NO345705B1

NO345705B1 - A method for optimizing an efficiency of a vessel on a voyage

Info

Publication number: NO345705B1
Application number: NO20181045A
Authority: NO
Inventors: Lily Rachmawati; Justin N Norman; Lars Ove Silseth
Original assignee: Kongsberg Maritime As
Priority date: 2018-08-03
Filing date: 2018-08-03
Publication date: 2021-06-21
Also published as: NO20181045A1; WO2020025745A1

Description

A method for optimizing an efficiency of a vessel on a voyage Background of the invention

Vessels, such as ships, fulfil many purposes, such as passenger transportation, goods transportation, transportation of oil and chemicals, cable laying, and fish trawling. Each ship has its own characteristics, such as size, weight, shape, engine configuration, and tank sizes. Furthermore, during voyages, the weight changes due to loading and unloading of for instance goods, ballast water, and fuel. This in turn causes a change in the hydrodynamic profile of the ship, both in terms of the draft of the ship, and in the ship’s profile above water. Currents and winds will have a different effect of the ship depending on the load.

It is desirable to improve a ship’s performance, since even a small improvement in the efficiency of a large ship will lead to large savings.

To improve the efficiency, the ship can be designed to have more advantageous characteristics, such as a better hydrodynamic profile, an improved light displacement, etc. More energy-efficient engines can be employed, low-resistance paint be used, and so on. For a given ship, however efficient, there may still be room for an increase in efficiency.

Certain ways of improving a ship’s efficiency are well known. For fuel efficiency, it is known that slower travel typically gives better fuel efficiency. For a propeller-driven ship, increasing propeller pitch and decreasing revolutions per minute (RPM) may also improve fuel efficiency. However, these relationships are qualitative only. To be useful, these relationships must be firmly established for each vessel in question. While it is reasonably straightforward to identify optima in propeller efficiency if the thrust and torque coefficient curves are known, it is not so for ship propulsion. In addition to the propeller characteristics, the net ship propulsion depends also on the hull and the interaction between the propeller and the hull. The modification of the flow introduced by the presence of the hull and the positioning of the propeller near the stern means that the propeller operates not in a uniform free stream as in open water tank test but in a complex wake field. In return the propeller also exerts a pressure field on the hull, sucking water back towards it with a mean effect of thrust deduction. Not only is it not feasible to perform a tank test for every possible scenario; it is impossible. Environmental factors, such as currents and wake field influence, are impossible to comprehensively test for in advance.

Vessel resistance is another important quantity that cannot be precisely computed from first principles in all practical sea states and vessel conditions. Vessel resistance includes frictional, residual and air resistance, all of which vary with the square of vessel speed at low speed and in fair weather. Frictional resistance may account for anywhere between 45-90% of total resistance and depends on the size and roughness of the wetted area, the ship speed and the hull form. The vessel frictional resistance changes with time in part due to fouling of propeller and hull. Fouling leads to an increased surface roughness, which translates into a corresponding increase in friction during vessel movement through water. The second source of resistance, residual resistance, is caused by waves and eddies created as the vessel moves through water. At low speeds, wave resistance increases with the square of the ship speed and contributes something like 8-25% of resistance, but it increases much more rapidly at higher speeds and creates a speed barrier where increase in propulsion power gets converted into wave energy instead of propulsion. At such high speeds, the residual resistance can account for 60% of total resistance. Air resistance, typically the smallest component, accounts for approximately 2-10% of total ship resistance, depending on the apparent wind speed and direction, the ship speed, and the cross-sectional area of the ship above waterline. Accurate calculation of vessel resistance in various conditions is notoriously difficult and is achieved, imperfectly, with the help of computational simulations as well as towing tank tests. Sea state (including apparent wind, current and wave amplitude) and fouling modify resistance in ways that are practically infeasible to be captured in computational fluid dynamics (CFD) simulations and towing tank tests. From studies, fouling may increase resistance by about 25-50% through the lifetime of the ship, while sea state could increase resistance well beyond 20%.

US 2014/336853 A1 discloses a method, non-transitory computer readable medium and performance optimization computing device for optimizing the performance of a ship. Data associated with one or more operational parameters associated with the ship is obtained. One or more performance values corresponding to the obtained data are identified. One or more optimal operational parameters are determined based on a comparison of the identified one or more performance values and one or more historical performance values. The historical performance values correspond to historical data associated with the one or more operational parameters. The determined one or more optimal operational parameters for the ship are provided.

WO 2013/178778 A1 discloses a computer-implemented method to produce a model that predicts the performance of a ship. The method comprises the steps of creating an initial model for simulating the performance of the ship by means of relationships between factors related to the ship during operation and parameters presenting primary dynamic input data that depend said factors depend on.

Measurement results are obtained from sensors during the operation of the ship for producing a set of new dynamic input data to be used in the model. A simulation result for the ship performance is calculated by using said produced input data in said model instead of the primary dynamic input data. The simulation result is then compared to the actual ship operation. If the difference between the simulation result and the actual operation exceeds defined threshold criteria, the model is improved by repeatedly producing new sets of dynamic input data to be used in the model by means of new measurement results until defined quality criteria are met. The invention is also concerned with a system and that performs such a method and a computer program.

