Fuel supply for maritime vessels

Abstract

A fuel control system for a maritime vessel is disclosed. The control system has an interface for receiving data identifying the maritime vessel and data identifying the voyage to be undertaken by the maritime vessel. A memory is also provided for storing computer program code for implementing a predictive model and parameter values ​​for the predictive model, the parameter values ​​being derived from optimizing the predictive model using historical data from multiple maritime vessels and multiple voyages. The control system also has a processor. The processor is configured to execute the computer program code to: instantiate the predictive model using the parameter values; supply the data identifying the maritime vessel and the data identifying the voyage to be undertaken by the maritime vessel to the predictive model; and apply the predictive model to determine an estimate of the fuel consumption of the maritime vessel during the voyage. A fuel supply system determines the amount of fuel to be supplied to at least one fuel tank of the maritime vessel based on the fuel consumption estimate.

Description

Technical Field

[0001] This invention relates to the supply of fuel to maritime vessels. Specifically, this invention relates to a fuel control system and method for maritime vessels, wherein fuel consumption estimates are determined. Background Technology

[0002] Shipping vessels, such as container ships, play a vital role in transporting goods, items, and materials around the world. These vessels are typically powered by diesel or natural gas propulsion systems that consume fuel to drive one or more propellers. These propellers can be directly driven, for example, by a low-speed reciprocating diesel engine, or driven via a gearbox. Diesel engines may consume marine-grade diesel fuel or heavy fuel oil.

[0003] Refueling of shipping vessels is typically carried out under the supervision of a ship operator, such as the captain or a crew member. Fuel can be supplied to fuel tanks during the voyage, and these tanks can be refilled in port if the fuel level is low.

[0004] Modern shipping adds extra complexity to refueling ships. For example, container ships may traverse a range of destinations on round trips around the world. Containers may be exchanged at various points along the route, and conditions may vary for each segment of the round trip. This makes it difficult to estimate how much fuel needs to be loaded onto the ship for each segment. When a ship will run out of fuel is unknown. In these situations, emergency refueling at sea (sometimes called "replenishment at sea") can be arranged. This can be risky.

[0005] Therefore, there is a need to improve how we estimate the fuel consumption of maritime vessels. In some cases, there is a need to improve the way we supply fuel to maritime vessels. Summary of the Invention

[0006] According to a first embodiment, a method for use on a maritime vessel is provided. The method includes: obtaining data identifying the maritime vessel and data identifying the voyage to be undertaken by the maritime vessel; calculating a fuel consumption estimate using a predictive model configured to receive the data identifying the maritime vessel and the data identifying the voyage as input, the parameters of the predictive model being fitted using historical data from multiple maritime vessels and multiple voyages; and determining the amount of fuel to be supplied to the maritime vessel based on the fuel consumption estimate.

[0007] Fuel consumption estimates can be used to properly supply (i.e., load) fuel to maritime vessels during their voyages across one or more network segments. Fuel consumption estimates can help optimize vessel loading to reduce fuel consumption and improve vessel efficiency, thereby reducing pollution levels. Fuel consumption estimates can also help avoid emergency refueling operations, such as at-sea refueling.

[0008] Optionally, the method includes controlling the supply of a predetermined amount of fuel to the seagoing vessel.

[0009] Optionally, the data identifying the maritime vessel includes an indication of the category of the maritime vessel, and the historical data of the plurality of maritime vessels includes an indication of the category of the plurality of maritime vessels.

[0010] Optionally, the data identifying the voyage to be undertaken includes identification of segments of the voyage, the segments of the voyage having arrival and departure ports, wherein the prediction model is configured to receive the identification of the segments of the voyage as input.

[0011] Optionally, the method includes determining the distance between the port of arrival and the port of departure; and determining an estimated mean speed of the maritime vessel during the voyage, wherein the prediction model is configured to receive the distance and the estimated mean speed as input.

[0012] Optionally, the prediction model is a linear model, and the parameters of the linear model are fitted using elastic network regularization.

[0013] Optionally, the prediction model includes one or more of the following: parameters independent of the data identifying the maritime vessel; parameters dependent on the data identifying the maritime vessel; parameters independent of the data identifying the voyage to be undertaken by the maritime vessel; and parameters dependent on the data identifying the voyage to be undertaken by the maritime vessel.

[0014] Optionally, the method includes: obtaining an initial fuel consumption estimate, wherein the prediction model is configured to receive the initial fuel consumption estimate as input.

[0015] Optionally, the initial fuel consumption estimate is calculated based on data indicating how fuel consumption changes with the speed of the seagoing vessel.

[0016] Optionally, the initial fuel consumption estimate is received as user input.

[0017] Optionally, the maritime vessel is configured to consume multiple types of fuel, and the calculation of fuel consumption estimates using a predictive model includes calculating fuel consumption estimates for the multiple fuel types.

[0018] Optionally, calculating the fuel consumption estimates for the multiple fuel types includes: calculating a first fuel consumption estimate for a first fuel type using a first prediction model; calculating a second fuel consumption estimate for a second fuel type using a second prediction model; and determining the fuel quantity includes determining the amount of each of the first and second fuel types to be supplied to the maritime vessel based on the corresponding first and second fuel consumption estimates.

[0019] Optionally, the method includes: obtaining an initial fuel consumption estimate for each of the first fuel type and the second fuel type, wherein the first prediction model is configured to receive the initial fuel consumption estimate for the first fuel type as input, and wherein the second prediction model is configured to receive the initial fuel consumption estimate for the second fuel type and an indication of whether the initial fuel consumption estimate for the second fuel type is non-zero as input.

[0020] According to another embodiment, a fuel control system for a maritime vessel includes: an interface for receiving data identifying the maritime vessel and data identifying the voyage to be undertaken by the maritime vessel; a memory for storing computer program code for implementing a predictive model and parameter values ​​of the predictive model derived from historical data of multiple maritime vessels and multiple voyages; and a processor for retrieving the computer program code and the parameter values ​​from the memory. The processor is configured to execute the computer program code to: instantiate the predictive model using the parameter values; supply the data identifying the maritime vessel and the data identifying the voyage to be undertaken by the maritime vessel to the predictive model; and apply the predictive model to determine an estimated fuel consumption of the maritime vessel during the voyage. The fuel control system further includes a fuel supply system for determining the amount of fuel to be supplied to at least one fuel tank of the maritime vessel based on the estimated fuel consumption.

