(19)
(11)EP 3 330 171 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
09.09.2020 Bulletin 2020/37

(21)Application number: 17204814.2

(22)Date of filing:  30.11.2017
(51)International Patent Classification (IPC): 
G01C 21/20(2006.01)
G01W 1/10(2006.01)
B63B 49/00(2006.01)
G01W 1/00(2006.01)

(54)

APPARATUS FOR PREDICTING A POWER CONSUMPTION OF A MARITIME VESSEL

VORRICHTUNG ZUR VORHERSAGE DES STROMVERBRAUCHS EINES SEESCHIFFES

APPAREIL PERMETTANT DE PRÉDIRE UNE CONSOMMATION D'ÉNERGIE D'UN VAISSEAU MARITIME


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(30)Priority: 30.11.2016 EP 16201572

(43)Date of publication of application:
06.06.2018 Bulletin 2018/23

(73)Proprietor: Offshore Navigation Limited
2640 George Hill (AI)

(72)Inventors:
  • DOKKEN, Sverre
    3071 Limassol (CY)
  • MAO, Wengang
    412 96 Gothenburg (SE)
  • ANAGNOSTOPOULOS, Alexis
    Macclesfield, Cheshire SK10 2AB (GB)

(74)Representative: Patentship Patentanwaltsgesellschaft mbH 
Elsenheimerstraße 65
80687 München
80687 München (DE)


(56)References cited: : 
US-A1- 2015 149 074
US-A1- 2016 251 064
  
  • "Training, test, and validation sets", Wikipedia , 29 November 2016 (2016-11-29), XP002782141, Retrieved from the Internet: URL:https://en.wikipedia.org/w/index.php?t itle=Training,_test,_and_validation_sets&o ldid=752127502 [retrieved on 2018-06-18]
  • "Regression analysis", Wikipedia , 28 November 2016 (2016-11-28), XP002782142, Retrieved from the Internet: URL:https://en.wikipedia.org/w/index.php?t itle=Regression_analysis&oldid=751970896 [retrieved on 2018-06-18]
  • "Artificial neural network", Wikipedia , 28 November 2016 (2016-11-28), Retrieved from the Internet: URL:https://en.wikipedia.org/w/index.php?t itle=Artificial_neural_network&oldid=75190 3155#Applications [retrieved on 2019-01-15]
  
Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


Description


[0001] The invention relates to an apparatus for predicting a power consumption of a maritime vessel, a method for operating an apparatus for predicting a power consumption of a maritime vessel and a computer program for performing the method.

[0002] In maritime environments, route and performance optimization of a maritime vessel may be performed in order to increase the efficiency and reliability of shipping and reduce emissions. For this purpose, the prediction of a power consumption of the maritime vessel at a given speed is of particular interest. Usually, theoretical models are used to predict the power consumption of a maritime vessel. However, these theoretical models may have large uncertainties and prediction errors. An accurate prediction of the power consumption is, however, desirable for providing high-quality ship-related services, e.g. accurate sail planning or weather routing systems. This may also be desirable for fuel and/or emission and/or fatigue and/or time optimization.

[0003] Document US 2016/0251064 A1 discloses a ship propulsion performance predicting apparatus including a theoretical propulsion performance computing unit computing theoretical propulsion performance for a desired navigation condition using a physical model of a propulsion system of a target ship, and a corrector correcting the theoretical propulsion performance using the smooth water correction term and the disturbance correction term stored in a correction term database. The smooth water correction term and the disturbance correction term stored in the correction term database are derived by a correction term deriver from operation experience data.

[0004] Document US 2015/0149074 A1 discloses a computer-implemented method for determination of a route for a ship by defining one or more performance variables of the ship and dynamic input parameters that affect to the performance variable. Information of ship operation and measurement results from sensors during the operation of the ship are obtained and a set of dynamic input parameters is produced. A model for simulating the performance of a ship is created by defining one or more relationships between the performance variables and the dynamic input parameters.

[0005] It is an object of the invention to provide an improved concept for predicting a power consumption of a maritime vessel.

[0006] It is an object of the invention to provide an improved concept for predicting a power consumption of a maritime vessel.

[0007] This object is achieved by the features of the independent claims. Further embodiments are described in the dependent claims, in the description as well as in the figures.

[0008] The invention is based on the finding that a predetermined power consumption model, e.g. a theoretical model, and a power consumption correction model representing a modeling error associated with the predetermined power consumption model, can be evaluated in parallel in order to improve the prediction of the power consumption of the maritime vessel. The power consumption correction model is a neural network model, and may be trained in order to minimize an error of overall power consumption prediction. The training may be performed upon the basis of training sets stored in a database, which may continuously be updated; thus providing an updated model parameter of the power consumption correction model. The maritime vessel may e.g. be a ship or an off-shore platform. Based on the predicted power consumption, a fuel consumption may be predicted.

[0009] According to a first aspect of the invention, the object is solved by an apparatus according to claim 1.

[0010] The apparatus may be allocated on board of the maritime vessel or may be allocated on shore and may be in communication with a computer system on board of the maritime vessel. The input interface and/or the output interface may be configured to transmit and/or receive radio signals or signals by cable. The processor may be a CPU and/or a GPU of a computer system, a microcontroller unit, or the like. The combination of the power consumption estimate with the power consumption correction may e.g. be performed by adding the power consumption estimate and the power consumption correction, or by subtracting the power consumption correction from the power consumption estimate.

[0011] The predetermined power consumption model may be a predetermined theoretical model. The power consumption correction model may be based on a machine learning approach, e.g. using supervised learning or un-supervised learning. The power consumption correction model may be updated during operation of the apparatus.

[0012] In an embodiment, the predetermined power consumption model comprises a calm water resistance model of the maritime vessel and an added resistance model of the maritime vessel.

[0013] The calm water resistance model can reflect the resistance of the maritime vessel in calm water, i.e. without the effects of waves and winds and other geophysical parameters affecting the resistance. The calm water resistance model can e.g. be based on the Holtrop-Mennen method or the like. The added resistance model can reflect the resistance of the maritime vessel due to waves and winds and other geophysical parameters affecting the resistance. The added resistance model can e.g. be based on potential flow panel methods or the like. The added resistance model can further reflect changes of the resistance over time, e.g. due to a varying friction.

