[ Original Paper ]
Journal of Advanced Marine Engineering and Technology - Vol. 46, No. 6, pp.387-393
ISSN: 2234-7925 (Print) 2765-4796 (Online)
Print publication date 31 Dec 2022
Received 15 Nov 2022 Revised 01 Dec 2022 Accepted 07 Dec 2022

# Development of energy management system using neural network for power electric system

SeongWan Kim1 ; SeokCheon Kang2 ; MinKi Son3 ; HyeonMin Jeon
1Professor, Ocean Polytech Team, Korea Institute of Maritime and Fisheries Technology, Tel: +82-51-620-5789 seongwan.kim@seaman.or.kr
2M. S. Candidate, Department of Marine System Engineering, Korea Maritime & Ocean University rkdtjrcjs294@g.kmou.ac.kr
3M. S. Candidate, Department of Marine System Engineering, Korea Maritime & Ocean University dlths63@g.kmou.ac.kr

Correspondence to: Professor, Department of Marine System Engineering, Korea Maritime & Ocean University, 727, Taejong-ro, Yeongdo-gu, Busan 49112, Korea, E-mail: jhm861104@kmou.ac.kr, Tel: +82-51-410-4841

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0), which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

## Abstract

With the international community's efforts to reduce carbon dioxide, the technology to improve the energy efficiency of ships has been significantly advanced owing to the development of propulsion and power systems. Among them, the power management system in the electric propulsion system, which is a eco-friendly technology, has been transformed to an energy management system(EMS) that integrates the battery management system or green technologies to improve system efficiency. The application of EMS can improve the efficiency of the power system by applying this technology not only to the electric propulsion system but also to the power generation system in merchant ships using the conventional direct mechanical propulsion system. Optimum control of the power generation source according to the rule-based strategy, including the load and state of charge of the battery, referring to the designer's intentions, is presented in this study. The neural network controller logic is developed, and the stability of the control system is analyzed using Matlab/Simulink based on various environments. As a test result, according to the step-by-step changes in load and battery status, the designed learning value shows a stable generator output command value in all operation areas and a stable generator optimal control output according to various conditions.

## Keywords:

Rule based strategy, Neural network, Power management system, Energy management system, Optimum control

## 1. Introduction

The international community has strengthened regulations on emissions from ships, and the IMO(International Maritime Organization) has tightened regulations on air pollution emitted from ships [1]. Accordingly, various alternatives such as EEDI(Energy Efficiency Design Index) power system improvement, low-carbon fuel, etc. have been proposed to reduce carbon dioxide [2], and the engine system is also being improved as a solution [3]. This affects not only the change of the propulsion system, but also the improve the efficiency of the power system. In the power system, the conventional parallel running load-sharing power generation system has limitations such as frequent maintenance cycle, increased emissions due to the fuel consumption increase by the low efficiency of the power source in the case of low-load operation [4]. To improve this, various fuels for engines, types of sources, and efficiency improvement measures for power generation sources have been introduced [5][6]. Among them, a hybrid power system using a battery system can also be a solution to improve system efficiency [7]. The efficiency of the system can be improved by the optimal operation of the power source and the battery system’s compensation. Comprehensive control of the battery management system linked with the existing power management system can be managed by the EMS(Energy Management System) [8]; the EMS controller is devised and used by various control methods [9]. In this study, by applying a rule-based strategy, the operation mode of the power generation system according to the load condition and battery SOC and the operation command for each generator were set. In addition, the EMS control results considering the SOC of the battery as a power source and the system without considering the SOC’s effects on the results of the EMS controller were also examined through simulation. To improve the efficiency of the system, the optimal operation point of the power source is designed to be operated in as many sections as possible, and the battery system is designed to compensate for variable load conditions.

## 2. Power management system and energy management system for ship

### 2.1 Conventional Power Management System

In the case of load sharing in the conventional electric power system of a ship, the output of the power generation load operated according to the time-varying load condition, as shown in Equation (1). It is equal to the value obtained by dividing the total power load required by the number of operating power sources. It is operated in parallel or in a programmed setting as some generators operate in the optimal operation mode and other power sources are allowed to handle the rest of the demanding load.