Summary of the invention

Embodiments of the present invention address at least some of the issues described above.

In a first aspect, the present invention provides a method for optimizing an efficiency of a vessel on a voyage, the vessel being controllable by changing a set of adjustable control parameters via a set of control levers, the efficiency of the vessel being represented by a set of one or more efficiency parameters. The method comprises:

- storing a set of historical operating data sets, each historical operating data set comprising operating values for the set of adjustable control parameters and corresponding operating values for a set of vessel performance parameters, and in some preferred embodiments a set of environmental parameters affecting the vessel,

- providing, based on the historical operating data sets, a current datadriven model describing the one or more efficiency parameters as corresponding one or more functions, each function depending on one or more of the adjustable control parameters and one or more of the vessel performance parameters and in some preferred embodiments one or more of the environmental parameters,

characterised in that the method further comprising

- determining that the vessel has reached a first steady state of travel,

- determining current values of the set of efficiency parameters at the first steady state of travel,

- determining, using the current data-driven model, one or more candidate sets of adjustable control parameter values having improved efficiency parameter values compared to the current values of the set of efficiency parameters,

- selecting a preferred candidate set among the one or more candidate sets, and

- applying the preferred candidate set of adjustable control parameters by adjusting the control levers to the values in the preferred candidate set.

The term “control lever” shall be construed broadly as means for changing the corresponding control parameter. For instance, on some vessels, the propeller speed is controlled, at least in part, using a physical lever, whereas on other vessels, the propeller speed is adjusted by interacting with a touch-sensitive screen displaying for instance an image resembling a physical propeller speed lever.

The optimization method for determining one or more candidate sets of adjustable control parameter values having improved efficiency parameter values compared to the current values of the set of efficiency parameters explores the space of possible control lever values to find a combination that yields more optimal performance as evaluated by underlying cost functions/computational models.

The set of one or more efficiency parameters may for instance comprise one or more of: a fuel efficiency, a mission efficiency, a life cycle cost index, a motion index (signifying for instance a degree of rolling around one or more axes of the vessel during the voyage).

In prior art systems, pitch and speed of the propeller are varied along combinator curves. The combinator curve is the optimum combination of pitch and shaft speed, or at least is thought to be optimal, with respect to fuel consumption. Combinator curves may be tuned from time to time and, an in special cases, trained crew members may input a value for propeller pitch directly into the ship’s control system deviating from the combinator curve. The combinator curve is usually nothing more than a static recommendation that tells the crew which combination of propeller pitch and propeller RPM achieves the best fuel efficiency. Selecting operating conditions that are not on the combinator curve requires knowledge that is not available. In practice, a crew will not choose to deviate from the combinator curve except under exceptional conditions.

The inventors found that the combinator curve, although a helpful tool, may end up actually hampering vessel efficiency. The combinator curve is not optimal when propeller(s) and engines are at different states from the state at which the propeller efficiency was initially measured to derive the combinator curve(s), due for instance to ageing and fouling. When the assumptions behind the combinator curve are no longer true, the combinator curve no longer represents efficient pairing of propeller pitch and RPM. Changes in sea state also changes the propeller efficiency. As part of the present invention, the propeller speed and pitch are allowed to vary independently. In other words, the optimization explores propeller pitch and RPM combinations not just on the combinator curve, but also away from the combinator curve. In a sense, embodiments of the invention that involve the propeller as an adjustable control parameter can be seen as providing an adaptive combinator curve. This leads to the discovery of previously unknown minima in the propulsion power as a function of propeller pitch and RPM, and of unknown minima in engine specific fuel consumption as a function of e.g. diesel engine RPM and load, an indicator of diesel engine fuel efficiency. As indicated, the method not only allows the discovery of new such minima in general, it also specifically allows for the discovery of minima associated with settings that are not on the combinator curve.

In steady state motion, moving a ship through water requires that effective thrust generated balances resistance encountered. Thrust is provided for instance by one or more diesel engines and/or electric motors through for instance a single controllable-pitch propeller. At any point in the vessel transit, the resistance encountered by the ship, which is a complex function for instance of ship displacement, ship speed and fluid characteristics for a given ship hull, must be balanced by the thrust force (or simply “thrust”).

From basic propeller theory, for a given fixed pitch propeller, an appropriate angular speed can be applied to achieve the required thrust for a specific advance speed while absorbing a certain shaft torque and power. Thrust and Torque are both functions of the advance speed, propeller diameter, angular speed, and fluid and flow characteristics. These functions are complex enough to defy a closed form mathematical expression, and analysis is performed by empirically identifying thrust and torque coefficient curves as functions of pertinent variables like advance speed, etc. For different pitch angles, the propeller behaviour needs to be captured with multiple thrust and torque coefficient curves, which tend to have the same general trend across different pitch values in a large range of propeller speed and advance speed but sufficiently different in absolute magnitudes.

In the operation of a controllable pitch propeller, there is an additional degree of freedom provided by the pitch if the combinator curve constraint is abandoned. Computationally speaking, this amounts to traversing two two-dimensional hyperplanes given by thrust and torque curves. Propeller efficiency, the portion of torque absorbed that is converted by the propeller into thrust, is a function of the thrust and torque coefficients that both vary with parameters mentioned above.