[0014] In an embodiment, the calm water resistance model is independent of the physical parameter and the added resistance model is dependent of the physical parameter.

[0015] In an embodiment, the physical parameter, i.e., the wetted surface of the maritime vessel, may further be combined with one of the following physical parameters: a rudder angle of the maritime vessel, a draft of the maritime vessel, a heading of the maritime vessel, a wind speed relative to the maritime vessel, a wind direction relative to the maritime vessel, a current speed relative to the maritime vessel, a current direction relative to the maritime vessel, a wave height around the maritime vessel, a wave period around the maritime vessel, a wave steepness around the maritime vessel, a wave spectrum around the maritime vessel, a water density around the maritime vessel, a water salinity around the maritime vessel, a water temperature around the maritime vessel, a geometric parameter of a hull of the maritime vessel, a sea ice thickness around the maritime vessel, a sea ice concentration around the maritime vessel, a fouling intensity of a hull of the maritime vessel. Similar physical parameters may also be applied.

[0016] These physical parameters, for example, influence the power consumption of the maritime vessel. The maritime vessel and/or the apparatus may comprise a respective sensor for detecting any one of the physical parameters.

[0017] Each predetermined power consumption model of the plurality of predetermined power consumption models is associated with a respective category of a maritime vessel. The respective category is indicative of a shape of a respective maritime vessel. In particular, different categories, e.g. a category for small-sized maritime vessels, a category for medium-sized maritime vessels, and a category for large-sized maritime vessels, may be used. As an example, large container ships and large cruise-liner ships may be regarded as large-sized maritime vessels. The number of different categories may be large and may be increased to an arbitrary, e.g. unlimited, number.

[0018] According to a second aspect of the invention, the object is solved by the use of the apparatus within a sail plan system for planning a sailing route of a maritime vessel. The sail plan system may comprise the apparatus. In particular, the functionality of the respective components of the apparatus may be implemented within the sail plan system, e.g. using software components. The sail plan system may perform an optimization.

[0019] The apparatus may analogously be used within a weather routing system for routing within the maritime environment. The weather routing system may comprise the apparatus. In particular, the functionality of the respective components of the apparatus may be implemented within the weather routing system, e.g. using software components. The weather routing system may perform an optimization.

[0020] According to a third aspect of the invention, the object is solved by a method according to claim 5.

[0021] The method may be performed by the apparat for predicting the power consumption of the maritime vessel. Further features of the method directly result from the functionality of the apparatus for predicting the power consumption of the maritime vessel.

[0022] According to a sixth aspect of the invention, the object is solved by a computer program comprising a program code for performing the method.

[0023] The invention can be implemented in hardware and software.

[0024] The inventive concept is now described, by way of example, with reference to the accompanying drawings, in which:
Fig. 1
shows a schematic diagram of an arrangement with an apparatus according to an embodiment of the invention;
Fig. 2
shows a schematic diagram of another apparatus not forming an embodiment of the invention;
Fig. 3
shows a schematic diagram of an arrangement with a ship and an apparatus according to an embodiment of the invention;
Fig. 4
shows a schematic flow diagram for a method according to an embodiment of the invention;
Fig. 5
shows another schematic flow diagram of a method not forming an embodiment of the invention;
Fig. 6
shows the total required power for calm water and the single components;
Fig. 7
shows a wind power prediction for relative wind speed from 0°, 45° and 90°;
Fig. 8
shows a wave resistance due to wave diffraction at 10 m/s;
Fig. 9
shows a wave resistance due to motions at 10 m/s;
Fig. 10
shows a wave added resistance at 10 m/s;
Fig. 11
shows a non dimensional added resistance with Alexandersson's method;
Fig. 12
shows a comparison of non dimensional added resistance with Alexandersson's method and the 'Improved formula for estimating the added resistance';
Fig. 13
shows the JONSWAP wave spectrum;
Fig. 14
shows a power prediction for the three different computed terms: the calm water power, wind power and wave power; and
Fig. 15
shows a flowchart of the Machine Learning steps.


[0025] Fig. 1 shows an arrangement with an apparatus 100 for predicting a power consumption of a maritime vessel. The apparatus 100 comprises an input interface 101, a processor 102, and output interface 103, and a memory 104.

[0026] The input interface 101 is configured to receive a measured physical parameter. The interface 101 is further configured to receive a measured speed of the maritime vessel. The processor 102 is a CPU or GPU of a computer system. In other embodiments, the processor 102 is a microcontroller unit. The output interface 103 is configured to provide the predicted power consumption of the maritime vessel e.g. to a computer system.

[0027] The memory 104 may e.g. be realized as a solid-state disk, a hard-disk drive or a cloud storage or processing system. The memory 104 is in functional communication with the input interface 101, the processor 102 and the output interface 103.

[0028] Fig. 2 shows another apparatus 200 not forming an embodiment of the invention. The apparatus 200 comprises a processor 201, a database 202 comprising a plurality of training sets 203, an input interface 204, and an output interface 205.

[0029] The processor 201 is a CPU of a computer system. The input interface 204 is in functional communication with the database 202 and the processor 201. The input interface 204 is configured to receive a training set 203, and store the received training set 203 in the database 202. The output interface 205 may be connected to a display or a computer system and is configured to provide a model parameter.

[0030] Fig. 3 shows an arrangement with a maritime vessel 300, in particular a ship. The maritime vessel 300 is moving within a maritime environment, e.g. on an ocean 301. The maritime vessel 300 comprises sensors 302. The sensors 302 are sensors to measure physical parameters of an environment of the maritime vessel 300. In particular, the sensors 302 are configured to measure or receive, e.g. from an external device such as an AIS (Automatic Identification System), at least one of the following physical parameters: a rudder angle of the maritime vessel 300, a draft of the maritime vessel 300, a heading of the maritime vessel 300, a wind speed relative to the maritime vessel 300, a wind direction relative to the maritime vessel 300, a current speed relative to the maritime vessel 300, a current direction relative to the maritime vessel 300, a wave height around the maritime vessel 300, a wave period around the maritime vessel 300, a wave steepness around the maritime vessel 300, a wave spectrum around the maritime vessel 300, a water density around the maritime vessel 300, a water salinity around the maritime vessel 300, a water temperature around the maritime vessel 300, a wetted surface of the maritime vessel 300, a sea ice thickness around the maritime vessel 300, a sea ice concentration around the maritime vessel 300, a fouling intensity of a hull of the maritime vessel 300. The sensors 302 may further measure a speed of the maritime vessel 300, in particular a speed over water of the maritime vessel 300. The sensors 302 may further measure an (actual) power consumption of the maritime vessel 300 to be used for training purposes, e.g. within training sets. A geometric parameter of a hull of the maritime vessel 300 as a physical parameter may be predetermined or may be provided to the apparatus 100 or the apparatus 200.

