Ship oil consumption prediction method based on data assimilation

Abstract

A ship fuel consumption prediction method based on data assimilation not only has the advantages of a white box model and a black box model, but also greatly reduces the number of required data, improves the precision, avoids the step of establishing a weak physical relationship between original data and fuel consumption by the black box model, and improves the fuel consumption prediction efficiency. And the method can be more flexibly applied to different ship types and different navigation working conditions. According to the technical scheme, the method comprises the following steps that S1, a white box model is established; s2, case ship navigation data processing; s3, determining a state vector; s4, generating a set; s5, calculating Hx; s6, data assimilation; and S7, calculating oil consumption.

Description

Technical Field

[0001] The invention belongs to the technical field of ship fuel consumption prediction methods, and in particular relates to a ship fuel consumption prediction method based on data assimilation. Background Art

[0002] Ship fuel consumption is not only related to the economic benefits of the operating company, but also a large amount of energy consumption and carbon dioxide emissions seriously endanger global resources and the environment. Since 2010, the International Maritime Organization (IMO) has introduced ship energy efficiency indexes such as EEOI (Energy Efficiency Operational Index) and EEDI (Energy Efficiency Design Index) to measure the energy consumption level of ships in the design and operation stages. Accurately predicting ship fuel consumption is a prerequisite for calculating ship energy efficiency indexes and optimizing navigation strategies.

[0003] Ship fuel consumption models can be roughly divided into white box models, black box models and gray box models.

[0004] White box models usually make reasonable assumptions and restrictions, and establish a simplified model between measurable physical quantities (input) and ship fuel consumption (output) through the physical connection within the system. Since 1975, Shoichi Nakamura, Coraddu, Fabian and others have successively proposed white box models such as fuel consumption regression formula, Matlab fuel consumption calculation toolbox, and mathematical relationship between ship speed and fuel consumption. These white box models can complete the calculation of ship fuel consumption, and because the internal parameter relationship of the physical model is transparent, it is relatively convenient to modify and improve each module in the white box model to improve its accuracy or real-time performance. However, the process of establishing the white box model depends on a large amount of ship performance parameters, navigation data, and environmental data such as wind, waves and currents. Some data are difficult to obtain, which greatly reduces the actual calculation accuracy of the white box model and limits its application.

[0005] The black box model is a model that directly describes the relationship between the input and output of the system. Compared with the white box model, the black box model does not need to understand the physical relationship between the internal parameters of the system, but only relies on data to establish the relationship between the input and output of the system. The physical meaning of the model parameters is often unclear. Machine learning has risen rapidly and has become a typical representative of the black box model. Petersen, Ye Rui, Mou Xiaohui and others used neural networks, random forests and other algorithms to calculate and predict the fuel consumption of different case ships in real time. The ship's navigation and environmental data are used as input, and the ship's historical fuel consumption data is used as output for training. The ship's own performance information is reflected in the parameters of the black box model, so this method can be applied to any ship type. However, due to the differences in ship performance and navigation environment, each ship's fuel consumption corresponds to a unique black box model. The training model requires a large data set, so a long period of data accumulation is required before each ship's fuel consumption model is established. In addition, since the accumulation of dirt on the bottom of the ship during navigation will change the resistance performance of the ship, the black box model needs to be changed regularly to accurately predict the ship's fuel consumption. In addition, due to its own principles, the black box model does not provide ideal estimation results for extrapolation situations when estimating fuel consumption.

[0006] The gray box model is a model between the black box model and the white box model, and has the advantages and disadvantages of both black box models. When establishing a gray box model, the white box part is established using known physical quantities and physical relationships, and the part with uncertain physical relationships is filled in by the black box model. In 2008, Leifson et al. first used the white box model to model the ship's resistance, main engine and propeller to obtain an estimated fuel consumption value, and then used the black box model to correct the fuel consumption value, thereby improving the accuracy of the fuel consumption estimation. The gray box model can make full use of known physical relationships to establish the initial model, and improve the model accuracy based on measured data when the relationship between some physical quantities is uncertain or the physical quantities themselves are uncertain, and complete effective ship fuel consumption calculations. However, the method proposed by Leifson et al. uses the black box model to directly correct the fuel consumption value, which is still to establish a weak physical relationship between the original data and fuel consumption, which requires a relatively large amount of data accumulation and model updating. Summary of the invention

[0007] The present invention solves the problem that the white box and black box models in the current traditional fuel consumption prediction methods each have their own advantages and disadvantages, and the gray box model combining the two models essentially separates the white box and black box models, and the two models still play their own roles, unable to give full play to their respective advantages, and unable to abandon their weaknesses, resulting in technical problems such as large data volume, low precision and inflexibility. A ship fuel consumption prediction method based on data assimilation is provided, which not only has the advantages of the white box and black box models, but also greatly reduces the amount of required data, improves precision, and avoids the step of establishing a weak physical relationship between the original data and fuel consumption by the black box model, so that it can be more flexibly applied to different ship types and different navigation conditions.