Optimizing propeller performance in open water is equivalent to identifying local and/or global optima in the propeller efficiency hyperplane, a tractable computational task with propeller thrust and torque curves available.

Generally, the shaft power required increases at a higher rate as advance speed of propeller in water increases. This is attributable to drag increasing with squared velocity and other factors. Looking at marginal rates of return (i.e. knot per kilowatt supplied), going from a lower speed to a higher speed is “expensive”. Furthermore, a minimum ratio of power/speed is obtained at the lowest advance speed and running at the highest pitch.

The amount of power required to produce the same advance speed in identical load (given the same weather and waters) can vary significantly for propellers with variable pitch. At higher speed values, there are fewer combinations of propeller RPM and pitch, and consequently less room for optimizing performance. At lower speed values, there are more combinations, and power intake could vary significantly while providing the same speed.

Generally, at lower advance speed, the shape of iso-speed curves is such that lowering RPM and increasing pitch corresponds to higher fuel efficiency. At higher advance speed, reducing RPM and increasing pitch does not always translate to better fuel efficiency; rather there is a local maximum. This means that two different sets of propeller RPM and pitch combination could correspond to the exact same fuel efficiency at the same advance speed. For example, at a speed of 16 knots, running the propeller at 165 RPM and pitch/diameter of 1.1 converts the same amount of torque into thrust as 138 RPM and a pitch/diameter of 1.4. (This is a example only. The relationship depends on a number of propeller characteristiscs, including for instance the number of blades and their individual size and shape.)

However, depending on the engine(s) employed to drive the propeller, this does not necessarily translate to an improved fuel efficiency. The specific fuel consumption of a diesel engine varies with rotational speed at the same load. Unless specific fuel consumption also happens to be equal (a probable scenario) at the two angular speeds, the fuel efficiency would be different. Different propeller RPM and pitch in a non-uniform flow expected when the propeller works as part of a ship may cause different thrust deduction effect. The difference may be negligible, though, and in practice other considerations, such as equipment wear and tear, should come into play when deciding between the two propeller RPM values. In other words, the life cycle cost could advantageously be included into the optimization procedure to optimize vessel performance.

The issues described above are all gracefully handled by embodiments of the present invention.

The set of adjustable control parameters may for instance comprise at least one or more of: a propeller speed, a propeller pitch, a power train setting, an engine part load, a commanded vessel heading, a vessel hull trim.

The set of vessel performance parameters may for instance comprise one or more of: a fuel flow rate, an actual vessel heading, a vessel speed over ground, a vessel course over ground, a shaftline torque, a vessel roll angle dependent parameter, a vessel pitch angle dependent parameter.

In some embodiments, the preferred candidate set is applied only subject to an improvement criterion being fulfilled. Such embodiments prevent continuous application of new adjustable control parameter values. An improvement criterion may for instance comprise a requirement that a change in an adjustable control parameter value must exceed a minimum absolute amount or exceed a minimum percentage amount or other relative amount relative to a current value of the adjustable control parameter value. For instance, a change in main engine RPM must exceed 1% or must exceed 5 RPM before the control levers are adjusted.

In some embodiments, the set of one or more efficiency parameters comprises a fuel efficiency parameter, and the current data-driven model comprises:

- a fuel flow function relating a fuel flow of the vessel to a first set of one or more of the adjustable control parameters, and

- a speed function relating a speed of the vessel to a second set of one or more of the adjustable control parameters,

and the fuel efficiency parameter is determined based on the fuel flow function and the speed function, which in some preferred embodiments may be determined as a ratio between the fuel flow function and the speed function.

The inventors found that this is a numerically efficient approach to optimizing the fuel efficiency during a voyage.

In some embodiments, the method further comprises:

- determining a predicted vessel performance parameter value using the current data-driven model and current values of the adjustable control parameters and in some embodiment also the current values of the environmental parameters,

- determining an actual vessel performance parameter value for comparison with the predicted vessel performance parameter value,

- determining that a discrepancy between the actual value and the predicted value of the vessel performance parameter exceeds a tolerance threshold and in response:

i. providing, based on the amended set of historical operating data sets, an updated data-driven model describing the one or more efficiency parameters as corresponding one or more functions, each function depending on one or more of the adjustable control parameters and one or more of the vessel performance parameters and in some preferred embodiment one or more of the environmental parameters.

Such embodiments test the current data-driven model to determine whether it predicts the performance parameters with sufficient accuracy. If it does not, the current data-driven model is replaced by an updated data-driven model. The updated data-driven model is based on historical data that emphasizes more recent historical data compared to the current data-driven model. The updated data-driven model then becomes the current data-driven model. By emphasizing more recent historical data, the updated data-driven model better reflects the recently observed behaviour of the vessel.