[0021] According to another embodiment, a non-transitory computer-readable storage medium stores instructions that, when executed by a processor, cause the processor to: obtain data identifying a maritime vessel and data identifying a voyage to be undertaken by the maritime vessel; determine the category of the maritime vessel based on the data identifying the maritime vessel; determine a voyage segment identifier, a voyage duration, and a voyage speed based on the data identifying the voyage; obtain an initial fuel consumption estimate for each of a first fuel type and a second fuel type, the initial fuel consumption estimate being based on the fuel consumption rate of the maritime vessel at the voyage speed and the voyage duration; and apply a first prediction model to output an adjusted fuel consumption estimate for the first fuel type, the first prediction model being the initial fuel consumption estimate for the first fuel type, the category of the maritime vessel, and the voyage speed. The first prediction model is a parameterized function of the segment identifier, the duration of the voyage, and the speed of the voyage, the parameters of which are optimized based on historical data of multiple shipping vessels and multiple voyages; a second prediction model is applied to output an adjusted fuel consumption estimate for the second fuel type, the second prediction model being a parameterized function of the initial fuel consumption estimate for the second fuel type, the category of the shipping vessel, the segment identifier, the duration of the voyage, and the speed of the voyage, the parameters of which are optimized based on the historical data; the amount of fuel of the first fuel type to be supplied to the shipping vessel is determined based on the adjusted fuel consumption estimate for the first fuel type; and the amount of fuel of the second fuel type to be supplied to the shipping vessel is determined based on the adjusted fuel consumption estimate for the second fuel type. Attached Figure Description

[0022] Embodiments of the invention will now be described by way of example only with reference to the accompanying drawings, in which:

[0023] Figure 1 A schematic diagram of a maritime vessel based on an example is shown;

[0024] Figure 2 A schematic diagram of a fuel control system for a maritime vessel, based on an example, is shown.

[0025] Figure 3 A schematic diagram of a maritime vessel using various fuel types is shown, based on an example.

[0026] Figure 4 A schematic diagram is shown, based on an example, of the variables that can be used to generate fuel consumption estimates;

[0027] Figure 5 A graph illustrating exemplary data showing how fuel consumption varies with the speed of a maritime vessel is shown;

[0028] Figure 6 A flowchart illustrating a method for refueling a maritime vessel, based on an example, is shown; and

[0029] Figure 7 A computer-readable medium according to the example storage instructions is shown. Detailed Implementation

[0030] Some examples described herein provide methods and systems for estimating the amount of fuel needed for a voyage undertaken by a maritime vessel. The estimated fuel consumption can be used to control, for example, the amount of fuel supplied to the maritime vessel during refueling or bunkering operations. Some examples provide improved estimates of maritime vessel fuel consumption based on comparative methods. Accurate estimates of fuel consumption enable optimal fuel loading for maritime vessels, such as supplying sufficient fuel to complete the voyage, without oversupply that could negatively impact the vessel's load and performance.

[0031] Some of the examples described in this paper enable the determination of fuel consumption estimates for different fuel types. This can allow for improvements in the supply of multiple fuel types for a voyage, where different fuel types may be suitable for different parts of the voyage. Some examples apply predictive models where the parameters of the predictive model can be fitted using historical data from multiple shipping vessels and multiple voyages. The predictive model may include multiple parameters that control the calculations applied to a set of inputs. The predictive model may take into account the characteristics of a particular shipping vessel and a particular voyage.

[0032] Figure 1 An example of a refueling operation on a maritime vessel 110 is shown. Although this example is described with reference to refueling, it can also be applied to initial refueling operations, such as for new or converted maritime vessels. In one case, the maritime vessel 110 may include a container ship. The maritime vessel 110 may be designed to transport one or more of cargo, articles, and materials from one port to another.

[0033] exist Figure 1In this example, the maritime vessel 110 includes a fuel control system 120. The fuel control system 120 may include electromechanical systems mounted on the maritime vessel 110 to control the supply of fuel to the maritime vessel 110. The maritime vessel 110 also includes a fuel tank 130, an engine 140, and a propeller 150. The fuel tank 130 is arranged to store fuel for the voyage. The fuel tank 130 is arranged to supply fuel to the engine 140 during the voyage to power the engine 140. The engine 140 is configured to consume fuel from the fuel tank 130 and rotate the propeller 150 to propel the maritime vessel 110 through a large body of water, such as the sea or a waterway. Although a propeller 150 is used in this example, other propulsion mechanisms may be used alternatively. The engine 140 may include a diesel engine, a gas turbine, or a liquefied natural gas (LNG) engine, etc. If the maritime vessel 110 includes a container ship, the engine 140 may include a reciprocating diesel engine. The reciprocating diesel engine may be a two-stroke or four-stroke engine. In other examples, more than one thruster 150 may be present. In these examples, each thruster may have one engine, or one engine may power multiple thrusters. Similarly, in other examples, more than one engine may be provided, each coupled to one or more thrusters. In some cases, a backup engine and a main engine 140 may be provided. The fuel supplied to and stored in the fuel tank 130 may depend on the engine 140 used. For a diesel engine, the fuel may include diesel fuel or fuel oil (including “heavy” and “residual” fuel oil). For an LNG engine, the fuel may include liquefied natural gas. For a gas turbine, the fuel may include petroleum. Other possible fuels include liquefied hydrogen. In one case, the fuel may include charge stored by a battery used to power an electric motor.

[0034] Figure 1 A shipping vessel 110 is shown docked at a refueling station 160. The refueling station may be located at a port or other destination accessible to the shipping vessel 110. The refueling station 160 includes a fuel tank 170 arranged to store fuel for supply to the shipping vessel. The fuel tank 170 is connectable to the fuel tank 130 (indicated by dashed lines) of the shipping vessel 110. During refueling operations, the fuel tank 170 can be connected to the fuel tank 130 to allow fuel to be supplied from the fuel tank 170 to the fuel tank 130. Figure 1In this vessel 110, the amount of fuel supplied from fuel tank 170 to fuel tank 130 is controlled by the fuel control system 120. For example, the fuel control system 120 can transmit values ​​to the refueling station 110, and the refueling mechanism, which forms part of fuel tank 170, can measure the amount of fuel supplied to fuel tank 130, for example, using a flow measurement sensor, and stop supplying fuel to fuel tank 130 when the amount of fuel supplied is equal to the value (or at least within a threshold amount of the value).