[0008] The technical solution of the present invention is:

[0009] A method for predicting ship fuel consumption based on data assimilation, characterized by comprising the following steps:

[0010] S1: Establishment of white box model: Using the traditional fuel consumption model as a white box, a fuel consumption calculation model under the influence of multiple (factors) parameters such as speed, hull parameters, wind and waves is established from a physical perspective;

[0011] S2: Pre-processing of case ship navigation data: Establish navigation data based on the original data of the case ship, and process the missing values, outliers, synchronization frequency and necessary physical calculations of the various types of measured original data during the navigation of the case ship to obtain various types of ship navigation data, that is, the data of different variables in the fuel consumption calculation model changing with the navigation time, including the ship's navigation speed v, significant wave height H 1 / 3 , the speed of the wave relative to the ship u and the mass of fuel consumed per unit time Q;

[0012] S3: Determine the state vector: Analyze the uncertain parameters of the target case ship according to the traditional fuel consumption model formula described in S1 to form the state vector in data assimilation;

[0013] S4: Generate a set: Set reasonable initial values ​​for each uncertain parameter according to the traditional model formula and determine the standard deviation, and generate a sample set of normal distribution;

[0014] S5: Calculate Hx: use the traditional fuel consumption model formula to calculate the corresponding fuel consumption under different parameters, and compare it with the actual fuel consumption Q;

[0015] S6: Data assimilation: The fuel consumption mass Q per unit time during a certain period of the voyage of the S2 case ship is used as the observation data, and other data related to the traditional model are substituted into the formula. The uncertain parameters in the traditional fuel consumption model are corrected by data assimilation to obtain the uncertain parameters that are most consistent with the actual system;

[0016] S7: Calculating fuel consumption: The accurate uncertain parameters after assimilation correction are brought into the traditional fuel consumption model to recalculate the fuel consumption.

[0017] The multiple parameters in step S1 include: hull parameters include ship length and ship width, speed parameters include design speed, and multiple parameters also include still water resistance coefficient, wave resistance coefficient, propeller propulsion efficiency, wet area, auxiliary engine power, fuel consumption rate, and fuel consumption.

[0018] The traditional fuel consumption model formula in step S1 is formula (1) and formula (2),

[0019]

[0020] Q=K f P (2)

[0021] Among them, K is the coefficient related to the propeller propulsion efficiency η and the ship speed, η is the ship speed is the design speed v d The propeller propulsion efficiency is usually 0.6 to 0.7. When the ship speed v decreases, K decreases according to the constants j and k (where j+k=1). P is the total power, where ρ is the density of water, and C T is the total drag coefficient, S is the wetted surface, v is the velocity; Cw is the wave drag coefficient, g is the gravitational acceleration, H 1 / 3 is the significant wave height, B is the ship width, L is the ship length, u is the speed of the wave relative to the ship, Paux is the auxiliary main engine power required to operate the water pump, lighting electricity and all support systems of the ship. In formula (2), Q is the fuel consumption per unit time (kg / s), K f is the fuel consumption rate (kg / J).

[0022] In S6, the data assimilation calculation formula is the EnKF data assimilation calculation formula, specifically formula (3), formula (4) and formula (5), and the data assimilation third-party library is Data assimilation third-party library "dafi" code;

[0023] x a(j) =x f(j) +K(y (j) -z f(j) ) (3)

[0024] z=Hx (4)

[0025] K=C xz (C z +C y ) -1 (5)

[0026] In the EnKF data assimilation calculation formula, x is the state vector, y is the actual observation data, the superscripts a and f represent the meaning of analysis and prediction respectively, and (j) represents the jth sample. z is the mapping of the sample state vector x in the observation space where y is located, H is the observation operator used to realize the mapping of x, K is the gain matrix, where C is the covariance matrix, and the subscripts xz, z, and y represent the covariance matrices between vectors x and z, and z and y respectively;

[0027] With respect to the traditional fuel consumption model formula described in claim 3, the uncertain parameters include j, P aux and C w , the three constitute the state vector x, generate the sample of the state vector normal distribution, and then through the gain matrix K and the error (yz) between the sample and the actual fuel consumption, make the uncertain parameter vector x approach the actual parameter value continuously.