Importantly, it allows learning the vessel behaviour from sensor data instead of relying on expensive benchmarking tests (e.g. tank tests) to characterize vessel behavior throughout the lifecycle of equipments and in various sea states. Much of the characteristics obtained in benchmarking must be considered constant through the lifecycle of the equipment because further benchmarking after the initial benchmarking and vessel launch would rarely, if ever, be performed, and would be very costly. Accordingly, outdated characteristics are in reality used for most of the vessel’s lifetime. The present invention provides a current data-driven model and provides for updating the data-driven model when the current data-driven model leads to an excessive discrepancy between the actual value and the predicted value of a vessel performance parameter. The updated data-driven model automatically accounts for changes in vessel behaviour, whatever the source(s) of the change. Embodiments that employ updating of the data-driven model of the vessel behaviour are particularly powerful, as they allow for a more precise prediction of control levers that provide increased efficiency. As noted before, sea state is a factor that obviously cannot be predicted. The present model, especially embodiments that include updating the data-driven model, can account very well for the sea state and other environmental parameters.

In some embodiments, providing the updated data-driven model may comprise the step of reducing an influence of older historical data sets relative to newer historical data sets. This may be achieved by partly decreasing the weight of older historical data relative to newer historical data, and/or by entirely removing older historical data.

In some embodiments, at least one of the one or more functions depends on the set of environmental parameters, and the set of environmental parameters comprises one or more of: an apparent wind speed, an apparent wind direction, a wave height, a wave direction relative to a hull of the vessel, a true wind speed, a vessel physical condition. The inventors found that including one or more of these parameters can improve the model’s predictive capabilities significantly. A physical condition of the vessel might for instance be the state of the vessel’s paint, the degree of wear of the propeller, and the amount of wear of one or more of the vessel’s engines.

Determining that the vessel has reached a steady state of travel may for instance comprise determining that a speed of the vessel does not vary more than allowed by a predefined steady-state speed criterion. For instance, a steady-state speed criterion might be that the measured speed is within 5 % of the average speed over the past 1 minute. As another example, the measured speed does not vary more than 1 knot during a period of 5 minutes. The details are merely a matter of design in view of the present disclosure.

Determining that the vessel has reached a steady state of travel may alternatively or additionally include one or more of:

- determining that a propeller speed does not vary more than allowed by a predefined steady-state propeller speed criterion,

- determining that a vessel heading does not vary more than allowed by a predefined steady-state vessel heading criterion,

- determining that a course-over-ground does not vary more than allowed by a predefined steady-state Course over Ground criterion,

- determining that a propeller torque does not vary more than allowed by a predefined steady-state propeller torque criterion,

- determining that a fuel flow does not vary more than allowed by a predefined steady-state fuel flow criterion.

For these parameters, absolute or relative variations might be used as criteria.

In some embodiments, the set of efficiency parameters comprises at least two efficiency parameters, and the step of determining one or more candidate sets comprises determining at least two non-dominated candidate sets. Such methods allow for a more complex optimization of the performance of the vessel. By considering both fuel efficiency and mission efficiency at the same time, the model can help find the optimal balance between fuel consumption and travel time. It might for instance be worth sacrificing some fuel efficiency if travel time is costly. Including a life cycle cost index will allow the model to take into account the cost of wear of parts such as engines.

In some embodiments, the preferred candidate is selected based on a preference of one efficiency parameter in the set of efficiency parameters over all other efficiency parameters in the set of efficiency parameters. Note that this is not the same as including only one efficiency parameter in the first place.

A second aspect of the invention provides digital computing and storage hardware configured specifically to perform a method in accordance with the first aspect of the invention.

A third aspect of the invention provides a computer-readable storage medium comprising program instructions that, when executed on suitable digital computing and storage hardware, cause the digital computing and storage hardware to perform a method in accordance with the first aspect of the invention.

Brief description of the drawings

Figure 1 illustrates an example of relationships between adjustable control parameters, performance parameters and environmental parameters.

Figure 2 is a flow chart illustrating a method in accordance with an embodiment of the invention.

Figure 3 illustrates an allowed space for exploring combined values of propeller pitch and propeller RPM during optimization.

Figure 4 is a flow chart illustrating a method in accordance with another embodiment of the invention.

Figure 5 schematically illustrates dominated and non-dominated solutions, as well as selection of preferred solutions in dependence of which efficiency parameter is preferred among multiple efficiency parameters.

Detailed description of selected embodiments

In the following, embodiments of the invention will be described with reference to the accompanying drawings.

The following example refers to a vessel that has one diesel main engine, two constant speed auxiliary engines, a controllable pitch propeller and bow thrusters. The vessel can be operated in six power modes:

- PTO (power take off) 60Hz

- PTO 50-60Hz,

- PTI (power take in) DE (diesel engine),

- PTI Boost,

- PTO Limited Combinator (Split), and

- Full Combinator.

In PTO 60Hz mode the main engine runs at MCR (maximum continuous rating) speed and supplies propulsion power as well as hotel loads. In PTO 50-60Hz mode the main engine can be operated from 625 RPM to 750 RPM, supplying power to the propeller as well as the switchboard and the rest of the vessel through a rotating converter. In Full Combinator mode the main engine operates at the largest range of speed variation (450 to 750 RPM) to power propulsion while the auxiliary engines provides power for the rest of the vessel. In PTO Limited Combinator (split), auxiliary engines supply hotel load while the main engine provides propulsion and thruster power. PTI Boost mode provides the largest possible propulsion power with the main engine running at MCR speed and auxiliary engines adding to the propulsion power through an Active Front End (AFE). When relatively small propulsion power is needed, the PTI DE mode where auxiliary engines provide all power needs can be utilized.