[0035] Figure 2 It shows what can be used for implementation Figure 1 The fuel control system 120 shown is a fuel control system 200. The fuel control system 200 includes an interface 210, a memory 220, a processor 230, and a fuel supply system 240. The fuel control system 200 can be implemented for use on marine vessels (such as...). Figure 1 The fuel control system 200 is part of the control system of a marine vessel (110). In one case, the fuel control system 200 may include an embedded computing system installed on the marine vessel.

[0036] Interface 210 may include an internal or external interface to the fuel control system 200. Figure 2 In this diagram, interface 210 is shown as an external interface. Interface 210 may include a network interface arranged to receive data via a wired or wireless network. Interface 210 may additionally or alternatively include a user interface for receiving data from a user. For example, data from a user may include data input by the user and / or selections of data stored in memory. If interface 210 includes an internal interface, it may provide an interface to a data storage device, such as a hard disk drive or solid-state storage. Figure 2 Interface 210 is communicatively coupled to processor 230. Interface 210 is configured to receive data identifying a maritime vessel and data identifying the voyage to be undertaken by the maritime vessel. The maritime vessel can be... Figure 1 The illustrated marine vessel 110 is, for example, a marine vessel equipped with a fuel control system 200. Although interface 210 is... Figure 2 The interface is shown as a single interface, but it can include two separate interfaces, each receiving either data identifying the vessel or data identifying the voyage. For example, the data identifying the vessel can be received via a data interface that couples the fuel control system 200 to the vessel's data storage device, and the data identifying the voyage can be selected from a list of possible voyages on the user interface.

[0037] exist Figure 2In this example, memory 220 stores computer program code to implement a predictive model. The predictive model may include machine learning models, including linear models, nonlinear models (including logistic regression models), neural network architectures, support vector machines, conditional random fields, Bayesian belief networks, and decision trees, etc. The computer program code may include machine learning libraries defined in one or more programming languages, including R, Python, Java, C++, and JavaScript. The computer program code may define out-of-the-box parameterizable models and / or may define the coupling and arrangement of low-level components. In this example, memory 220 also stores parameter values ​​for the predictive model. These may include weight values, bias values, hyperparameters, probability values, and parameters of one or more probability distributions, etc. Parameter values ​​may be used in conjunction with the computer program code defining the predictive model to instantiate the predictive model, i.e., to provide a specific configuration instance of the predictive model. Although in Figure 2 The diagram shows a memory 220, but in some cases, memory 220 may include two separate memories, one for storing computer program code (e.g., erasable read-only firmware memory) and the other for storing parameter values ​​(such as random access memory).

[0038] The predictive model is implemented using computer program code and parameter values ​​from memory 220. Processor 230 is configured to retrieve the computer program code and parameter values ​​from memory 220 and execute the computer program code to determine fuel consumption estimates for maritime vessels and voyages, for example, using data identified at interface 210. Processor 230 is configured to execute the computer program code to instantiate the predictive model first using the parameter values. For example, this may include loading weight values ​​for a linear model or values ​​of the weight and bias matrices in a neural network architecture. The parameter values ​​are derived from optimizing the predictive model using historical data from multiple maritime vessels and multiple voyages. Depending on the predictive model used, optimization may include a training or fitting phase where parameter values ​​are determined by minimizing the cost or loss function of data items within the historical data. The training or fitting phase is performed before determining the fuel consumption estimates. In one case, the training or fitting phase may be performed offline and / or at a remote computing device, for example, by retrieving parameter values ​​via a network and loading them into memory 220. The training or fitting phase may be performed periodically to update the parameter values ​​based on new data.

[0039] Once the processor 230 instantiates the prediction model, it executes computer program code to supply the prediction model with data identifying shipping vessels and data identifying the voyages the vessels are to undertake. For example, data identifying shipping vessels may include or be mapped to numerical values ​​representing shipping vessels in a set of shipping vessels (e.g., via array indexing, etc.). Data identifying voyages may similarly include or be mapped to numerical values ​​representing voyages in a set of possible voyages. In another case, data identifying voyages may include data identifying the properties of the voyage, or data used to retrieve these properties during preprocessing operations. For example, the properties of a voyage may include one or more of the following: voyage distance, voyage duration, normal voyage speed, network segments (e.g., in a multi-segment journey), port of arrival, and port of destination, etc. These properties may be provided as numerical values ​​(e.g., distance may include distance values ​​in kilometers, or network segments may be identified by integer array indexing). Depending on the configuration of the prediction model, the term "voyage" may refer to one or more network segments.

[0040] The predictive model can be configured to receive an array or an N-dimensional data structure as input. The array may include a multidimensional array. The predictive model can implement one or more functions in its application to determine an estimate of a maritime vessel's fuel consumption during a voyage. The predictive model can provide a fuel consumption estimate as output. The fuel consumption estimate can be a continuous value or an integer value (e.g., in the former case, a floating-point data type) indicating the amount of fuel the model predicts will be consumed during the voyage. The fuel consumption estimate can be provided as a quantity in tons, liters, or US gallons. This output can be stored in volatile or non-volatile memory and / or communicated with other components via a system bus.

[0041] exist Figure 2 In the example, the fuel consumption estimate calculated by the processor 230 using a predictive model is accessible to the fuel supply system 240. The fuel supply system 240 is configured to determine the amount of fuel to be supplied to at least one fuel tank of a maritime vessel based on the fuel consumption estimate. For example, the fuel supply system 240 can send data indicating the fuel consumption estimate to... Figure 1 The fuel station 160 shown is configured to supply fuel from fuel tank 170 to fuel tank 130 based on fuel consumption estimates. Alternatively, the fuel supply system 240 may control a fluid flow system (e.g., one or more valves) that controls the flow of fuel to... Figure 1 The fuel supply to fuel tank 130 is provided. If the gas station 160 is a conventional gas station, the fuel supply system 240 can provide fuel consumption estimates via a user interface, enabling manual control of the fuel supply based on the fuel consumption estimates.

[0042] In some cases, historical data used to optimize the predictive model may include data indicating fuel consumption for multiple shipping vessels over multiple voyages. These multiple shipping vessels may or may not include those associated with this fuel control system 200. If the shipping vessel associated with this fuel control system 200 is new or a converted vessel, the predictive model may be constructed to ignore function terms associated with that vessel; for example, in some cases, any contribution to the predictive model based on a specific shipping vessel may be effectively ignored. In these cases, function terms associated with shipping vessels in general, independent of any particular shipping vessel, may be retained. In some cases, data identifying new or unseen shipping vessels may be processed by the predictive model to have a incidental effect on any fuel consumption estimates.