[0028] The case ship uses the measured data of the passenger and vehicle ferry MS Smyril during its voyage as the original data. This data set covers the voyage data from February 16, 2010 to April 12, 2010, recording a total of 246 voyages. The case ship collected 17 different types of data during this voyage.

[0029] The above-mentioned traditional fuel consumption model of the case ship involves 6 types of data, namely: starboard draft, fuel density, fuel volume flow rate, ground speed, wind speed and wind direction.

[0030] The technical effect of the present invention is as follows: a ship fuel consumption prediction method based on data assimilation first establishes a white box model of a traditional fuel consumption model, and establishes real navigation data based on the original data of the case ship to obtain the real fuel consumption, and then analyzes the uncertain parameters in the traditional fuel consumption model, and processes these uncertain parameters using the data assimilation method, that is, by artificially generating samples of a state vector normal distribution based on the uncertain parameters, and then obtaining the state vector that is most consistent with the actual system through the data assimilation calculation formula and the data assimilation third-party library, that is, through the uncertain parameter gain matrix K and the error (yz) between the sample and the real fuel consumption, the uncertain parameter vector x is continuously approached to the real parameter value, thereby avoiding the weak physical relationship between the original data and the fuel consumption established based on the black box model.

[0031] Therefore, the present invention takes into account the advantages of the white box model's "strong physical logic" and "easy to improve" and the black box model's "wide range of applications" through the organic combination of the black box model and the white box model, while avoiding the limitation of the large amount of data required by both, and establishes a data-driven model under strong physical constraints using very little data.

[0032] Although the present invention belongs to the fuel consumption gray box model, compared with the fuel consumption gray box model proposed by Leifson et al., the model proposed by the present invention does not separate the white box and black box models, that is, it simply adopts the method of first predicting with the white box and then correcting the fuel consumption value with the black box. Instead, it implants the EnKF (black box) into the traditional fuel consumption model (white box), and uses the EnKF black box to obtain more accurate white box parameters, so as to achieve the effect of correcting physical relationships, physical quantities and improving model accuracy. Compared with the method proposed by Leifson et al., since it avoids the step of establishing a weak physical relationship between the original data and fuel consumption through a neural network, and avoids the step of establishing a weak physical relationship between the original data and fuel consumption by a black box model, the amount of required data is greatly reduced, and it can be more flexibly applied to different ship types and different navigation conditions.

[0033] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flowchart of the prediction method of the present invention;

[0035] Figure 2 is Basic parameters of the ship according to the embodiment of the present invention;

[0036] Figure 3 It is a comparison chart between the predicted value of the traditional fuel consumption prediction method and the actual fuel consumption;

[0037] Figure 4 is Comparison chart of the prediction results of the fuel consumption prediction method of the present invention, the traditional fuel consumption prediction results and the actual fuel consumption. DETAILED DESCRIPTION

[0038] Figure 1 In the present invention, a ship fuel consumption prediction method based on data assimilation comprises the following steps:

[0039] S1: Establishment of white box model: Using the traditional fuel consumption model as a white box, a fuel consumption calculation model under the influence of multiple (factors) parameters such as speed, hull parameters, wind and waves is established from a physical perspective;

[0040] S2: Pre-processing of the original data of the case ship: Establish the navigation data based on the original data of the case ship, and process the missing values, abnormal values, synchronization frequency and necessary physical calculations of the multiple types of measured original data during the navigation of the case ship to obtain the navigation data of multiple types of ships, that is, the data of different variables in the fuel consumption calculation model changing with the navigation time, including the ship's navigation speed v, significant wave height H 1 / 3 , the speed of the wave relative to the ship u and the mass of fuel consumed per unit time Q;

[0041] S3: Determine the state vector: Analyze the uncertain parameters of the target case ship according to the traditional fuel consumption model formula described in S1 to form the state vector in data assimilation.