The vessel also has a number of fresh water tanks, diesel oil tanks and urea tanks with varying capacity. To adjust the trim of the vessel and the resistance in water, distribution of diesel oil in the appropriate tanks can be varied.

In the present example, the data-driven model considers the vessel to be completely characterised by the following parameters:

1. Propeller RPM

2. Propeller pitch

3. Shaftline torque

4. Shaftline power

5. Speed through water (measured for instance with a pitometer)

6. Speed over Ground (SoG, measured with Global Positinging System, GPS) 7. Main Engine RPM (Diesel mechanical)

8. Main Engine Load (Diesel mechanical)

9. Main Engine fuel flow rate

10. Power Configuration

11. Auxiliary engine 1 (AUX1) fuel flow rate

12. Auxiliary engine 1 (AUX1) power output

13. Auxiliary engine 2 (AUX2) fuel flow rate

14. Auxiliary engine 2 (AUX2) power output

15. Efficiency of propulsion, efficiency of shaft generator, efficiency of diesel electric engines as taken from datasheets

16. Apparent Wind Speed

17. Apparent Wind Angle

18. Tanks contents

19. Vessel roll angle (“Euler x”)

20. Vessel pitch angle (“Euler y”)

21. Course over ground

22. Heading

23. Hotel load

24. Minimum metacentric height, distance from keel to metacenter, Longitudinal Center of Gravity, Vertical Center of Gravity.

Fig. 1 schematically illustrates relationships in an exemplary data-driven model, relating adjustable control parameters, performance parameters and environmental parameters to one another. Arrows point from inputs to outputs. Main Engine RPM (ME RPM), for instance, depends on the control levers “power mode” and “propeller RPM”. The chart may be used to track the effect of changing a control lever as well as trace back the control levers to be varied to achieve a change in a dependent variable.

Adjustable control parameters are shown in simple square boxes in the diagram. Propeller RPM, propeller pitch, Heading, and Main Engine load (“ME load”) are examples of adjustable control parameters (“control levers”).

The model also includes performance variables, which depend on the control levers, and in some cases on non-controllable variables. These are shown in ovals.

Examples include Speed, Main Engine fuel flow rate, Speed over Ground (“SoG”), Course over Ground (“CoG”), and hull attitudes (“Euler x”, “Euler y”). These are measureable parameters.

Some parameters are considered to be relatively constant are shown underlined. These include true wind direction (“True wind”), Hotel load and Total tank content.

The remaining parameters are dependent variables whose values can be determined from input variables by a simple computation. These are shown in boxes having cut corners.

Frequently during the voyage, data are stored in order to record how the vessel performed under specific conditions. Speed over ground, wind direction, apparent wind speed, fuel rates, and so on, are stored together with the settings of the adjustable control parameters at those times. These historical data sets form the basis for the data-driven model. The data-driven model is established by adjusting parameters in the underlying computational models to best fit the historical data, or at least some of it, namely that which is considered important. This can be selected either automatically or at least partially manually. Usually, the most recent historical data sets are used when fitting the parameters of the computational models, as these best reflect the recent performance of the vessel and therefore, all else being equal, will result in a data-driven model which has a strong predictive power.

Fig. 2 illustrates an embodiment of the invention. Historical data sets are stored during a voyage, as illustrated by step 201. The historical data sets are stored in a database 202, preferably a database stored on a computer system. In step 203, a current data-driven model 204 is determined. Before applying the data-driven model for the purpose of determining one or more candidate sets of adjustable control parameter, it is determined, in step 205, that the vessel is in a steady state, i.e. that the vessel is not in the process of, as examples, significantly slowing down or speeding up or changing heading. In step 207, current values of the one or more efficiency parameters are determined. In step 209, one or more adjustable control parameter candidate sets are determined, which, if applied, will (or at least should, based on the model’s strength in view of current conditions) result in an improved efficiency. In step 211, a particular candidate set is selected. If there are more than one candidate set, a preferred candidate set may be selected based on a criterion or criteria, or it may be selected randomly. In step 213, the preferred candidate set is applied. As a result, the vessel should exhibit an impoved efficiency.

The limited validity of data-driven models induced from historical data is a fundamental issue as the optimization is especially interesting in practice when it picks up values of control levers previously unexplored but estimates of the effect of such control lever values are unreliable as there is no data support.

Design data, e.g. from propeller open water test or tank tests, a diesel engine map and an auxiliary engine datasheet, offer some information on areas unexplored by the historical operation of the vessel. Data-driven models based on historical data are invoked only to evaluate a candidate solution in the optimization if it involves control lever values within a set percentage around past data values. Beyond the set percentage, data driven models based on design data may for instance be invoked. This is illustrated in Fig. 3.

Combinations of propeller RPM and pitch available are limited in comparison to the possible set of values. The area illustrates propeller RPM of 80 to 140 and pitch of 0.8 to 1.5 in p/d. All combinations are in principle allowed. The dots (that form curve-like structures in the plot) signify operating points found in historical data sets. A 7% region beyond (in all directions) the support of historical data sets is illustrated as dark areas in the plot. Available test data from the design phase may be utilized beyond that boundary.