[0043] Although reference Figure 1 Fuel control system 120 describes fuel control system 200, but in one case, it can be implemented outside of the shipping vessel 110. For example, fuel control system 200 can be part of a control system installed as part of a refueling station. In yet another case, fuel control system 200 can include a remotely located control system configured to transmit data to one or more of the shipping vessel 110 and refueling station 160 to arrange the amount of fuel supplied to the shipping vessel. For example, interface 210 can receive data from the shipping vessel 110 via a wireless communication link and issue a fuel consumption estimate processed by fuel supply system 240. One or more of the shipping vessel 110 and refueling station 160 can include a control system with a corresponding interface to receive the fuel consumption estimate and arrange the supply of fuel based on the estimate. In one case, Figure 2 Some components of the fuel control system 200 shown may be distributed among two or more of the following: the marine vessel 110, the refueling station 160, and the remote data processing station. For example, the fuel supply system 240 may be located locally on the marine vessel 110 or at the refueling station 160.

[0044] and Figure 1 and Figure 2Fuel control systems, as illustrated in one or more examples, enable the supply of accurate amounts of fuel to maritime vessels. Accurate estimates can be provided by using predictive models trained on historical data, reducing the risk of vessels running out of fuel during voyages while also avoiding over-fueling, which could otherwise lead to reduced carrying capacity or inefficiency. For example, if a maritime vessel with a fixed capacity carries more fuel than required for the voyage, it may consume more fuel. Therefore, the fuel consumption estimates provided by this example allow maritime vessels to operate more efficiently and reduce pollution and fuel consumption. By preventing vessels from running out of fuel at sea, risky resupply operations at sea can be avoided, thus improving crew safety. Accurate fuel estimates are particularly useful for large container ships (e.g., up to 10,000 or more 20-foot equivalent units - TEUs). For example, a large container ship can consume 10,000 to 14,000 liters of heavy fuel oil per hour, with the fuel weight per hour potentially ranging from 10 to 15 tons (10-15 x 10). 3 Between kg).

[0045] Some of the examples described in this article can be applied to provide fuel consumption estimates for multiple fuel types. Figure 3 Example 300 is provided, which is Figure 1 An adjustment to the example, wherein the marine vessel 310 includes a fuel control system 320 that determines fuel consumption estimates for a first fuel type and a second fuel type. In this example, the marine vessel 310 has fuel tanks for each fuel type: a first fuel tank 330 for storing the first type of fuel and a second fuel tank 335 for storing the second type of fuel. Figure 3 In one example, either type of fuel can be used to power engine 340 to drive propeller 350. In other examples, each fuel type can have a separate engine.

[0046] exist Figure 3 In this example, gas station 360 includes fuel tanks for each fuel type. A first fuel tank 330 is connectable to a first fuel tank 370 supplying the first fuel type, and a second fuel tank 335 is connectable to a second fuel tank 375 supplying the second fuel type. In other examples, the gas station may have fuel tanks supplying only a subset of fuel types (e.g., a fuel tank supplying the second fuel type but not the first). In some other examples, a single fuel tank may have separate compartments for different fuel types. Figure 3In this example, when the shipping vessel 310 is docked at the refueling station 360 (as shown by the dashed line), the first fuel tank 370 and the second fuel tank 375 can be connected to the first fuel tank 330 and the second fuel tank 335, respectively. In this example, while docked, the fuel control system 320 can communicate with the refueling station 360, for example, via wired or wireless communication channels, including communication with the refueling facilities associated with the first fuel tank 370 and the second fuel tank 375.

[0047] The fuel control system 320 can be connected with Figure 1 and Figure 2 The fuel control systems 120 and 200 are implemented in a similar manner. Figure 3 In the example, the fuel control system 320 is configured to calculate fuel consumption estimates for multiple fuel types, including a first fuel type and a second fuel type. In one case, a common prediction model can be used for both fuel types. In this case, the common prediction model may have different parameters for each fuel type, resulting in different instantiated models for each fuel type. Alternatively, in this case, the common prediction model can implement a joint model and provide a multidimensional output representing fuel consumption estimates for multiple fuel types. In other cases, different prediction models may, for example, use different neural network architectures or different model components for different fuel types. In both cases, a first fuel consumption estimate is generated for the first fuel type and a second fuel consumption estimate is generated for the second fuel type. These two fuel consumption estimates can be used, for example, in conjunction with those for... Figure 1 The method described for a single fuel type is similar to the method used to determine the amount of fuel of different fuel types supplied from the corresponding fuel tanks 370, 375 to the corresponding fuel tanks 330, 335.

[0048] In some cases, if two prediction models are used for each corresponding fuel type, the output of one prediction model can be used as the input to the other. In one scenario, the expected proportion of each fuel type can be supplied as input to the prediction model. This can be used as a constraint in the prediction model. In some cases, the nature of the shipping vessel and / or the voyage to be undertaken can be used as input to one or more prediction models. These, in turn, can be used as constraints in the prediction model. For example, fuel tanks 330 and 335 may have different capacities, or different fuel types may be consumed by engine 340 (or multiple different engines) at different rates. Similarly, a fuel type may not be suitable for use at least part of the voyage; for example, the voyage may have a route through canals (such as the Suez Canal or the Panama Canal), where certain fuel types may be unsuitable due to space constraints.

[0049] In one case, different fuel types may include immiscible fuel types, such as liquefied natural gas and heavy fuel oil. In other cases, different fuel types may be mixed, such as different grades of heavy fuel oil, or one fuel type may have additives for regular engine maintenance while another fuel type does not. Mixable fuel types may be mixed in the fuel tank, in the engine feed, and / or in the engine itself. In one case, a first fuel type includes fuel oil having a first chemical composition, such as a first level of sulfur, and a second fuel type includes fuel oil having a second chemical composition, such as a second level of sulfur, wherein the first level differs from the second level.

[0050] In some cases, calculations by the fuel control system 320 are used to provide fuel estimates across multiple voyages (or across a voyage consisting of multiple segments), enabling the supply of sufficient fuel of each fuel type for the voyage. This can be useful when a particular fuel type is not available at one or more final destinations.

[0051] Figure 4 Example 400 of a voyage that can be undertaken by a maritime vessel is shown. Example 400 will be used to describe certain inputs that can be provided to a predictive model. In some cases, variables may be provided in one form, and the computation of variables in different forms may implicitly derive from a function applied by the predictive model, and / or may derive from one or more preprocessing stages of the predictive model. According to other examples, Figure 4 These are schematic diagrams provided to illustrate certain concepts related to this implementation method, with necessary simplification.