 ${P}_{load}={P}_{G1}+{P}_{G2}+{P}_{G3}$ (1)

In this case, the efficiency of the entire power generation source decreases during equivalent parallel operation, which leads to low system efficiency; this causes an increase in carbon dioxide emissions due to an increase in fuel consumption [4]. Even in the case of some optimal operation mode, it is difficult to evaluate whether the efficiency of the entire system has been improved to a satisfactory level because the efficiency of the low-load power generation source operated asymmetrically is low. When the battery SOC is not considered as reserve power, system maintains the SOC level of the battery properly, but when the battery SOC is considered, the battery SOC will be completely consumed after a specific operation time. The load balancing of the system with the battery is represented in Equation (2)

 ${P}_{load}={P}_{G1}+{P}_{G2}+{P}_{G3}+{P}_{BAT}$ (2)

### 2.2 Control Strategy of Energy Management System not Considering Battery SOC

As shown in Table 1, if the battery SOC is not considered a reserve power source, it operates in five operating modes. The generator's operation command is designed to be taken as G_STOP, G_OPT, α = PLOAD + PG_min is set in consideration of the minimum generator operation output value in the design standard of the load standard to designate the initial battery charge and the charging section of the battery according to the section without considering the battery SOC as a reserve power. The output command rule for each source was created.

Rule based strategy not considering battery SOC (α = PLOAD + PG_min, PG_min = 20%, G_OPT = 75% )

### 2.3 Control Strategy of Energy Management System Considering Battery SOC

When the battery is considered as the reserve power of the load, the overall generator output command value is reduced because the system power is compensated by the battery. As shown in Table 2, unlike Table 1, by setting β = PLOADPBATt, a different operation mode is obtained according to the load condition category. The generator operation status is set to G_STOP, G_VAR, G_OPT.

Rule based strategy considering battery SOC (β = PLOAD − PBATT)

## 3. Design of energy management system using neural network

### 3.1 Specification and test environment

To design the controller for EMS according to the designed rule, the system capacity (system load, generator engine, and battery) was selected as shown in Table 3, and the NCR output of the generator engine was selected as 2,000 kW considering the engine margin. All the design specifications were calculated to control the generator engine’s output, as shown in Table 4.

System specification

Design specification of generator engine

### 3.2 Training data for neural network

To test the responsiveness and stability of the generator output command controller of the system according to the time-varying load and SOC variation, the environment shown in Figures 1 and 2 was selected as the final experimental environment.

Specific SOC profile

Table 5 shows the neural network training values that do not consider the battery SOC as a reserve power based on the rule-based strategy for the generator output command according to the ship’s load and the battery SOC value in consideration of the simulation environment. Table 6 shows the neural network learning values considering the SOC value.

Neural network training value not considering battery SOC

Neural network training value considering battery SOC

The gray section indicates the operation where the battery is no longer charged from the system when the battery SOC value reaches 90% or more, and PG_min is set to 0 at α = PLOAD + PG_min and trained.

### 3.3 Neural network application

The designed neural network controller had a single-layer neural network; the hidden layers were composed of 30 layers; and a neural network controller using Bayesian regularization was applied. As shown in Figures 3 and 4, the error in the output value of the learned neural network controller can be compensated by the battery system, so it does not affect the output or operation performance.

Result of errors in neural network training not considering battery SOC

Result of errors in neural network training considering battery SOC

### 3.4 Training result by neural network controller

In addition, to verify the control stability of the designed neural network, the load and battery SOC values as shown in Figures 5 and 6 that were used for learning were applied to verify that the generator output command value matches the learned value and to confirm that there are no irregular or uncontrollable areas in the control result. Figures 7 and 8 depicts the results, which are stable.

Specific SOC profile

Result of neural network controller output tested in step load and step battery SOC not considering battery SOC

Result of neural network controller output tested in step load and step battery SOC considering battery SOC

### 3.5 Simulation

The system shown in Figure 9 was designed through Matlab/Simulink to apply the unspecified ship power load and battery SOC value as input values using the verified neural network controller to make the output command value for the generator output through the controller which prove the stability of the controller.

EMS using neural network

As shown in Figures 10 and 11, if the battery SOC is not considered, the required output of the power source is higher. However, as a result of both designs, the power generation source was designed to have a parallel operation form above a certain load exceeding the optimum operation output value of the power source.