Propeller open water test results may be employed to predict shaftline torque and power, vessel speed and speed over ground when the propeller RPM and pitch are beyond the specified boundary around historical data. The diesel engine fuel map may be employed in lieu of the historical data driven model for main engine fuel consumption for points beyond historical data, and datasheets for auxiliary engines 1 and 2 may be employed at all times. Note that this design data should not be expected to closely represent operational behaviour of the vessel, as the tank test result represents open water behaviour for a stock propeller. If substituted with more specific information, e.g. self-propelled towing test for the right hull shape and the actual propeller instead of the stock propeller, the resulting models would be closer to the true behaviour. However, the model points in the right general direction. It is also noted that design data supplementing historical data is generally only available for a very limited number of variables. Motion parameters, such as roll and pitch, may be predicted using models derived from historical data.

Parameters significant in theoretical computations like fluid characteristics, hull form descriptions (e.g. wetted area, area above the water, block coefficient, hull roughness, hull efficiency), resistance coefficients, displacement/draught, wave period and amplitudes, current, etc. are not included in the computational models. These factors are not available as measurements/data. As described previously, they are taken into account implicitly in the parameters of the data-driven models learnt from historical data sets.

Selected embodiments of the invention provide online learning of the data-driven model. These are needed for the computational models to implicitly capture the factors just described and no data is available to indicate when they vary and how their variation affect for example shaft torque. The rate at which these models need to be updated would differ: Shaftline Torque for a given speed, propeller RPM and pitch, may not change as quickly as the Course Over Ground for a given heading.

The purpose of embodiments of the invention is to find controllable parameters that optimize one or more efficiencies. In some embodiments, this means finding values for the adjustable control parameters, X, that minimize cost functions, F, representing the efficiency parameters. In the present example, the adjustable control parameters

are

x1: main engine rotational speed (in RPM)

x2: propeller rotational speed (in RPM),

x3: propeller pitch (in p/d)

x4: power mode

x5: main engine output power (in kW)

x6: auxiliary engine 1 (AUX1) output power (in kW)

x7: auxiliary engine 2 (AUX2) output power (in kW)

x8: heading (in degrees)

x9: tank 1 content

x10: tank 2 content

x11: tank 3 content

x12: tank 4 content

x13: tank 5 content

x14: tank 6 content

These are the variables that the crew can directly control, and a goal is to find combinations of values of these parameters that minimize one or more cost functions, as desired, such as:

Fuel efficiency:

Target speed:

Rocking:

Life cycle cost:

where LCC(X) is a catch-all life cycle cost function. As an example, the life cycle cost might represent the rate of wear of the main engine. The main engine wears more quickly the higher the power output.

The optimization is a numerical search procedure and can yield values of X that are not allowed. For instance, the propeller pitch might be limited to values between 0 and 1.4. During the optimization process, the search process takes such contraints into account by discarding those values of X that do not fulfill all constraints.

Another example is that the propeller RPM must be below 130. The engines also have a limited output power capability, and therefore constraints exist for those parameters as well. Obviously, the heading must be in the range from 0 to 360, which is therefore also imposed when determining whether a solution X is valid.

In some embodiments of the invention, the optimization method finds solutions to optimize more than one efficiency parameter at the same time. In practice this involves finding a set of recommended operating points that represent trade off in the performance metrics, with an option to prioritize one objective over another, and in practice typically within a user specified deadline in order to obtain a solution within a reasonable amount of time.

A multi-objective optimization problem involves the optimization of a set of conflicting objectives such that there is no single solution that outperforms all other solutions in all objectives. Solutions to a multi objective problem are optimal in the sense of Pareto optimality, i.e. a solution is optimal if improvement in one objective implies degradation in at least one other objective. A multi-objective optimization problem has a non-unique set of solutions which represent trade-offs between the objective values and form a non-dominated front in the objective space.

The main challenge in a multi-objective problem arises from the partial order in the objective space. Suppose there are three solutions: A, B and C. Suppose solution A is better in all objectives than solution B, and solution A and C are non-dominated (i.e. they represent trade off in objective values and they are incomparable as there is no basis for deciding that A is better than C). That does not necessarily mean that solution C is better than solution B. Solution B cannot possibly dominate solution C but the two can be non-dominated.

As optimization algorithms operate by generating a search path in the objective space that potentially minimizes the cost function, when there is no total order in the cost function, another approach is necessary. Typical approaches to multiobjective optimization include:

1. Conversion into a single-objective problem e.g. by means of aggregation of objectives or formulation of objectives as constraints, or goal vectors. By converting the problem into a single-objective problem, a total order is induced and the resulting optimization problem can be addressed as usual, i.e. by finding the optimal solution to the now single-objective problem.

2. Pareto dominance based search. In Pareto dominance based search the optimization algorithm aims to find a set of non-dominated solutions lying as close as possible to the theoretical Pareto front in the objective space. Apart from proximity to the optimal front, a uniform distribution of solutions in the objective space is usually considered good, as it gives a useful set of alternatives representing trade off in objectives to choose from.

