-V., Thern, M., &

Andersson, K. (2018). Energy and Exergy Analysis of

a Cruise Ship. Energies, 11(10), 2508. doi:

10.3390/en11102508

Barri, E., Bouras, C., Gkamas, A., Karacapilidis, N.,

Karadimas, D., Kournetas, G., & Panaretou, Y. (2020).

Towards an informative simulation-based application

for energy saving in large passenger and cruise ships. In

Proc. of the 6th IEEE International Energy Conference

(ENERGYCON 2020), Gammarth, Tunisia.

Chebotarova, Y., Perekrest, A., & Ogar, V. (2019).

Comparative Analysis of Efficiency Energy Saving

Solutions Implemented in the Buildings. In Proc. of the

IEEE International Conference on Modern Electrical

and Energy Systems (MEES 2019), pp. 434-437. doi:

10.1109/mees.2019.8896691

Checkland, P.B. & Holwell, S. (1998). Action Research: Its

Nature and Validity. Systemic Practice and Action

Research, 11(1), pp. 9–21.

Cunningham, P., & Delany, S. J. (2007). k-Nearest

neighbour classifiers. Multiple Classifier Systems,

34(8), 1-17.

Deist, T., Patti, A., Wang, Z., Krane, D., Sorenson, T. &

Craft, D. (2018). Simulation assisted machine learning.

Bioinformatics. 35. 10.1093/bioinformatics/btz199.

Guangrong, Z., Kinnunen, A., Tervo, K., Elg, M., Tammi,

K. & Kovanen, P. (2013). Modeling ship energy flow

with multi-domain simulation. In Proc. of the 27th

CIMAC World Congress on Combustion Engines,

Shanghai, China.

Guzhov, S., & Krolin, A. (2018). Use of big data

technologies for the implementation of energy-saving

measures and renewable energy sources in buildings. In

Proc. of the Renewable Energies, Power Systems &

Green Inclusive Economy Conference (REPS-GIE

2018), pp. 1-5. doi: 10.1109/repsgie.2018.8488861

International Maritime Organization. (2018). Current

awareness bulletin, No. 5. Available at: http://www.

imo.org/en/KnowledgeCentre/CurrentAwarenessBulle

tin/Documents/CAB%20258%20MAY%202018.pdf

Karacapilidis, N. & Moraitis, P. (2001). Building an agent-

mediated electronic commerce system with decision

analysis features. Decision Support Systems, 32(1), 53-

69.

Marty, P., Corrignan, P., Gondet, A., Chenouard, R. &

Hetet, J-F. (2012). Modelling of energy flows and fuel

consumption on board ships: application to a large

modern cruise vessel and comparison with sea

monitoring data. In Proc. of the 11th International

Marine Design Conference (IMDC 2012), June 2012,

Glasgow.

Rajasekaran, R. G., Manikandaraj, S., & Kamaleshwar, R.

(2017). Implementation of Machine Learning

Algorithm for predicting user behavior and smart

energy management. In Proc. of the International

Conference on Data Management, Analytics and

Innovation (ICDMAI 2017), pp. 24-30. doi:

10.1109/icdmai.2017.8073480

Rokach, L. & Maimon, O. (2008). Data mining with

decision trees. Theory and applications. World

Scientific.

Seyedzadeh, S., Pour Rahimian, F., Glesk, I., & Roper, M.

(2018). Machine learning for estimation of building

energy consumption and performance: a review.

Visualization in Engineering 6, p. 5. doi:

10.1186/s40327-018-0064-7.

Zhang, T., Siebers, P. & Aickelin, U. (2010). Modelling

Office Energy Consumption: An Agent Based

Approach. SSRN Electronic Journal. Available at:

http://dx.doi.org/10.2139/ssrn.2829316

Blending Simulation and Machine Learning Models to Advance Energy Management in Large Ships

109