[0052] Figure 4 The example illustrates a voyage comprising four network segments: L1, L2, L3, and L4. Network segment L1 extends between departure port A and arrival port B; network segment L2 extends between departure port B and arrival port C; network segment L3 extends between departure port C and arrival port D; and network segment L4 extends between departure port D and arrival port A. The voyage described is a round trip, meaning the shipping vessel begins and ends at port A. Although the term "port" is used in this example, the departure and / or arrival locations do not necessarily have to be ports in the conventional sense; rather, they can include, for example, offshore drilling rigs, mooring positions, berthing positions, and / or refueling stations independent of any cargo loading or unloading. In one case, the shipping vessel may include a cargo ship, and thus cargo may be loaded and / or unloaded at one or more of ports A through D.

[0053] Each network segment can be identified using a unique identifier from a predefined list, such as global and / or local routes. Each network segment can be considered as part of a larger journey or the journey itself. Each network segment L i With associated flight segment distance di In this example, each network segment L i It also has an estimated duration t i Segment distance can be the distance between ports, such as the distance between a port of departure and a port of arrival. Segment distance can include the distance between geographical points and / or the distance of a specific route between two points. For example, if a particular network segment must avoid certain territories or areas due to known geographical features (such as rocky outcrops or islands) and / or piracy risks, then the segment distance d... i It may be longer than the distance between points. Segment distance can be provided as a numerical value indicating distance in kilometers. In some cases, segment distance can be retrieved from local or remote data storage devices based on the network segment identifier. Segment distance can be updated over time, for example, retrieved from a remote repository of updated segment distances. Segment duration can be an estimate of the time required to complete a network segment.

[0054] Figure 4 Two shipping vessels are shown making a round trip. There is a first shipping vessel in a first shipping vessel category VC1 and a second shipping vessel in a second shipping vessel category VC2. Shipping vessel categories can include a group of shipping vessels or vessels with a common design. Shipping vessel categories can be distinguished from shipping vessel types, which can indicate capacity and / or intended use. Data identifying shipping vessels can indicate one or more of the shipping vessel categories and types. Shipping vessel categories can be set by a classification society and can conform to one or more standards, such as those defined by the International Maritime Organization (IMO). Predefined numbers or codes can be assigned to shipping vessels to identify categories. One or more shipping vessel categories and types can be indicated by codes or integer indices from a predefined list of categories or types. The input and historical data for the predictive model use a consistent set of categories and / or types. Using indications of shipping vessel categories and / or types allows the predictive model to model category and / or type-specific factors. This can improve the accuracy of fuel consumption estimates. If the shipping vessel does not involve any pre-existing category or type, categories and / or types indicating "unknown" or "other" categories can also be provided. The type of shipping vessel can affect the distance and speed processed by the predictive model, for example, it may generate parameters that model the specific interaction between shipping vessel type and distance and / or speed.

[0055] Figure 4 The diagram also shows two shipping vessels with specified speeds s1 and s2. The specified speed can be a suggested average speed for a network segment. It can be provided as an independent variable or set based on the category or type of shipping vessel. It can also be determined by dividing the segment distance by the segment duration, i.e., d. i / t iIn some cases, the speed s i Distance d i and duration t i One or more of these can be retrieved based on a lookup operation on a set of historical data, which is based on data identifying shipping vessels and / or data identifying voyages. For example, speed s i Distance d i and duration t i Retrieval can be based on average values ​​of specific identified shipping vessels and network segments. In some cases, only speeds can be retrieved. i Distance d i and duration t i Both of them, and a third variable can be calculated accordingly.

[0056] Marine Vessel Category VC i Network segment L i Speed ​​s i Distance d i and duration t i One or more of these can be provided as numerical inputs to the predictive model to calculate fuel consumption estimates for a specific maritime vessel and network segment. Fuel consumption estimates for each individual network segment can be combined to provide estimates for round trips or different segments of the journey.

[0057] In one scenario, the predictive model may also receive an initial fuel consumption estimate. In this case, the initial fuel consumption estimate can be a "coarse" estimate refined by the predictive model. This initial fuel consumption estimate can be provided, for example, by a member of the crew or refueling personnel based on experience. In another scenario, the initial fuel consumption estimate can be provided by initial calculations (e.g., the output of a conventional fuel control system). Conventional fuel control systems can operate using data from indicated speed curves or fuel gauges.

[0058] Figure 5 A graph 500, as shown in the example, indicates how a conventional fuel control system can calculate an initial fuel consumption estimate. Graph 500 shows data 510 indicating how the fuel consumption rate (e.g., in million tons per day) can vary with speed (in knots). Data 510 may represent recorded and / or modeled data. A curve 520 can be fitted to data 510. Graph 500 may be unique for a specific maritime vessel or maritime vessel class. In one case, the relationship shown by graph 500 may be provided as a data table, etc., as part of the specification for a specific maritime vessel or maritime vessel class. A conventional fuel control system can determine a specified speed for a voyage (e.g., from data indicating the relationship shown in graph 500) by retrieving data indicating the relationship shown in graph 500. Figure 4 s iThe initial fuel consumption estimate is calculated by multiplying the fuel consumption rate by the voyage duration (in days). This initial fuel consumption estimate can then be fed as a numerical value into the predictive model. Similarly, historical data used to train the predictive model can also include fuel consumption estimates for recorded voyages from conventional fuel consumption systems. Used in this way, the fuel control system described herein can be retrofitted to upgrade existing conventional fuel control systems, for example, those provided on older marine vessels.

[0059] In one scenario, if a conventional fuel control system outputs initial fuel consumption estimates for multiple fuel types, the predictive model may include flags or other input variables indicating whether one or more types of fuel will be consumed for a particular network segment.

[0060] In some cases, a linear model can be used to implement the prediction model, which operates on one or more of the following: vessel speed, network segment distance, and initial fuel consumption estimate. In this case, the parameter values ​​can provide one or more of the following: a deviation term independent of vessel category and network segment, a deviation term dependent on vessel category, a deviation term dependent on network segment, a weight term for speed independent of vessel category and network segment, a weight term dependent on vessel category, a weight term for distance independent of vessel category and network segment, a weight term dependent on vessel category, a weight term for initial fuel consumption estimate independent of vessel category and network segment, a weight term dependent on initial fuel consumption estimate dependent on network segment, a weight term for the product of speed and distance independent of vessel category and network segment, and a weight term dependent on the product of speed and distance dependent on vessel category. Predictive models for specific fuel types can include fuel type-specific parameter values. In some cases, non-zero initial fuel consumption estimates from conventional fuel control systems can be used as binary variables, where one or more weights and / or bias terms are provided for one or more of the binary variable and / or a specific item for a maritime vessel category or network segment. The binary variable can be used, for example, as an indicator for a prediction model for a second fuel type, allowing parts of the prediction model to be turned on and off.