[0031] The maritime vessel 300 may further comprise the apparatus 100. The apparatus 100 is in functional communication with the sensors 302 and/or external devices to receive identical or similar parameters. The input interface 101 of the apparatus 100 is configured to receive the physical parameters and/or the speed of the maritime vessel measured from the sensors 302. These parameters may also be received from an external device, e.g. an AIS system.

[0032] The maritime vessel 300 may further comprise the apparatus 200. The apparatus 200 is in functional communication with the sensors 302. The input interface 204 is configured to receive training sets comprising a speed of the maritime vessel 300, a physical parameter of a maritime environment of the maritime vessel 300, and an (actual) power consumption of the maritime vessel 300, and to store the received training set in the database 202.

[0033] In another embodiment, the apparatus 100 and/or the apparatus 200 are arranged on-shore, and communicate with the respective sensors 302 of the maritime vessel 300 over a communication network, e.g. the Internet or a dedicated communication network.

[0034] Consequently, predicting a power consumption and determining a model parameter of a power consumption correction model can be realized. The model parameter can thus be updated and can be used for predicting the power consumption of the maritime vessel 300.

[0035] Fig. 4 shows a schematic flow diagram 400 for a method according to an embodiment of the invention. The described method may be performed by the apparatus 100 on board of the maritime vessel 300.

[0036] In step 401 the speed of the maritime vessel 300 is received by the input interface 101. The speed of the maritime vessel 300 is measured by the sensors 302 of the maritime vessel 300 or received from an external device. The sensors 302 send the measured speed of the maritime vessel directly towards the input interface 101. In another embodiment, the sensors 302 store the measured speed of the maritime vessel 300 in the memory 104 or another memory. The input interface 101 of the apparatus 100 can receive the speed value of the maritime vessel 300 by reading the stored value from the memory 104.

[0037] In step 402, the input interface 101 receives the physical parameter of the maritime environment of the maritime vessel 300. The physical parameter is measured by the sensors 302 of the maritime vessel 300 received from an external device, and is sent directly to the input interface 101 in the above-described way. In another embodiment, also as described above, the sensors 302 store the physical parameter in the memory 104 after the measurement or reception.

[0038] In step 403, the processor 102 determines a power consumption estimate. The power consumption estimate is based on the received speed of the maritime vessel 300 and the physical parameter of the maritime environment of the maritime vessel 300. For determining the power consumption estimate, the processor 102 uses a predetermined power consumption model. The predetermined power consumption model may be predefined. The predetermined power consumption model may also be updated according to a supervised or unsupervised machine-learning approach.

[0039] In step 404, the processor determines a power consumption correction. The power consumption correction is also based on the speed and the physical parameter. The processor uses for determining the power consumption correction a power consumption correction model. The power consumption correction model represents a modeling error of the predetermined power consumption model.

[0040] In step 405, the processor combines the determined power consumption estimate and the power consumption correction. In the described embodiment, the processor adds the power consumption estimate and the power consumption correction. The processor may alternatively subtract the power consumption correction from the power consumption estimate. In other embodiments other mathematical methods are used to combine the determined power consumption estimate and the power consumption correction.

[0041] In step 406, the output interface provides the predicted power consumption of the maritime vessel 300, e.g. to the memory 104. In another embodiment, the predicted power consumption is output additionally or alternatively to a display or another computer system.

[0042] Fig. 5 shows a schematic flow diagram 500 of a method not forming an embodiment of the invention. In this example, the method is performed by the apparatus 200. The apparatus 200 may be allocated on board of a vessel, in particular in the maritime vessel 300, or on shore.

[0043] The database 202 comprises a plurality of training sets 203. Each training set 203 comprises the speed of the maritime vessel 300 and a physical parameter of the maritime environment of the maritime vessel 300. For example, each training set 203 comprises a wind speed relative to the maritime vessel 200 as a physical parameter. In another embodiment the physical parameter may be a vector comprising several single physical parameters. In another embodiment, the training sets 203 comprise a plurality of physical parameters.

[0044] The processor 201 is configured to read the training sets 203. The processor 201 determines in step 501 a power consumption estimate based on the respective speed and the physical parameter. For determining the power consumption estimate, the process 201 uses a predetermined power consumption model. The predetermined power consumption model in this embodiment correlates to the predetermined power consumption model described according to the method described with reference to Fig. 4. Step 501 is repeated for each training set 203.

[0045] In step 502, the processor determines, for each training set 203, a difference between the determined power consumption estimate and the respective power consumption. The respective power consumption may have been measured during a previous voyage of the maritime vessel as a historical value and may serve as a reference to be compared with the determined power consumption estimate.

[0046] In step 503, the processor appends to each training set 203 the respective power consumption difference from the step 502. Therefore, the processor generates a respective appended training set.

[0047] In step 504, the processor trains the power consumption correction model on the basis of the plurality of appended training sets. For this purpose, a neural network is used.

[0048] In the following, further exemplary embodiments of the apparatus 100 and/or the apparatus 200 are described. In particular, suitable predetermined power consumption models are described in more detail.

[0049] Power prediction can be performed with empirical and semi-empirical methods according to the ITTC (International Towing Tank Conference) procedure aimed at meeting the IMO (International Maritime Organization) EEDI (Energy Efficiency Design Index) requirements. The total resistance here can be calculated as the sum of three terms: calm water resistance, wind added resistance and wave added resistance. Smaller terms that contribute to the resistance such as drifting, steering for course keeping, shallow water may not be considered as it is considered that their contribution may be very small and may be neglected. Once the total resistance is calculated, the required power is found and plotted to compare it to the actual propulsion taken from the data.