Result of neural network controller output tested in specific load and battery SOC not considering battery SOC

Result of neural network controller output tested in specific load and battery SOC considering battery SOC

As shown in Figures 12 and 13, the command value converted from the generator output command value using the load-RPM table designed in Table 4. It is for a command as the generator angular speed command. It was confirmed that irregular values or oscillation situations did not occur.

Result of RPM command tested in specific load and battery SOC not considering battery SOC

Result of RPM command tested in specific load and battery SOC considering battery SOC

## 4. Conclusion

Through this study, our novel research made the following contributions and findings.

• 1) It was simulated through Matlab/Simulink to design a neural network controller that contributes to the improvement of system efficiency by applying an EMS that manages the battery hybrid system to the power system of a general merchant ship.
• 2) The error between the designed rule-based control value and the neural network controller output was minimized using the Bayesian regularization technique.
• 3) In the controller design, the output of the neural network controller according to the time-varying load is set by setting the optimal operation command of the power source according to the designer's intention and comparing the case where the battery SOC is considered a reserve power source with the case where the battery SOC is not considered under the same load condition.
• 4) It was confirmed that the control characteristics of the command value were stable in designed load and SOC conditions and in a specific test environment too.

However, for future study, the speed range of the engine should be considered for generator control in consideration of the lubricating and mechanical characteristics of the engine.

## Acknowledgments

This work was supported by the Technology development program(P0016077) funded by the Ministry of SMEs and Startups(MSS, Korea).

### Author Contributions

Conceptualization, S. W. Kim; Methodology, S. W. Kim; Software, S. W. Kim and S. C. Kang; Formal Analysis, M. K. Son and S. C. Kang; Investigation, H. M. Jeon and S. C. Kang; Resources, H. M. Jeon; Data Curation, M. K. Son and S. C. Kang; Writing-Original Draft Preparation, S. W. Kim; Writing-Review & Editing, H. M. Jeon; Visualization, M. K. Son and S. C. Kang; Supervision, S. W. Kim and H. M. Jeon; Project Administration, H. M. Jeon; Funding Acquisition, H. M. Jeon.

## References

• T. C. Van, et al., “Global impacts of recent IMO regulations on marine fuel oil refining processes and ship emissions,” Transportation Research Part D: Transport and Environment, vol. 70, pp. 123-134, 2019. [https://doi.org/10.1016/j.trd.2019.04.001]
• E. A. Bouman, et al., “State-of-the-art technologies, measures, and potential for reducing GHG emissions from shipping – A review,” Transportation Research Part D: Transport and Environment, vol. 52, Part A, pp. 408-421, 2017. [https://doi.org/10.1016/j.trd.2017.03.022]
• T. -H. Joung, et al., “The IMO initial strategy for reducing Greenhouse Gas (GHG) emissions, and its follow-up actions towards 2050,” Journal of International Maritime Safety, Environmental Affairs, and Shipping, vol. 4, no. 1, pp. 1-7, 2020. [https://doi.org/10.1080/25725084.2019.1707938]
• M. Jaurola, et al., “Optimising design and power management in energy-efficient marine vessel power systems: a literature review,” Journal of Marine Engineering & Technology, vol. 18, no. 2, pp. 92-101, 2019. [https://doi.org/10.1080/20464177.2018.1505584]
• V. C. Pham, et al., “Effects of various fuels on combustion and emission characteristics of a four-stroke dual-fuel marine engine,” Journal of Marine Science and Engineering, vol. 9, no. 10, p. 1072, 2021. [https://doi.org/10.3390/jmse9101072]
• C. W. Noor, M. M. Noor, and R. Mamat, “Biodiesel as alternative fuel for marine diesel engine applications: A review,” Renewable and sustainable energy reviews, vol. 94, pp. 127-142, 2018. [https://doi.org/10.1016/j.rser.2018.05.031]
• R. Barrera-Cardenas, O. Mo, and G. Guidi. “Optimal sizing of battery energy storage systems for hybrid marine power systems,” 2019 IEEE Electric Ship Technologies Symposium (ESTS), pp. 293-302, 2019. [https://doi.org/10.1109/ESTS.2019.8847932]
• J. Han, J. -F. Charpentier, and T. Tang, “An energy management system of a fuel cell/battery hybrid boat,” Energies, vol. 7, no. 5, pp. 2799-2820, 2014. [https://doi.org/10.3390/en7052799]
• S. Kim and J. Kim, “Optimal energy control of battery hybrid system for marine vessels by applying neural network based on equivalent consumption minimization strategy,” Journal of Marine Science and Engineering, vol. 9, no. 11, p. 1228, 2021. [https://doi.org/10.3390/jmse9111228]