[0061] Figure 6 A method 600 for refueling a maritime vessel according to an example is shown. Method 600 includes three boxes 610, 620 and 630.

[0062] At box 610, data identifying the shipping vessel and the voyage to be undertaken by the shipping vessel is obtained. This may include, for example, receiving or retrieving a shipping vessel identifier or category identifier. The identifier may be a unique code, a string label mapped to an integer value, or an embedded multidimensional array representing the shipping vessel. The data identifying the voyage to be undertaken may include identifiers of one or more network segments of the route. Data may be received and / or selected using a user interface, received via a network connection, and / or retrieved from storage devices or memory.

[0063] At box 620, a fuel consumption estimate is calculated using a predictive model. The predictive model is configured to receive data identifying the shipping vessel and data identifying the voyage to be undertaken as input. If the data identifying the voyage includes multiple network segment identifiers, box 620 can be repeated for each network segment. In this example, historical data from multiple shipping vessels and multiple voyages are used to fit the parameters of the predictive model. For example, the predictive model may include a linear model, a logistic regression model, or a neural network architecture, and may be trained on historical data. Historical data may include multiple data samples, each providing at least a triplet of data identifying the historical shipping vessel, data identifying network segments that have been traversed in the past, and data indicating the recorded fuel consumption of the historical shipping vessel in the traversed network segments. The term "historical" is used here to refer to data at a past point in time; it may include shipping vessels still in operation and network segments.

[0064] At box 630, the amount of fuel to be supplied to the maritime vessel is determined based on a fuel consumption estimate. This may include controlling fluid supply equipment such as pumps and valves to supply a given amount of fuel to the maritime vessel. It may also include transmitting the fuel consumption estimate and / or fuel quantity to a remote system to instruct refueling or bunkering operations.

[0065] Method 600 can be repeated for fleets spanning cross-ship networks. Method 600 can be repeated for multiple voyages within a predefined time period. This set of fuel consumption estimates can be used to control fuel tanks and refueling stations, as well as refueling of maritime vessels. This set of fuel consumption estimates can be stored and compared with actual measured fuel consumption of maritime vessels. The predictive model can be updated as more data is received, for example, by continuous or periodic retraining to incorporate new data and update parameter values.

[0066] In some cases, data identifying shipping vessels includes indications of their categories, and historical data for multiple shipping vessels may include indications of multiple categories. For example, a fleet might contain shipping vessels of 10 categories. These categories can be identified using IMO codes or numbers. A list of available categories can be generated, and each string or numbered category can be mapped to an integer representing an index in the list. The predictive model can be configured to receive integer values ​​representing shipping vessel categories as input, such as as a dimension of an input array.

[0067] In some cases, the data identifying a voyage includes the identifiers of the voyage segments. These segments, or network segments, may have associated arrival and departure ports, such as... Figure 4 Example 400 illustrates this. Similar to the above, a list of network segments can be compiled, such as representing portions of known flight routes. This can be presented as a list, where a particular network segment, identified, for example, by a string or numeric identifier, can be mapped to an integer representing an index in the list. The integer representation of the network segments can then be fed as input (e.g., as a dimension of an input array) to the prediction model.

[0068] In some cases, the data identifying a voyage includes or is used to retrieve or calculate one or more of the following: the distance between the port of arrival and the port of departure, and the estimated average speed of the shipping vessel during the voyage. These can be supplied as continuous data values ​​(e.g., as floating-point or long data types). They can be supplied as input to a predictive model, for example, each forming one dimension of an input array.

[0069] In one scenario, the predictive model includes one or more of the following: parameters independent of the data identifying the shipping vessel; parameters dependent on the data identifying the shipping vessel; parameters independent of the data identifying the voyage to be undertaken by the shipping vessel; and parameters dependent on the data identifying the voyage to be undertaken by the shipping vessel. In one case, the predictive model is a linear model, and the parameters of the linear model are fitted using elastic network regularization. In another case, the predictive model is a neural network architecture, and the parameter values ​​are fitted using backpropagation and stochastic gradient descent. Elastic network regularization can be used to remove unimportant predictor variables, where interactions between input variables are taken into account. Variables removed through elastic network regularization may be specific to each combination of shipping vessel category and network segment. In practice, this can allow for the construction of custom models for each combination of shipping vessels and network segments.

[0070] In one instance, the method includes obtaining an initial fuel consumption estimate prior to block 620. This may include estimates from a conventional fuel control system (e.g., a reference). Figure 5(as described) and / or estimates from crew members (e.g., supplied via a user interface). In this case, the predictive model can be configured to receive initial fuel consumption estimates as input, for example, as another dimension of the input array.

[0071] In one instance, box 620 includes calculating fuel consumption estimates for multiple fuel types that can be used on a maritime vessel. This may include calculating a first fuel consumption estimate for a first fuel type using a first prediction model and calculating a second fuel consumption estimate for a second fuel type using a second prediction model. Alternatively, a common prediction model may be arranged to output a two-dimensional array, the array including estimates for each fuel type as each array element (e.g., [FC1, FC2]). In one instance, box 630 includes determining the amount of each of the first and second fuel types to be supplied to the maritime vessel based on the corresponding first and second fuel consumption estimates, for example, by referring to... Figure 3 As described. If a second prediction model is used, it can be configured to receive an initial fuel consumption estimate for a second fuel type and an indication of whether the initial fuel consumption estimate for the second fuel type is non-zero as input. In some cases, prediction models for one or more fuel types may include rigid constraints such that if, for example, the initial fuel consumption estimate from a conventional fuel control system is non-zero, the prediction model outputs only a non-zero fuel consumption estimate. Prediction models for each fuel type may have separate sets of parameter values; for example, in this case, parameter values ​​are not shared between the first and second prediction models. In other cases, parameter sharing may exist.

[0072] In one scenario, the input array for the predictive model may include [VC, L, s, d, LFC], where VC is an integer value representing the vessel category, L is an integer value representing the network segment, s is a numerical value representing the vessel speed, d is a numerical value indicating the inter-port distance of the network segment, and LFC is a conventional or initial fuel consumption estimate for a specific fuel type. Therefore, in one implementation, the input to the predictive model may include a 1x5 array. In some cases, multiple conventional fuel consumption estimates may be provided for each fuel type. The output of the predictive model may include a single numerical value representing a fuel consumption estimate and / or an array of numerical values ​​representing fuel consumption estimates for a set of fuel types. Method 600 may include processing one or more of the data obtained at box 610 and historical data into a defined input format (e.g., a 1x5 array as described above).