Resistance & Power prediction:



[0050] The resistance of a ship at a given speed is the force needed to tow the ship at that speed through water. The total required resistance is the sum of more components, caused by different factors. Here the total resistance is broken down into three components:
  • Calm water resistance
  • Wind added resistance
  • Wave added resistance


[0051] The methods used to compute these resistances are described in the following.

Resistance & Power prediction for calm water:



[0052] The calm water resistance is usually considered to be the sum of four components:
  • Frictional resistance, caused by the motion of the hull through a viscous fluid.
  • Wave making resistance, caused by the energy that the hull transfers to the wave system created on the surface of the water.
  • Eddy resistance, caused by the energy taken away by eddies shed from the hull and appendages.
  • Air resistance by the above-water part of the main hull and the superstructures due to the motion of the ship through the air.


[0053] The wave making resistance and the eddy resistance may often be considered together as residual resistance. As an example, the resistance and the required power for calm water are calculated following a specific method. The method can e.g. be based on the recommended procedures of ITTC '57, and the required power is found by regression analysis of model test results from different combinations of speed, draught and trim calibrated by full scale speed trials at delivery.

[0054] The total required power in calm water is calculated as shown in eq. 2.1:

where:
  • PF is the power required for the frictional resistance
  • PR is the power required for the residual resistance

Frictional Resistance & Power:



[0055] The frictional resistance RF is the sum of the tangential stresses along the wetted surface in the direction of the motion. According to the ITTC '57 formula it is calculated as:

where:

ρ = 1025 kg/m3 is the sea water density

S = wetted surface of the hull

V = speed of the ship

and the frictional coefficient CF is defined, according to ITTC '57 as:

where:


  is the Reynolds number
ks = 150 µm  is the hull surface roughness

[0056] The first part of eq. 2.3 accounts for the frictional coefficient, while the second part is an increment coefficient, also known as CA which accounts for the roughness of the hull computed according to Townsin's equation as reported by Watson.

[0057] Once, the resistance is known, the required power can be computed as shown in eq. 2.4:

where V is the speed, CP is the correction factor for delivered power and ηtot the total efficiency. The value of CP is 1.058 and the total efficiency is equal to 0.732. Both values were provided. The results are shown in Fig. 6.

Residual Power:



[0058] The residual power is founded on regression analysis of model test results from different combinations of speed, draught and trim calibrated by full scale speed trials at delivery. Fig. 6 shows the power prediction for calm water. The curves of PD and PR have been calculated through the regression analysis, while PF is calculated as shown in Eq. 2.1.

[0059] In Table 2.1 a range of speeds is displayed to show the values of the power related to the more common operational conditions of a ship. The Froude number is calculated as:

where:

V: speed of the ship in m/s

g: gravitational constant

Lpp: Length between perpendiculars

Table 2.1: Required power for calm water.
SpeedFNPCW
[kn] [-] [MW]
16.0 0.177 7.323
17.0 0.189 8.457
18.0 0.200 9.829
19.0 0.211 11.523
20.0 0.222 13.631
21.0 0.233 16.241

Resistance increase due to the effect of the wind:



[0060] A ship, sailing, experiences a resistance when moving through the air depending on the ship's speed and the area and the shape of the superstructures. When wind is blowing the resistance also depends on the speed of the wind and its relative direction. In addition, the wind raises waves which are cause for further increase in resistance. The effects of waves are discussed later. Here, some terms regarding the air and wind resistance are explained to clarify their usage. The term "true" wind is intended to be the wind that exists independently from the ship's motion, and therefore, true wind equal to zero is still air. The "relative" wind is the vectorial summation of the speed and the direction of the ship with the speed and direction of the true wind.

[0061] In case that wind resistance tests have been performed in a wind tunnel, the wind resistance coefficients derived by these tests shall be used to calculate the wind resistance of the ship. In this case the wind resistance coefficients are based on the data derived from the performed model tests from the wind tunnel. The resistance increase due to relative wind is calculated according to ITTC by:

where:

AXV: Area of maximum transverse section exposed to the wind

CX: wind resistance coefficient

VWR: relative wind speed

ρA : mass density of air

ψWR relative wind direction, with heading wind at 0°



[0062] As previously done for the calm water, the power required is calculated by multiplying the resistance RAA by the relative speed VWR and the efficiencies as shown in eq. 2.7.



[0063] The resulting required power for the wind for different wind directions is plotted in Fig. 7. For the plotting, a relative wind speed in the range from 0 to 22 knots is shown, and the three wind directions are 0°, 45° and 90°. For 0° it is considered head wind. From Fig. 7 it can be seen that the highest required power is for 0°, i.e. head wind, and it decreases as the angle becomes higher. The pattern of the curves is cubic, since the power has the term of the speed V3. The wind resistance coefficients have been calculated from the wind tunnel tests.

Resistance increase due to the effect of waves:



[0064] The wave added resistance is defined as the mean force caused by the encountered waves. In this section two methods for the evaluation of the added resistance due to waves are presented. The reason why two different methods are chosen is to see how well they perform in the end in the prediction of the power. The first method is the "Improved Formula for estimating Added resistance of Ships in Engineering applications" by Liu et al. and the second is a simplified method developed by Alexandersson.

Improved Formula for Estimating Added Resistance of Ships in Engineering Applications:



[0065] The method presented here to calculate added resistance is a further refined formula based on the semi-empirical formula introduced by the authors of "Fast approach to the estimation of the added resistance of ships in waves" by Liu and Papanikolaou. The formula is based on best fitting of available experimental data for different hull forms and is made to be easy to be used for engineering applications as it uses only the speed, the main characteristics of the ship and the wave environment as input. This paper lies within the framework of the IMO Marine Environment Protection Committee (MEPC).232(65) EEDI guidelines for estimating the minimum powering of ships in adverse weather conditions. The formula is suitable for both low and moderate speeds and thus it can be used in a procedure for determining the minimum power required in adverse conditions [Liu et al.]. One of the limitations of this method is that it is to calculate the added resistance only for head waves.

Formula:



[0066] The predicted added resistance of a ship in head waves at any wave length is calculated as:

where RAWR is the added resistance in regular waves due to diffraction effect, caused from short waves, and RAWM is the added resistance in regular waves due to motion effect, caused by long waves.

Added resistance in regular waves due to diffraction:



[0067] RAWR is calculated as shown in eq. 2.9:

where:



[0068] Eq. 2.9 is a simplified expression using typical design data for a ship, based on the previous study of Liu et al. LE has been redefined as the distance from the forward perpendicular (F.P.) to the point where the ship breadth (beam) reaches 99%.