### Figure 2:

Specific SOC profile

### Figure 3 :

Result of errors in neural network training not considering battery SOC

### Figure 4 :

Result of errors in neural network training considering battery SOC

### Figure 6 :

Specific SOC profile

### Figure 7 :

Result of neural network controller output tested in step load and step battery SOC not considering battery SOC

### Figure 8 :

Result of neural network controller output tested in step load and step battery SOC considering battery SOC

### Figure 9 :

EMS using neural network

### Figure 10 :

Result of neural network controller output tested in specific load and battery SOC not considering battery SOC

### Figure 11 :

Result of neural network controller output tested in specific load and battery SOC considering battery SOC

### Figure 12 :

Result of RPM command tested in specific load and battery SOC not considering battery SOC

### Figure 13 :

Result of RPM command tested in specific load and battery SOC considering battery SOC

### Table 1 :

Rule based strategy not considering battery SOC (α = PLOAD + PG_min, PG_min = 20%, G_OPT = 75% )

Power Mode PG1 PG2 PG3
α ≤ G1_MAX 1 G_VAR = α G_STOP G_STOP
G1_MAX < α <
G1_OPT + G2_OPT
2-1 G_OPT G_VAR = α −PG1 G_STOP
G1_OPT + G2_OPT ≤ α
< G1_MAX + G2_MAX
2-2 G_STOP
G1_MAX + G2_MAX ≤ α
< G1_OPT + G2_OPT + G3_OPT
3-1 G_OPT G_OPT G_VAR= α −PG1PG2
G1_OPT + G2_OPT + G3_OPT ≤ α
< G1_MAX + G2_MAX + G3_MAX
3-2

### Table 2 :

Rule based strategy considering battery SOC (β = PLOAD − PBATT)

Power Mode PG1 PG2 PG3
β ≤ 0 0’ G_STOP G_STOP G_STOP
0 < β ≤ G1_MAX 1’ G_VAR = β G_STOP G_STOP
G1_MAX ≤ β < G1_OPT + G2_OPT 2’-1 G_OPT G_VAR = β −PG1 G_STOP
G1_OPT + G2_OPT ≤ β
< G1_MAX + G2_MAX
2’-2 G_STOP
G1_MAX + G2_MAX ≤ β
< G1_OPT + G2_OPT + G3_OPT
3’-1 G_OPT G_OPT G_VAR = β −PG1PG2
G1_OPT + G2_OPT + G3_OPT ≤ β
< G1_MAX + G2_MAX + G3_MAX
3’-2

### Table 3 :

System specification

System specification Capacity[kW]
Generator engine 3,000
Battery 1,000

### Table 4 :

Design specification of generator engine

[kW]
TORQUE
[N ∙ m]
RPM

VOLTAGE
[V]
AMP
[A]
ω
30 740.3 387 690 25.1 40.5
60 1174.1 488 690 50.2 51.1
90 1537.5 559 690 75.3 58.5
120 1863.3 615 690 100.4 64.4
150 2163.7 662 690 125.5 69.3
300 3430.9 835 690 251 87.4
600 5441.2 1053 690 502.1 110.3
900 7132.3 1205 690 753.1 126.2
1200 8641.9 1326 690 1004.1 138.9
1500 10023.8 1429 690 1255.1 149.6
1800 11308.4 1520 690 1506.2 159.2
2100 12549.1 1598 690 1757.2 167.3
2400 13715.3 1671 690 2008.2 175.0
2700 14834.9 1738 690 2259.3 182.0
3000 15915.5 1800 690 2510.3 188.5