[0073] Figure 6 Method 600 can be generated by a processor (such as...) Figure 2The method 600 is implemented by a processor 230. The method 600 may be instructed by instructions stored in a non-transitory computer-readable storage medium such as memory 220. These instructions can be executed by the processor to perform the method.

[0074] Figure 7 Another example 700 of processor 710 is shown, the processor being arranged to execute instructions stored on a non-transitory computer-readable storage medium 720. Storage medium 720 may include volatile or non-volatile memory, such as random access memory. Storage medium 720 may alternatively include non-volatile data memory, such as hard disk drive or solid-state memory. In one embodiment, processor 710 and storage medium 720 may form part of a fuel control system for a marine vessel. In this case, processor 710 and storage medium 720 may form part of an embedded computing system. In another embodiment, processor 710 and storage medium 720 may form part of a fuel control system implemented in a ground control center.

[0075] Instruction 725 instructs processor 710 to obtain data identifying a maritime vessel and data identifying the voyage to be undertaken by the maritime vessel. This may include retrieving the maritime vessel identifier. One or more of the data identifying the maritime vessel and the data identifying the voyage to be undertaken by the maritime vessel may be retrieved from storage medium 720 or another memory. In one case, these instructions may include obtaining a data array comprising tuples having [data_vessel, data_voyage]. Instruction 730 instructs processor 710 to determine the category of the maritime vessel based on the data identifying the maritime vessel. In one case, the data identifying the maritime vessel may include the category of the maritime vessel; in another case, the data identifying the maritime vessel may be used, for example, to retrieve the category of the maritime vessel using an application programming interface (API) call. Instruction 735 instructs processor 710 to determine the voyage segment identifier, voyage duration, and voyage speed based on the data identifying the voyage. Again, in one case, the data identifying the voyage may include this data, or in another case, the data identifying the voyage may be used to retrieve or calculate this data. Instruction 740 instructs processor 710 to obtain an initial fuel consumption estimate for each of a first fuel type and a second fuel type. Initial fuel consumption estimates can be based on the fuel consumption rate of a maritime vessel at its range speed and the duration of its voyage. In one case, they can be based on... Figure 5 The data shown is similar to the data used to calculate the initial fuel consumption estimate, where different groups of data are provided for each fuel type.

[0076] Execution of instructions 725 through 740 may result in the generation of one or more input arrays for the execution of subsequent instructions. In one case, the input arrays may include data such as [VC, L, s, d, LFC] as described above. Separate input arrays may be provided for each fuel type, where the initial entries in the arrays are common to the first input array (e.g., VC, L, s, and d).

[0077] Instruction 745 instructs processor 710 to apply a first prediction model to output an adjusted fuel consumption estimate for a first fuel type. In this example, the first prediction model is a parameterized function (e.g., an input array as described above) of the initial fuel consumption estimate for the first fuel type, the shipping vessel category, the voyage segment identifier, the voyage duration, and the voyage speed. In one case, any string or numeric identifier can be mapped to an integer value, also as described above. The first prediction model includes a set of parameters. These parameters are optimized based on historical data from multiple shipping vessels and multiple voyages. For example, the values ​​of the parameters may be derived from training the prediction model on historical data. Historical data may include data samples having an input array (e.g., forming a portion of array X) with a form similar to that described previously and recorded fuel consumption for the first fuel type (e.g., y = [FC1]). The data samples may be batch-processed for training. Training may include training the prediction model on a large number of data samples (e.g., at least 1000, at least 10,000, or more samples). In one scenario, during training, the first prediction model can be configured to receive a multidimensional array or matrix X of size 5 times the number of data samples (per batch) and a “foundational truth” output vector Y of size 1 times the number of data samples (per batch).

[0078] Instruction 745 instructs processor 710 to apply a second predictive model to output an adjusted fuel consumption estimate for the second fuel type. This can include operations similar to those of instruction 750. The second predictive model is a parameterized function of the initial fuel consumption estimate for the second fuel type, the maritime vessel category, the voyage segment identifier, the voyage duration, and the voyage speed. The parameters of the second predictive model are also optimized based on historical data. The format of the historical data can be similar to that used for the first predictive model, but with the initial fuel consumption estimate for the second fuel type and the "true base" fuel consumption measurement for the second fuel type (e.g., y = [FC2] instead of the first fuel type).

[0079] After executing instructions 745 and 750, processor 710 can access fuel consumption estimates for each fuel type during the voyage. These estimates may be for specific network segments within a route. Instruction 755 instructs processor 710 to determine the amount of fuel of the first fuel type to be supplied to the maritime vessel based on the adjusted fuel consumption estimate for the first fuel type. In some cases, this instruction may also include indicating the supply of fuel of the first fuel type. Instruction 760 instructs processor 710 to determine the amount of fuel of the second fuel type to be supplied to the maritime vessel based on the adjusted fuel consumption estimate for the second fuel type. In some cases, this instruction may also include indicating the supply of fuel of the second fuel type. Indicating the supply of fuel may include sending data indicating the fuel quantity to a fuel supply control system. The fuel supply control system may be part of the maritime vessel or external to the maritime vessel. In some cases, instructions 745 and 750 may be repeated to generate total fuel consumption estimates for multiple network segments of the voyage. In this case, instructions 755 and 760 may be used to supply fuel of both fuel types for a trip or round trip including multiple network segments.

[0080] Some examples of methods and systems characterized by generating fuel consumption estimates for maritime vessels have been described. These examples use predictive models to compute the estimates. These predictive models can be fitted to parameters based on historical data. Historical data can include performance data available to the fleet, such as data over a period of several months or years. The estimates can be used to properly supply (i.e., load) fuel to maritime vessels on one or more network segments. These estimates can avoid refueling, such as additional emergency refueling operations. They may also help optimize maritime vessel loading to reduce fuel consumption and improve maritime vessel efficiency, thereby reducing pollution levels. For example, after the introduction of a new route and / or a new maritime vessel or significant retrofitting of a maritime vessel, the predictive model can be configured to handle maritime vessels or voyages for which there is no prior historical data. The predictive model can include terms or coefficients relating to a general contribution applied to all maritime vessels and network segments, as well as specific contributions related to a particular maritime vessel category and network segment.