Pareto dominance based search is typically achieved by ranking solutions based on pair-wise dominance relation, and archiving the best non-dominated solutions found so far, as a basis for further exploration in the search space. Solution A dominates solution B if solution A performs at least as well as B in all objective functions and outperforms B in at least one objective. Pareto ranking schemes assign a solution quality metric, akin to the cost function, based on the pair-wise dominance relations. Pareto rank is also a population-dependent metric, in that the rank of a solution is evaluated relative to other solutions present as alternatives and would change with variation in the alternatives. Multi-objective optimization algorithms that aim to find a set of non-dominated solutions are typically population-based, where the fitness of a decision vector is defined relative to other decision vectors in the population. The preference-based multi-objective optimization aims to converge to a subset of the Pareto front that favors the preferred objective versus the rest. This may for instance be achieved by placing selective bias on the candidate solutions throughout the optimization process by controlling the distribution of objective vectors. Fig. 5 schematically illustrates a number of solutions and the values of two efficiency parameters, f1, and f2. To optimize with respect to f1, solutions 501 are preferred. Solution 502 in particular provides the lowest value of f2 subject to the lowest value of Solution 502 is therefore preferred in case efficiency parameter f1 is critical. If efficiency parameter f2 is of some importance, solution 504 might be the preferred solution among solutions 503. Although solution 504 does not provide the lowest value of f1, it may be the preferred solution because what is lost in terms of efficiency in respect of is more than gained in terms of the increased efficiency of f2.

Solutions 505 are dominated because solution 506 is better in terms of both f1, and f2. Therefore none of these would be selected as a preferred solution and their corresponding control lever settings would not be applied. No other solutions in Fig. 5 are dominated.

To optimize with respect to f2r solutions 507 are preferred. Solution 508 in particular provides the lowest value of f1 subject to the lowest value of f2. Solution 508 is therefore preferred in case efficiency parameter f2 is critical. If efficiency parameter f1 is of some importance, solution 509 might be the preferred solution. Although solution 509 does not provide the lowest value of f2, it may be the preferred solution because what is lost in terms of efficiency in respect of f2 is more than gained in terms of the increased efficiency of f1.

In the discussion above, it was assumed that the process was one of minimizing the objective functions. Generally, the present invention is not limited to methods that specifically require minimization or specifically require maximization. The person skilled in the art will recognize, in view of the present description, that the efficiency parameters are generally optimized, whether it be by minimization or by maximization, or a combination.

Fig. 4 illustrates updating the current data-driven model. Usually, the method will applied in conjunction with the method illustrated in Fig. 2 and described above. Therefore, the method is illustrated as connecting to A, labelled “215”, of Fig. 2. Using the current data-driven model 204, a vessel performance parameter is predicted in step 403. (As mentioned previously, and as illustrated in Fig. 1, performance parameters include for instance Speed, Main Engine fuel flow rate, Speed over Ground (“SoG”), Course over Ground (“CoG”), and hull attitude.) Furthermore, an actual value of the performance parameter is obtained in step 405. If the discrepancy between the predicted and the actual value of the performance parameter exceeds a tolerance level, as determined in step 407, the current datadriven model is updated as shown in step 410 resulting in an updated data-driven model which then replaces the current data-driven model as illustrated in Fig. 4. The updating is based on the currently stored historical data sets 202, which have typically been amended by new operating data sets compared to the time when the current data-driven model, before being updated, was determined. These data sets reflect all of the parameters discussed above. On a relatively short time scale, environmental factors, such as wind speed and/or direction and sea state, may have changed. As discussed, these affect the efficiency of the vessel in a way that may favour a different combination of propeller RPM and pitch compared to the combination that were favoured when the last data-driven model was determined. On a longer time scale, fouling, changing load, and so on affect the efficiency.

Independent of the time scale, the presence of new operating data sets lead an updated data-driven model that better accounts for the present conditions. As a consequence, the updated data-driven model 404 better predicts control parameter values that will improve the efficiency of the vessel, which is the very purpose of the data-driven model.

Typically, all the historical data that the vessel has generated are stored in a persistent media, e.g. a disk. During the optimization, only a subset of those data sets are used to determine the data-driven model. The subset includes those data sets that are considered relevant for the situation at hand, e.g. resembles current condition or more recent. Therefore, the embodiment in Fig. 4 illustrates a step 409 of selecting historical data sets for providing the updated data-driven model. The simplest approach is to simply use a number of the latest historical data sets.

However, older historical data sets may be included in the subset because of a particular relevance. For instance, if wave conditions have changed in the course of hours, for instance due to a change in heading, there will be recent historical data sets for different wave conditions. If the waves are strong, they will have a large influence in the optimization. If the vessel now returns to the heading that were applied some hours ago, the historical data sets from that time may give a better prediction of optimal control lever settings.

S’ may contain duplicates of members of S that are particularly relevant to the current condition, i.e. members that are emphasized. Members that are de emphasized may have lower weightage or not included at all. These for example could be datapoints that conflict with the most recent datapoint (e.g. same propeller pitch and RPM but different torque).

The subset used for determining the data-driven model typically resides in active memory/RAM of the computer system involved in establishing the data-driven model based on the historical data sets.