[0042] S4: Generate a set: Set reasonable initial values ​​for each uncertain parameter according to the traditional model formula and determine the standard deviation, and generate a sample set of normal distribution;

[0043] S5: Calculate Hx: use the traditional fuel consumption model formula to calculate the corresponding fuel consumption under different parameters, and compare it with the actual fuel consumption Q;

[0044] S6: Data assimilation: The fuel consumption per unit time Q of the S2 case ship during a certain period of time during a voyage is used as the observation data, and other data related to the traditional model are substituted into the formula. The uncertain parameters in the traditional fuel consumption model are corrected by data assimilation to obtain the uncertain parameters that are most consistent with the actual system.

[0045] S7: Calculating fuel consumption: The accurate uncertain parameters after assimilation correction are brought into the traditional fuel consumption model to recalculate the fuel consumption.

[0046] The multiple parameters in step S1 include: hull parameters include ship length and ship width, speed parameters include design speed, and multiple parameters also include still water resistance coefficient, wave resistance coefficient, propeller propulsion efficiency, wet area, auxiliary engine power, fuel consumption rate, and fuel consumption.

[0047] The traditional fuel consumption model formula in step S1 is formula (1) and formula (2), which analyzes the uncertain parameters of the target ship and constitutes the state vector in data assimilation;

[0048]

[0049] Q=K f P (2)

[0050] Where K is a coefficient related to propeller propulsion efficiency η and ship speed. η is the ship speed v d The propeller propulsion efficiency is usually 0.6 to 0.7. When the ship speed v decreases, K decreases according to the constants j and k (where j+k=1). P is the total power, where ρ is the density of water, C T is the total drag coefficient, S is the wetted surface, v is the velocity. Cw is the wave drag coefficient, g is the gravitational acceleration, H 1 / 3 is the significant wave height, B is the ship width, L is the ship length, and u is the speed of the wave relative to the ship. Paux is the auxiliary main engine power required to operate water pumps, lighting electricity, and all support systems of the ship. In formula (2), Q is the mass of fuel consumed per unit time (kg / s), and Kf is the fuel consumption rate (kg / J).

[0051] In step S6, the data assimilation calculation formula is the EnKF data assimilation calculation formula. The uncertain parameters in the traditional model are corrected according to the actual navigation fuel consumption data of the case ship. The EnKF data assimilation calculation formula is used, specifically formula (3), formula (4) and formula (5). The data assimilation third-party library is Data assimilation third-party library "dafi" code, to obtain the state vector that is most consistent with the actual system, that is, the uncertain parameters;

[0052] x a(j) =x f(j) +K(y (j) -z f(j) ) (3)

[0053] z=Hx (4)

[0054] K=C xz (C z +C y ) -1 (5)

[0055] In EnKF, x is the state vector and y is the actual observation data. The superscripts a and f represent the meaning of analysis and prediction respectively, and (j) represents the jth sample. z is the mapping of the sample state vector x in the y observation space, H is the observation operator used to realize the mapping of x, K is the gain matrix, where C is the covariance matrix, and the subscripts xz, z, y represent the covariance matrices between vectors x and z, and z, y.

[0056] With respect to the traditional fuel consumption model formula described in claim 3, the uncertain parameters include j, P aux and C w , the three constitute the state vector x, generate the sample of the state vector normal distribution, and then through the gain matrix K and the error (yz) between the sample and the actual fuel consumption, make the uncertain parameter vector x approach the actual parameter value continuously.

[0057] The case ship uses the measured data of the passenger and vehicle ferry MS Smyril during its voyage as the original data. This data set covers the voyage data from February 16, 2010 to April 12, 2010, recording a total of 246 voyages. The case ship collected 17 different types of data during this voyage.

[0058] The above-mentioned traditional fuel consumption model of the case ship involves 6 types of data, namely: starboard draft, fuel density, fuel volume flow rate, ground speed, wind speed and wind direction.