### Table 5 :

Neural network training value not considering battery SOC

[kW]
7000 6300 5600 4900 4200 3500 2800 2100 1400 700 350 0
■ = Disable for battery charging
■ = 1 set generator engine running with 20% additional output for battery charging
■ = 2 sets generator engine running with 20% additional output for battery charging
■ = 3 sets generator engine running with 20% additional output for battery charging
■ = 3 sets generator engine running with under 20% additional output for battery charging
□ = Disable
SOC
[kW]
1000
100,
100,
100
90,
90,
90
78.4,
78.4,
78.4
75,
75,
50
82.5,
82.5,
0
75,
55,
0
95,
0,
0
60,
0,
0
25,
0,
0
0,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
900
100,
100,
100
90,
90,
90
78.4,
78.4,
78.4
75,
75,
50
82.5,
82.5,
0
75,
55,
0
95,
0,
0
60,
0,
0
25,
0,
0
0,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
800
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
700
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
600
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
500
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
400
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
300
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
200
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
100
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0
SOC
[kW]
0
100,
100,
100
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
37.5,
0,
0
20,
0,
0

### Table 6 :

Neural network training value considering battery SOC

[kW]
7000 6300 5600 4900 4200 3500 2800 2100 1400 700 350 0
■ = Battery only,
■ = 1 set generator engine running,
■ = 2 set generator engine running
■ = 3 set generator engine running
■ = Maximum operation
□ = Disable
SOC
[kW]
1000
100,
100,
100
90,
90,
90
78.4,
78.4,
78.4
75,
75,
50
82.5,
82.5,
0
75,
55,
0
95,
0,
0
60,
0,
0
25,
0,
0
0,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
900
100,
100,
100
90,
90,
90
78.4,
78.4,
78.4
75,
75,
50
82.5,
82.5,
0
75,
55,
0
95,
0,
0
60,
0,
0
25,
0,
0
0,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
800
100,
100,
100
91.7,
91.7,
91.7
80,
80,
80
75,
75,
55
85,
85,
0
75,
60,
0
75,
25,
0
65,
0,
0
30,
0,
0
0,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
700
100,
100,
100
93.4,
93.4,
93.4
81.7,
81.7,
81.7
75,
75,
60
87.5,
87.5,
0
75,
65,
0
75,
30,
0
70,
0,
0
35,
0,
0
0,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
600
100,
100,
100
95,
95,
95
83.4,
83.4,
83.4
75,
75,
65
90,
90,
0
75,
70,
0
75,
35,
0
75,
0,
0
40,
0,
0
5,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
500
100,
100,
100
96.7,
96.7,
96.7
85,
85,
85
75,
75,
70
92.5,
92.5,
0
75,
75,
0
75,
40,
0
80,
0,
0
45,
0,
0
10,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
400
100,
100,
100
98.4,
98.4,
98.4
86.7,
86.7,
86.7
75,
75,
75
95,
95,
0
77.5,
77.5,
0
75,
45,
0
85,
0,
0
50,
0,
0
15,
0,
0
0,
0,
0
0,
0,
0
SOC
[kW]
300
100,
100,
100
100,
100,
100
88.4,
88.4,
88.4
76.7,
76.7,
76.7
97.5,
97.5,
0
80,
80,
0
75,
50,
0
90,
0,
0
55,
0,
0
20,
0,
0
2.5,
0,
0
0,
0,
0
SOC
[kW]
200
100,
100,
100
100,
100,
100
93.4,
93.4,
93.4
81.7,
81.7,
81.7
75,
75,
60
87.5,
87.5,
0
75,
65,
0
75,
30,
0
70,
0,
0
35,
0,
0
17.5,
0,
0
0,
0,
0
SOC
[kW]
100
100,
100,
100
100,
100,
100
93.4,
93.4,
93.4
81.7,
81.7,
81.7
75,
75,
60
87.5,
87.5,
0
75,
65,
0
75,
30,
0
70,
0,
0
35,
0,
0
17.5,
0,
0
0,
0,
0
SOC
[kW]
0
100,
100,
100
100,
100,
100
93.4,
93.4,
93.4
81.7,
81.7,
81.7
75,
75,
60
87.5,
87.5,
0
75,
65,
0
75,
30,
0
70,
0,
0
35,
0,
0
17.5,
0,
0
0,
0,
0