[0069] Fig. 8 shows the pattern of the wave added resistance due to the diffraction of the waves at a speed of the ship of 10 m/s = 20 knots. It can be seen that the shorter the waves, where the ratio λ/LPP is lower in the left part of the plot, the higher the resistance. On the right part, where the λ/LPP ratio is higher the resistance decreases exponentially. This is because for longer waves the motions of the ship are the main component to the wave added resistance, which is presented in the next paragraph.

Added resistance in regular waves due to motions:



[0070] For long waves, the formula of Jinkine and Ferdinande can be modified to eq. 2.10:

where a1 and a2 defined as shown in eq. 2.11 and 2.12 are the peak value factor and forward speed factor respectively.





[0071] The natural frequency ω̅ is defined as in eq. 2.13

where kyy is the longitudinal mass radius of gyration.

[0072] The parameters b1 and d1 are defined as shown in eq. 2.14, 2.15, 2.16 and 2.17 for CB < 0.75



for CB ≥ 0.75





[0073] By summing RAWR to RAWM the wave added resistance may be computed.

A simplified method to calculate added resistance based on Gerritsma and Beukelman's method:



[0074] Alexandersson proposed a simplified method to calculate the wave added resistance based on the method proposed by Gerritsma and Beukelman's. Gerritsma and Beukelman's method considers the radiated energy to calculate the added resistance by utilizing strip theory and is calculated with eq. 2.18:

where, k is the wave number, β is heading angle, ωe is encountered frequency, L is the length of the waterline of the ship, |VZb| is the amplitude of the velocity of water relative to the strip, b' is the sectional damping coefficient for speed and xb is the x coordinate on the ship.

[0075] Gerritsma and Beukelman's method even though is one of the most widely used methods for calculating the added resistance, as it is regarded an accurate method for different ship types, results quite complex when calculating the ship's motions by strip theory and might result difficult or unnecessary for certain applications.

[0076] To overcome the problem of not having the geometry of the hull and avoiding strip theory calculations Alexandersson proposed a simplified method by using Gerritsma and Beukelman's method. It was applied to a large series of case studies to determine the added resistance, then, by using linear regression, a series of simplified formulas were derived to determine the added resistance. This method is therefore considered semi-empirical. The limitations of this method are the following: the accuracy decreases in bow, beam, following waves, it only considers the added resistance from the ship motions and not considering the diffraction, and the trim of the ship is not included.

[0077] This method is briefly described below along with the equations that were used.

Method:



[0078] Alexandersson's method consists in fitting a normal distribution function to the transfer function to be described. This way, the transfer function can be described with eq. 2.19 by the use of only three parameters:

where:

RAW,p is the peak value

ωp is the peak frequency

c is the width.



[0079] The three parameters are calculated through linear regression and the equations derived by Alexandersson's method can be seen through eq. 2.20 to eq. 2.25:











[0080] The parameters have been made non-dimensional through the relations shown in Table 2.2:
Table 2.2: Parameters of the model non-dimensionalized.
 ParametersNon-dimensional parameters
Length of waterline LWL 1
Prismatic coefficient CP CP
Longitudinal Center of buoyancy LCB[m] LCB,norm = LCB/LWL
Beam B [m] Bnorm = B/LWL
Draught T [m] Tnorm = T/LWL
Radius of pitch gyration ryy [m] kyy = ryy/LWL
Speed V [m/s] Fn


[0081] Alexandersson has calculated all these parameters for 7 ships, and the ones most suitable for this study, are the ones for a Ro-Ro ship.

[0082] Once the added resistance is calculated it is made non-dimensional through eq. 2.25:



[0083] Fig. 11 shows the curve for the added resistance with Alexandersson's method. The curve is commented in the following.

Comparison of the two methods:



[0084] A graphic comparison is made here to see how similar the two methods are with each other. In Fig. 12 the two methods are plotted. It can be seen that both yield very similar results, with the "Improved Formula" by Liu et al. have a slightly higher peak, and coinciding with longer waves than the ship. On the other hand, Alexandersson's method, provides no added resistance for short waves, as it was expected from the problem formulation, and it peaks when the waves are approximately the same length as the ship. For the plotting, a speed of 10 m/s has been used.

Mean Added Resistance in Irregular Waves:



[0085] The mean added resistance in irregular waves is calculated as shown in eq. 2.26:

where:

RAW(ω) the added resistance response function in regular waves

S(ω) is the wave Spectrum

ζa the regular wave amplitude



[0086] For this study the JONSWAP (Joint North Sea Wave Project) wave spectrum has been chosen. The JONSWAP spectrum is valid for not fully developed sea states but it is also used to represent fully developed seas. The spectrum is expressed as:

where:





[0087] The peakedness parameter γ is generally 3.3, and it is also the value used here. The JONSWAP is the same as the Pierson Moskowitz, Bretschneider and ITTC wave spectra, given Hs, Tp, when γ = 1. In Fig. 13 the JONSWAP spectrum is plotted.

[0088] From the mean added resistance the power is calculated in the same way as the calm water and the wind. Thus, the power is expressed as shown in eq. 2.28:



[0089] In Fig. 14 the three different components of the power can be seen. The calm water accounts for the highest contribution to the resistance as expected. The wave added power is computed here only for the improved formula by Liu et al., but later both methods are be computed. For the wave added resistance a significant wave height, Hs of 4 m is chosen, and a Tp peak period of 7 s, just to provide an example.

[0090] To compute the performance of the ship while sailing the power is computed as the sum of the calm water, the wind and the waves power.



[0091] In the following, further exemplary embodiments of the apparatus 100 and/or the apparatus 200 are described. In particular, suitable machine-learning and training approaches are described in more detail.

[0092] An overview of the followed approach in this machine-learning problem is shown in Fig. 15. This kind of method showed is used for supervised learning, which is different than other machine-learning methods.

[0093] Following, each step is broken down and described.