[0059] The present invention establishes a gray box model based on data assimilation and the traditional fuel consumption physical model, achieving a more accurate, efficient and flexible fuel consumption prediction capability. The traditional fuel consumption model is used as a white box, which establishes a fuel consumption calculation model under the influence of multiple factors such as speed, hull parameters, wind and waves from a physical perspective:

[0060]

[0061] Q=K f P (2)

[0062] Where K is a coefficient related to propeller propulsion efficiency η and ship speed. η is the ship speed v d The propeller propulsion efficiency is usually 0.6 to 0.7. When the ship speed v decreases, K decreases according to the constants j and k (where j+k=1). Ps represents the ship's static water power, where ρ is the density of water, CT is the total resistance coefficient, S is the wetted surface, and v is the speed. Pw represents the power required for the ship to overcome wave resistance, Cw is the wave resistance coefficient, g is the acceleration of gravity, H1 / 3 is the significant wave height, B is the ship's width, L is the ship's length, and u is the speed of the wave relative to the ship. Paux is the auxiliary main engine power required to operate water pumps, lighting electricity, and all support systems of the ship. In formula (2), Q is the mass of fuel consumed per unit time (kg / s), and Kf is the fuel consumption rate (kg / J).

[0063] The measured data of the passenger and vehicle ferry MS Smyril during the voyage is used as the original data. This data set covers the voyage data from February 16, 2010 to April 12, 2010, and a total of 246 voyages are recorded. The case ship collected 17 different types of data during this voyage, and the above-mentioned traditional fuel consumption model involves 6 of them, as shown in Table 1. After the original data is processed by missing values, outliers, synchronization frequency and necessary physical calculations, the navigation data of v, H1 / 3, u, and Q are obtained. The basic parameters of the case ship are shown in 2. By selecting appropriate parameters j and k, this traditional fuel consumption model can estimate the actual ship fuel consumption.

[0064] Table 1. Data types collected by the example ship

[0065]

[0066]

[0067] However, the effect of this traditional model is not satisfactory. Guo Qiang compared the actual fuel consumption of another case ship with the calculation results of this traditional model. Figure 3As shown in the figure, the error between the traditional fuel consumption model and the actual value. In the figure, blue represents the actual fuel consumption and orange represents the calculation result of the traditional model. This model has a certain degree of accuracy, but there is still a large error. At the same time, the predicted fuel consumption value trend basically cannot fully reflect the trend of the actual fuel consumption value. This phenomenon may be because the physical relationship of the system is not accurately described by the traditional model, or because the physical quantity in the model deviates from the actual value.

[0068] In order to correct the traditional model as much as possible, the present invention adopts EnKF (ensemble Kalmen filter) data assimilation method to correct the uncertain parameters in the traditional model through the actual navigation fuel consumption data of the case ship. EnKF is a Kalmen filter based on Monte Carlo theory. Its principle can be simply understood as artificially creating a set of n samples with normal distribution, and correcting the distribution of n samples through filtering methods, so that the system corresponding to n samples is close to the actual situation of the system under study. The EnKF calculation formula is as follows:

[0069] x a(j) =x f(j) +K(y (j) -z f(j) ) (3)

[0070] z=Hx (4)

[0071] K=C xz (C z +C y ) -1 (5)

[0072] In EnKF, x is the state vector and y is the actual observation data. The superscripts a and f represent the meaning of analysis and prediction respectively, and (j) represents the jth sample. z is the mapping of the sample state vector x in the y observation space, and H is the observation operator, which is used to calculate the physical quantity y' corresponding to x. K is the gain matrix, where C is the covariance matrix, and the subscript xz represents the covariance matrix between vectors x and z. For the traditional model of the case ship mentioned above, the observation data y is the fuel consumption value Q of the case ship during navigation. The uncertain parameters include j, Paux and Cw, and these three constitute the state vector x. Artificially generate samples of the state vector with normal distribution, and then use the gain matrix K and the error (yz) between the sample and the actual fuel consumption to make the uncertain parameter vector x continuously approach the actual parameter value.

[0073] Substituting the assimilated and corrected parameters into the traditional model, the accuracy of the corrected model results is significantly improved, such as Figure 4In the figure, the blue dotted line represents the direct calculation result of the traditional fuel consumption model, the orange solid line represents the calculation result of the gray box model proposed in the present invention, and the gray solid line represents the actual measured fuel consumption of the case ship.