Claims

1. A method for optimizing an efficiency of a vessel on a voyage, the vessel being controllable by changing a set of adjustable control parameters (X) via a set of control levers, the efficiency of the vessel being represented by a set of one or more efficiency parameters (f), the method comprising:

- storing a set of historical operating data sets, each historical operating data set comprising operating values for the set of adjustable control parameters (X) and corresponding operating values for a set of vessel performance parameters,

- providing, based on the historical operating data sets, a current datadriven model describing the one or more efficiency parameters (f) as corresponding one or more functions, each function depending on one or more of the adjustable control parameters (X) and one or more of the vessel performance parameters,

characterised in that the method further comprising

- determining that the vessel has reached a first steady state of travel, - determining current values of the set of efficiency parameters (f) at the first steady state of travel,

- determining, using the current data-driven model, one or more candidate sets of adjustable control parameter (X) values having improved efficiency parameter (f) values compared to the current values of the set of efficiency parameters (f),

- selecting a preferred candidate set among the one or more candidate sets, and

- applying the preferred candidate set of adjustable control parameters (X) by adjusting the control levers to the values in the preferred candidate set.

2. A method in accordance with claim 1, further comprising applying the preferred candidate set subject to an improvement criterion, the improvement criterion comprises a requirement that a change in an adjustable control parameter (X) value must exceed a minimum absolute amount or exceed a minimum percentage amount or other relative amount relative to a current value of the adjustable control parameter (X) value.

3. A method in accordance with claim 1 or 2, wherein the set of one or more efficiency parameters (f) comprises a fuel efficiency parameter (f), and the current data-driven model comprises:

- a fuel flow function relating a fuel flow of the vessel to a first set of one or more of the adjustable control parameters (X), and

- a speed function relating a speed of the vessel to a second set of one or more of the adjustable control parameters (X),

and the fuel efficiency parameter (f) is determined based on the fuel flow function and the speed function.

4. A method in accordance with one of the preceding claims, further comprising:

- determining a predicted vessel performance parameter value using the current data-driven model and current values of the adjustable control parameters (X),

- determining an actual vessel performance parameter value for comparison with the predicted vessel performance parameter value, - determining that a discrepancy between the actual value and the predicted value of the vessel performance parameter exceeds a tolerance threshold and in response:

i. providing, based on the updated set of historical operating data sets, an updated data-driven model describing the one or more efficiency parameters (f) as corresponding one or more functions, each function depending on one or more of the adjustable control parameters (X) and one or more of the vessel performance parameters.

5. A method in accordance with claim 4, wherein providing the updated data-driven model comprises reducing an influence of older historical data sets relative to newer historical data sets.

6. A method in accordance with one of the preceding claims, wherein the set of one or more efficiency parameters (f) comprises one or more of: a fuel efficiency, a mission efficiency, a life cycle cost index, a motion index.

7. A method in accordance with one of the preceding claims, wherein the set of adjustable control parameters (X) comprises one or more of: a propeller speed, a propeller pitch, a power train setting, an engine part load, a commanded vessel heading, a vessel hull trim.

8. A method in accordance with one of the preceding claims, wherein the set of vessel performance parameters comprises one or more of: a fuel flow rate, an actual vessel heading, a vessel Speed over Ground (SoG), a vessel Course over Ground (CoG), a shaftline torque, a vessel roll angle dependent parameter, a vessel pitch angle dependent parameter.

9. A method in accordance with one of the preceding claims, wherein at least one of the one or more functions depends on the set of environmental parameters, and the set of environmental parameters comprises one or more of: an apparent wind speed, an apparent wind direction, a wave height, a wave direction relative to a hull of the vessel, a true wind speed, a vessel physical condition.

10. A method in accordance with one of the preceding claims when dependant on claims 3, 7 and 8, wherein determining that the vessel has reached the first steady state of travel comprises one or more of:

- determining that the speed of the vessel does not vary more than allowed by a predefined steady-state speed criterion,

- determining that the propeller speed does not vary more than allowed by a predefined steady-state propeller speed criterion,

- determining that the vessel heading does not vary more than allowed by a predefined steady-state vessel heading criterion,

- determining that the Course over Ground (CoG) does not vary more than allowed by a predefined steady-state Course over Ground (CoG) criterion, - determining that the propeller torque does not vary more than allowed by a predefined steady-state propeller torque criterion,

- determining that the fuel flow does not vary more than allowed by a predefined steady-state fuel flow criterion.

11. A method in accordance with one of the preceding claims, wherein the set of efficiency parameters (f) comprises at least two efficiency parameters (f), and wherein the step of determining one or more candidate sets comprises determining at least two non-dominated candidate sets.

12. A method in accordance with claim 11, wherein the preferred candidate is selected based on a preference of one efficiency parameter (f) in the set of efficiency parameters (f) over all other efficiency parameters (f) in the set of efficiency parameters (f).

13. Digital computing and storage hardware configured specifically to perform a method in accordance with one of claims 1-12.

14. A computer-readable storage medium comprising program instructions that, when executed on suitable digital computing and storage hardware, cause the digital computing and storage hardware to perform a method in accordance with one of claims 1-12.