[0094] In Fig. 15 the steps followed in this study are shown in a flowchart. The first step consists in loading the raw data, which may be in the form of .csv (comma separeted values). Once the data is uploaded and interpreted the features and the target for creating the machine-learning models are selected. At this point, the subset of data that is selected is split between training set and test set data, that will serve to train and test the model created. The three selected models are at this point applied on the data and trained. The trained model is subsequently applied on the unseen test data, and the difference between the prediction and the actual data is computed. The aforementioned steps are explained in greater detail in the following sections.

Raw Data & Data Analysis:



[0095] Starting from the engine room, the sensors measure the fuel flow and temperature of the auxiliary engine, the electric power and the main engine fuel temperature and flow. In addition, the shaft power and torque, along with the RPM of the shaft and the electric power of the shaft generator are collected. Furthermore the water temperature is measured, and also the rudder angle and the rate of turn of the rudder. At the aft and the bow of the ship sensors measure the draft. The ship is also equipped with an echo sounder sensors to measure the depth of the water. At the wheelhouse, the GPS tracker is positioned that measures the position of the ship, the speed over ground and the course of the ship. Furthermore, a gyroscope collects data regarding the heading of the ship and the roll and pitch motions. In addition a motion sensor also records the six degrees of freedom (DOF) of the ship. Finally, an anemometer records the wind speed and direction.

Data selection:



[0096] For predicting the performance with machine-learning models, the selection of the data sets may be relevant in order to obtain good results. The first step is to define what is expected to be predicted. The goal here is to predict the propulsive power when the ship is sailing in open sea, so the parts of voyage of the ship when it's maneuvering, accelerating and decelerating will not be considered. To do this, the data is divided by the voyages by port of arrival and port of departure.

Features and Target selection:



[0097] The target is the variable that is going to be predicted with the machine-learning, while for features, is meant the variables that form the data frame excluding the target and contribute to predict the selected target. In the feature selection here, a subset of the features may be chosen to build the model. For the ship performance analysis various types of parameters can be chosen. Some of them are: hull performance, main engine performance, the ship's overall efficiency and fuel efficiency. Even though many variables may be available, the focus is on the performance of the hull. The variables, or features as called in machine-learning, that describe the performance of the hull are: the speed over ground, wind speed and direction, ship course and heading, draft forward and aft, rudder angle, and the motions of the ship. These features were chosen based on the empirical and semi-empirical calculations, and their influence on the propulsion power. The measurements of the waves, even though could be useful for such a study may not be considered if the available data is considered to be too little, and may in some cases be invalid. To this previous list, the power delivered to the propeller is omitted, and this is because the delivered power is the target, what is wanted to be predicted, of this machine-learning problem.

[0098] Since the propulsive power is the target, a brief comparison of the two data sets is carried out here. The target of a supervised machine-learning problem may be relevant for both the training and the testing as discussed in a following subsection.

Training Data & Test Data:



[0099] Once the data to be used has been prepared, and the features and the target have been selected, the data is split for the training of the model and the testing of the model. Since machine-learning works in a way that it learns properties of a data frame and then applying them to new data it is common to split the data in two sets. These two data sets are called: training set, which is used to learn the properties of the data, and test set, on which the learned properties may be tested.

[0100] Here, a random split of the data into training and test sets is computed according to Pedregosa et al. The splitting used is of 80 - 20, i.e. 80% of the data is used for training and 20% for testing. This may be a common split in machine-learning problems, but also not the only possible option.

Training & Testing:



[0101] In the training part of the machine-learning, the models are applied on the training data. This way, the models learn the patterns of the data, the linearities and nonlinearities of the features depending on the model. For the training, the fit function of Pedregosa et al. is used to train the model. The theory on which the used models are based, is explained further below.

[0102] Once the model is trained, the unseen test data is given to the trained model to predict a target variable, which in this case is the propulsion power. The test data, therefore, that is given as input to the model, is uncoupled from the propulsion power. Once that the model has predicted the propulsive power, it is compared to the actual data of the propulsion to see how good or bad is the model at predicting. To verify the accuracy of the models, the approach of Pedregosa et al. uses a metrics module that has different functions to measure regression performance. For this case the R2 and the mean absolute error are used. Below a brief description of these metric functions is given.

Mean Absolute error:



[0103] The mean_absolute_error function from Pedregosa et al. calculates mean absolute error, a metric that corresponds to the expected value of the absolute error loss. Assuming that y^i is the predicted value from the model, of the i-th sample, and yi is the corresponding value taken from the data, then the mean absolute error (MAE) estimated over nsamples used in the model is defined as:


R2 score, the coefficient of determination:



[0104] The r2_score function called, the coefficient of determination provides a measure of how well new samples are likely to be predicted by the model. The best possible score is 1.0, while a negative score will indicate a worse prediction. For example, a model that always predicts the expected value of y, disregarding the input features, would get a R2 score of 0.0. Assuming y^i is the predicted value of the i-th sample and yi is the value from the data, then the score R2 estimated over nsamples is defined as:

where:


Machine Learning Models:



[0105] Exemplarily, three different models may be used: one statistical and two further machine-learning models, making a so called hybrid analysis. For the statistical method linear regression can be used, while the machine-learning models can be the XGBoost, short for extreme gradient boosting, a model of ensemble trees, and multi-layer perceptron regression, part of neural network models. All of the aforementioned are regression models as the desired output is a continuous variable. The reason for using three models is to compare the results of a statistical model to more complex machine-learning models that take into account nonlinearities within data. Both XGBoost, and the multi-layer regression take into account non-linearities with different approaches.

REFERENCE SIGNS



[0106] 
100, 200
apparatus
101, 103, 204, 205
interface
102, 201
processor
104
memory
202
database
203
training set
300
maritime vessel
301
ocean
302
sensor
400, 500
flowchart
401-406
step
501-504
step



Claims

1. An apparatus (100) for predicting a power consumption of a maritime vessel (300), the apparatus comprising:

an input interface (101) configured to receive a category indicator indicating a category of the maritime vessel, the category being indicative of a shape of the maritime vessel, wherein the input interface (101) is further configured to receive a speed of the maritime vessel (300) and a physical parameter, the speed and the physical parameter influencing the power consumption of the maritime vessel (300), the physical parameter being a wetted surface of the maritime vessel (300);

a processor (102) configured to select a predetermined power consumption model from a plurality of predetermined power consumption models upon the basis of the category indicator, wherein the processor (102) is further configured to determine a power consumption estimate based on the speed and the physical parameter using the predetermined power consumption model, to determine a power consumption correction based on the speed and the physical parameter using a power consumption correction model, the power consumption correction model representing a modeling error associated with the predetermined power consumption model, the power consumption correction model depending on a model parameter associated with the physical parameter, the power consumption correction model being a neural network model, the model parameter being a weight of a neuron associated with the physical parameter, and to combine the power consumption estimate with the power consumption correction in order to obtain a predicted power consumption of the maritime vessel (300); and

an output interface (103) configured to provide the predicted power consumption of the maritime vessel (300).