Claims

1. A ship fuel consumption prediction method based on data assimilation, characterized by The following steps are involved: S1: Establishment of white box model: Using the traditional fuel consumption model as a white box, a fuel consumption calculation model under the influence of multiple (factors) parameters such as speed, hull parameters, wind and waves is established from a physical perspective; S2: Pre-processing of case ship navigation data: Establish navigation data based on the original data of the case ship, and process the missing values, outliers, synchronization frequency and necessary physical calculations of the various types of measured original data during the navigation of the case ship to obtain various types of ship navigation data, that is, the data of different variables in the fuel consumption calculation model changing with the navigation time, including the ship's navigation speed v, significant wave height H 1 / 3 , the speed of the wave relative to the ship u and the mass of fuel consumed per unit time Q; S3: Determine the state vector: Analyze the uncertain parameters of the target case ship according to the traditional fuel consumption model formula described in S1 to form the state vector in data assimilation; S4: Generate a set: Set reasonable initial values ​​for each uncertain parameter according to the traditional model formula and determine the standard deviation, and generate a sample set of normal distribution; S5: Calculate Hx: use the traditional fuel consumption model formula to calculate the corresponding fuel consumption under different parameters, and compare it with the actual fuel consumption Q; S6: Data assimilation: The fuel consumption mass Q per unit time during a certain period of the voyage of the S2 case ship is used as the observation data, and other data related to the traditional model are substituted into the formula. The uncertain parameters in the traditional fuel consumption model are corrected by data assimilation to obtain the uncertain parameters that are most consistent with the actual system; S7: Calculating fuel consumption: The accurate uncertain parameters after assimilation correction are brought into the traditional fuel consumption model to recalculate the fuel consumption.

2. The ship fuel consumption prediction method based on data assimilation according to claim 1 is characterized by: The multiple parameters in step S1 include: hull parameters include ship length and ship width, speed parameters include design speed, and multiple parameters also include still water resistance coefficient, wave resistance coefficient, propeller propulsion efficiency, wet area, auxiliary engine power, fuel consumption rate, and fuel consumption.

3. The ship fuel consumption prediction method based on data assimilation according to claim 1 or 2, characterized in that The traditional fuel consumption model formula in step S1 is formula (1) and formula (2), Q=K f P (2) Among them, K is the coefficient related to the propeller propulsion efficiency η and the ship speed, η is the ship speed is the design speed v d The propeller propulsion efficiency is usually 0.6 to 0.

7. When the ship speed v decreases, K decreases according to the constants j and k (where j+k=1). P is the total power, where ρ is the density of water, and C T is the total drag coefficient, S is the wetted surface, v is the velocity; Cw is the wave drag coefficient, g is the gravitational acceleration, H 1 / 3 is the significant wave height, B is the ship width, L is the ship length, u is the speed of the wave relative to the ship, Paux is the auxiliary main engine power required to operate the water pump, lighting electricity and all support systems of the ship. In formula (2), Q is the fuel consumption per unit time (kg / s), K f is the fuel consumption rate (kg / J).

4. The ship fuel consumption prediction method based on data assimilation according to claim 3 is characterized in that in S6, the data assimilation calculation formula is the EnKF data assimilation calculation formula, specifically formula (3), formula (4) and formula (5), and the data assimilation third-party library is Data assimilation third-party library "dafi" code; x a(j) =x f(j) +K(y (j) -z f(j) ) (3) z=Hx (4) K=C xz (C z +C y ) -1 (5) In the EnKF data assimilation calculation formula, x is the state vector, y is the actual observation data, the superscripts a and f represent the meaning of analysis and prediction respectively, and (j) represents the jth sample. z is the mapping of the sample state vector x in the observation space where y is located, H is the observation operator used to realize the mapping of x, K is the gain matrix, where C is the covariance matrix, and the subscripts xz, z, and y represent the covariance matrices between vectors x and z, and z and y respectively.

5. The ship fuel consumption prediction method based on data assimilation according to claim 4 is characterized by: With respect to the traditional fuel consumption model formula described in claim 3, the uncertain parameters include j, P aux and C w , the three constitute the state vector x, generate the sample of the state vector normal distribution, and then through the gain matrix K and the error (yz) between the sample and the actual fuel consumption, make the uncertain parameter vector x approach the actual parameter value continuously.

6. The ship fuel consumption prediction method based on data assimilation according to claim 4 is characterized in that The case ship uses the measured data of the passenger and vehicle ferry MS Smyril during its voyage as the original data. This data set covers the voyage data from February 16, 2010 to April 12, 2010, recording a total of 246 voyages. The case ship collected 17 different types of data during this voyage.

7. The ship fuel consumption prediction method based on data assimilation according to claim 4 is characterized in that The above-mentioned traditional fuel consumption model of the case ship involves 6 types of data, namely: starboard draft, fuel density, fuel volume flow rate, ground speed, wind speed and wind direction.

Images