[0081] In testing, the fuel consumption estimates calculated using the examples described herein showed a significant improvement in accuracy compared to conventional fuel control systems. For example, the average error indicative of each voyage (e.g., actual fuel consumption minus the root mean square error of the fuel consumption estimate) was reduced by approximately 25% to 30%, depending on the fuel type.

[0082] It should be noted that although each example is described individually, features from each example can be combined, and a feature of one example can be combined with features of one or more other examples. Examples of the invention have been discussed. However, it should be understood that changes and modifications can be made to the examples described within the scope of the invention.

Claims

1. A method comprising: Obtain data identifying the first maritime vessel and data identifying the voyage to be undertaken by the first maritime vessel; A predictive model is used to calculate a fuel consumption estimate. This predictive model is a machine learning model configured to receive data identifying the first maritime vessel and data identifying the voyage to be undertaken as input. The parameters of the predictive model are fitted using historical data from multiple maritime vessels and multiple voyages. The predictive model is a parameterized function of an initial fuel consumption estimate, the class of the first maritime vessel, the voyage segment identifier, the voyage duration, and the voyage speed. The amount of fuel to be supplied to the first seagoing vessel is determined based on the estimated fuel consumption.

2. The method according to claim 1, comprising: Control the supply of a predetermined amount of fuel to the first seagoing vessel.

3. The method of claim 1, wherein the data identifying the first maritime vessel includes an indication of the category of the first maritime vessel, and the historical data of the plurality of maritime vessels includes an indication of the category of the plurality of maritime vessels.

4. The method of claim 1, wherein the data identifying the voyage to be undertaken includes identification of segments of the voyage, the segments of the voyage having an arrival port and a departure port, wherein the prediction model is configured to receive the identification of the segments of the voyage as input.

5. The method according to claim 4, comprising: Determine the distance between the port of arrival and the port of departure; as well as Determine the estimated average speed of the first maritime vessel during the voyage. The prediction model is configured to receive the distance and the estimated mean velocity as input.

6. The method of claim 1, wherein the prediction model is a linear model, and the parameters of the linear model are fitted using elastic network regularization.

7. The method according to claim 1, wherein the prediction model comprises: Parameters independent of the data identifying the first maritime vessel; The parameters depend on the data identifying the first maritime vessel; Parameters independent of the data identifying the voyage to be undertaken by the first maritime vessel; as well as The parameters depend on the data that identifies the voyage to be undertaken by the first maritime vessel.

8. The method according to claim 1, comprising: Obtain initial fuel consumption estimates. The prediction model is configured to receive the initial fuel consumption estimate as input.

9. The method of claim 8, wherein the initial fuel consumption estimate is calculated based on data indicating how fuel consumption changes with the speed of the first seagoing vessel.

10. The method of claim 8, wherein the initial fuel consumption estimate is received as user input.

11. The method of claim 1, wherein the first maritime vessel is configured to consume multiple types of fuel, and wherein calculating fuel consumption estimates using a predictive model comprises: Calculate the estimated fuel consumption for the various fuel types.

12. The method of claim 11, wherein calculating the fuel consumption estimates for the multiple fuel types comprises: The first prediction model is used to calculate the estimated first fuel consumption for the first fuel type; The second prediction model is used to calculate the estimated second fuel consumption for the second fuel type, and Determining the fuel quantity includes determining the amount of each of the first fuel type and the second fuel type to be supplied to the first maritime vessel based on the corresponding first fuel consumption estimate and second fuel consumption estimate.

13. The method of claim 12, comprising: Obtain initial fuel consumption estimates for each of the first and second fuel types. The first prediction model is configured to receive the initial fuel consumption estimate for the first fuel type as input, and The second prediction model is configured to receive an initial fuel consumption estimate for the second fuel type and an indication of whether the initial fuel consumption estimate for the second fuel type is non-zero as input.

14. A fuel control system for marine vessels, comprising: An interface is provided for receiving data identifying the first maritime vessel and data identifying the voyage to be undertaken by the first maritime vessel. The memory, which is used to store: Computer program code for implementing a prediction model, the prediction model being a machine learning model, and the prediction model being a parameterized function of an initial fuel consumption estimate, the class of the first maritime vessel, a voyage segment identifier, a voyage duration, and a voyage speed; The parameter values ​​of the prediction model are derived from the optimization of the prediction model using historical data from multiple shipping vessels and multiple voyages; A processor, configured to retrieve the computer program code and the parameter values ​​from the memory, the processor being arranged to execute the computer program code to: The prediction model is instantiated using the parameter values; The data identifying the first maritime vessel and the data identifying the voyage to be undertaken by the first maritime vessel are supplied to the prediction model; as well as The prediction model is applied to determine the estimated fuel consumption of the first maritime vessel during the voyage; as well as A fuel supply system for determining the amount of fuel to be supplied to at least one fuel tank of the first maritime vessel based on the fuel consumption estimate.

15. A non-transitory computer-readable storage medium storing instructions, said instructions causing the processor, when executed by a processor, to: Obtain data identifying the first maritime vessel and data identifying the voyage to be undertaken by the first maritime vessel; The category of the first maritime vessel is determined based on the data identifying the first maritime vessel; The flight segment identifier, flight duration, and flight speed are determined based on the data identifying the flight route; Obtain initial fuel consumption estimates for each of the first fuel type and the second fuel type, the initial fuel consumption estimates being based on the fuel consumption rate of the first maritime vessel at the range speed and the range duration; A first prediction model is applied to output an adjusted fuel consumption estimate for the first fuel type. The first prediction model is a machine learning model and is a parameterized function of the initial fuel consumption estimate for the first fuel type, the category of the first maritime vessel, the voyage segment identifier, the voyage duration, and the voyage speed. The parameters of the first prediction model are optimized based on historical data from multiple maritime vessels and multiple voyages. A second prediction model is applied to output an adjusted fuel consumption estimate for the second fuel type. The second prediction model is a machine learning model and is a parameterized function of the initial fuel consumption estimate for the second fuel type, the category of the first maritime vessel, the voyage segment identifier, the voyage duration, and the voyage speed. The parameters of the second prediction model are optimized based on the historical data. The amount of the first type of fuel to be supplied to the first maritime vessel is determined based on the adjusted fuel consumption estimate of the first fuel type. as well as The amount of the second type of fuel to be supplied to the first maritime vessel is determined based on the adjusted fuel consumption estimate of the second fuel type.

Images