 
2. The apparatus (100) of claim 1, wherein the predetermined power consumption model comprises a calm water resistance model of the maritime vessel (300), and an added resistance model of the maritime vessel (300).
 
3. The apparatus (100) of claim 2, wherein the calm water resistance model is independent of the physical parameter, and wherein the added resistance model is dependent of the physical parameter.
 
4. Use of an apparatus (100) of any one of the claims 1 to 3 within a sail plan system for planning a sailing route of a maritime vessel (300).
 
5. A method for operating an apparatus (100) for predicting a power consumption of a maritime vessel (300), the apparatus comprising an input interface (101), a processor (102) and an output interface (103), the method comprising:

receiving, by the input interface (101), a category indicator indicating a category of the maritime vessel, the category being indicative of a shape of the maritime vessel;

receiving, by the input interface (101), a speed of the maritime vessel (300) and a physical parameter, the speed and the physical parameter influencing the power consumption of the maritime vessel (300), the physical parameter being a wetted surface of the maritime vessel (300);

selecting, by the processor (102), a predetermined power consumption model from a plurality of predetermined power consumption models upon the basis of the category indicator;

determining, by the processor (102), a power consumption estimate based on the speed and the physical parameter using the predetermined power consumption model;

determining, by the processor (102), a power consumption correction based on the speed and the physical parameter using a power consumption correction model, the power consumption correction model representing a modeling error associated with the predetermined power consumption model, the power consumption correction model depending on a model parameter associated with the physical parameter, the power consumption correction model being a neural network model, the model parameter being a weight of a neuron associated with the physical parameter;

combining, by the processor (102), the power consumption estimate with the power consumption correction in order to obtain a predicted power consumption of the maritime vessel (300); and

providing, by the output interface (103), the predicted power consumption of the maritime vessel (300).


 
6. Computer program comprising a program code for performing the method of claim 5.
 


Ansprüche

1. Vorrichtung (100) zum Vorhersagen eines Stromverbrauchs eines Seefahrzeuges (300), mit:

einer Eingabeschnittstelle (101), die ausgebildet ist eine Kategoriekennziffer, die eine Kategorie des Seefahrzeuges (300) angibt, zu empfangen, wobei die Kategorie eine Form des Seefahrzeuges bezeichnet, wobei die Eingabeschnittstelle (101) ferner ausgebildet ist eine Geschwindigkeit des Seefahrzeuges (300) und einen physikalischen Parameter zu empfangen, wobei die Geschwindigkeit und der physikalische Parameter den Stromverbrauch des Seefahrzeuges (300) beeinflussen, wobei der physikalische Parameter eine benetzte Oberfläche des Seefahrzeuges (300) ist;

einem Prozessor (102), der ausgebildet ist basierend auf der Kategoriekennziffer ein vorbestimmtes Stromverbrauchsmodell aus einer Vielzahl von vorbestimmten Stromverbrauchsmodellen auszuwählen, wobei der Prozessor (102) ferner ausgebildet ist eine Stromverbrauchsschätzung basierend auf der Geschwindigkeit und dem physikalischen Parameter unter Verwendung des vorbestimmten Stromverbrauchsmodells zu bestimmen, um eine Stromverbrauchskorrektur basierend auf der Geschwindigkeit und dem physikalischen Parameter unter Verwendung eines Stromverbrauchskorrekturmodells zu bestimmen, wobei das Stromverbrauchskorrekturmodell einen dem vorbestimmten Stromverbrauchsmodell zugeordneten Modellierungsfehler abbildet, wobei das Stromverbrauchskorrekturmodell von einem dem physikalischen Parameter zugeordneten Modellparameter abhängt, wobei das Stromverbrauchskorrekturmodell ein neuronales Netzwerkmodell ist, wobei der Modellparameter ein Gewicht eines dem physikalischen Parameter zugeordneten Neurons ist, und wobei der Prozessor (102) ausgebildet ist die Stromverbrauchsschätzung mit der Stromverbrauchskorrektur zu kombinieren, um einen vorhergesagten Stromverbrauch des Seefahrzeuges (300) zu erhalten; und

einer Ausgangsschnittstelle (103), die ausgebildet ist den vorhergesagten Stromverbrauch des Seefahrzeuges (300) bereitzustellen.


 
2. Vorrichtung (100) nach Anspruch 1, wobei das vorbestimmte Stromverbrauchsmodell ein Widerstandsmodell bei ruhiger See des Seefahrzeuges (300) und ein Modell für zusätzlichen Widerstand des Seefahrzeuges (300) umfasst.
 
3. Vorrichtung (100) nach Anspruch 2, wobei das Widerstandsmodell bei ruhiger See von dem physikalischen Parameter unabhängig ist und wobei das Modell für zusätzlichen Widerstand von dem physikalischen Parameter abhängig ist.
 
4. Verwendung einer Vorrichtung (100) nach einem der Ansprüche 1 bis 3 innerhalb eines Segelplansystems zur Planung einer Segelroute eines Seefahrzeugs (300).
 
5. Verfahren zum Betreiben einer Vorrichtung (100) zum Vorhersagen eines Energieverbrauchs eines Seefahrzeugs (300), wobei die Vorrichtung eine Eingangsschnittstelle (101), einen Prozessor (102) und eine Ausgangsschnittstelle (103) umfasst, mit:

Empfangen einer Kategoriekennziffer, die eine Kategorie des Seefahrzeuges (300) angibt, durch die Eingabeschnittstelle (101), wobei die Kategorie eine Form des Seefahrzeuges bezeichnet;

Empfangen einer Geschwindigkeit des Seefahrzeuges (300) und eines physikalischen Parameters durch die Eingabeschnittstelle (101), wobei die Geschwindigkeit und der physikalische Parameter den Stromverbrauch des Seefahrzeuges beeinflussen (300), wobei der physikalische Parameter eine benetzte Oberfläche des Seefahrzeuges (300) ist;

Auswählen, durch den Prozessor, eines vorbestimmten Stromverbrauchsmodells aus einer Vielzahl von vorbestimmten Stromverbrauchsmodellen basierend auf der Kategoriekennziffer;

Bestimmen einer Stromverbrauchsschätzung durch den Prozessor (102) basierend auf der Geschwindigkeit und dem physikalischen Parameter unter Verwendung des vorbestimmten Energieverbrauchsmodells;

Bestimmen einer Stromverbrauchskorrektur durch den Prozessor (102) basierend auf der Geschwindigkeit und dem physikalischen Parameter unter Verwendung eines Stromverbrauchskorrekturmodells, wobei das Stromverbrauchskorrekturmodell einen dem vorbestimmten Stromverbrauchsmodell zugeordneten Modellierungsfehler abbildet, wobei das Stromverbrauchskorrekturmodell von einem dem physikalischen Parameter zugeordneten Modellparameter abhängt, wobei das Stromverbrauchskorrekturmodell ein neuronales Netzwerkmodell ist, wobei der Modellparameter ein Gewicht eines dem physikalischen Parameter zugeordneten Neurons ist;

Kombinieren der Stromverbrauchsschätzung durch den Prozessor (102) mit der Stromverbrauchskorrektur, um einen vorhergesagten Stromverbrauch des Seefahrzeuges (300) zu erhalten; und

Bereitstellung des vorhergesagten Stromverbrauchs des Seefahrzeuges (300) durch die Ausgangsschnittstelle (103).


 
6. Computerprogramm, mit einem Programmcode zum Durchführen des Verfahrens nach Anspruch 5.
 


Revendications

1. Appareil (100) destiné à prédire la consommation d'énergie d'un navire maritime (300), l'appareil comportant :

une interface (101) d'entrée configurée pour recevoir un indicateur de catégorie indiquant une catégorie du navire maritime, la catégorie étant indicative d'une forme du navire maritime, l'interface (101) d'entrée étant en outre configurée pour recevoir une vitesse du navire maritime (300) et un paramètre physique, la vitesse et le paramètre physique influençant la consommation d'énergie du navire maritime (300), le paramètre physique étant une surface mouillée du navire maritime (300) ;

un processeur (102) configuré pour sélectionner un modèle prédéterminé de consommation d'énergie parmi une pluralité de modèles prédéterminés de consommation d'énergie sur la base de l'indicateur de catégorie, le processeur (102) étant en outre configuré pour déterminer une estimation de consommation d'énergie d'après la vitesse et le paramètre physique à l'aide du modèle prédéterminé de consommation d'énergie, pour déterminer une correction de consommation d'énergie d'après la vitesse et le paramètre physique à l'aide d'un modèle de correction de consommation d'énergie, le modèle de correction de consommation d'énergie représentant une erreur de modélisation associée au modèle prédéterminé de consommation d'énergie, le modèle de correction de consommation d'énergie dépendant d'un paramètre de modèle associé au paramètre physique, le modèle de correction de consommation d'énergie étant un modèle de réseau neuronal, le paramètre de modèle étant le poids d'un neurone associé au paramètre physique, et pour combiner l'estimation de consommation d'énergie avec la correction de consommation d'énergie afin d'obtenir une consommation d'énergie prédite du navire maritime (300) ; et

une interface (103) de sortie configurée pour fournir la consommation d'énergie prédite du navire maritime (300) .


 
2. Appareil (100) selon la revendication 1, le modèle prédéterminé de consommation d'énergie comportant un modèle de résistance en eau calme du navire maritime (300), et un modèle de résistance ajoutée du navire maritime (300).
 
3. Appareil (100) selon la revendication 2, le modèle de résistance en eau calme étant indépendant du paramètre physique, et le modèle de résistance ajoutée étant dépendant du paramètre physique.
 
4. Utilisation d'un appareil (100) selon l'une quelconque des revendications 1 à 3 au sein d'un système de plan de navigation pour planifier un itinéraire de navigation d'un navire maritime (300).
 
5. Procédé d'utilisation d'un appareil (100) pour prédire une consommation d'énergie d'un navire maritime (300), l'appareil comportant une interface (101) d'entrée, un processeur (102) et une interface (103) de sortie, le procédé comportant :

la réception, par l'interface (101) d'entrée, d'un indicateur de catégorie indiquant une catégorie du navire maritime, la catégorie étant indicative d'une forme du navire maritime ;

la réception, par l'interface (101) d'entrée, d'une vitesse du navire maritime (300) et d'un paramètre physique, la vitesse et le paramètre physique influençant la consommation d'énergie du navire maritime (300), le paramètre physique étant une surface mouillée du navire maritime (300) ;

la sélection, par le processeur (102), d'un modèle prédéterminé de consommation d'énergie parmi une pluralité de modèles prédéterminés de consommation d'énergie sur la base de l'indicateur de catégorie ;

la détermination, par le processeur (102), d'une estimation de consommation d'énergie d'après la vitesse et le paramètre physique à l'aide du modèle prédéterminé de consommation d'énergie ;

la détermination, par le processeur (102), d'une correction de consommation d'énergie d'après la vitesse et le paramètre physique à l'aide d'un modèle de correction de consommation d'énergie, le modèle de correction de consommation d'énergie représentant une erreur de modélisation associée au modèle prédéterminé de consommation d'énergie, le modèle de correction de consommation d'énergie dépendant d'un paramètre de modèle associé au paramètre physique, le modèle de correction de consommation d'énergie étant un modèle de réseau neuronal, le paramètre de modèle étant un poids d'un neurone associé au paramètre physique ;

la combinaison, par le processeur (102), de l'estimation de consommation d'énergie avec la correction de consommation d'énergie afin d'obtenir une consommation d'énergie prédite du navire maritime (300) ; et

la fourniture, par l'interface (103) de sortie, de la consommation d'énergie prédite du navire maritime (300).


 
6. Programme d'ordinateur comportant un code de programme destiné à réaliser le procédé selon la revendication 5.
 




Drawing


















































Cited references

REFERENCES CITED IN THE DESCRIPTION



This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

Patent documents cited in the description




Non-patent literature cited in the description