Dual-fuel ship energy efficiency operational indicator prediction method based on physical data hybrid driving

Abstract

The application discloses a kind of based on physical data hybrid drive's dual-fuel ship energy efficiency operation index prediction method, comprising: collecting liquefied natural gas dual-fuel ship historical energy efficiency data and pre-processing;Obtain input characteristics by feature selection.The physical mechanism model is constructed to predict fuel consumption, and the two-way recurrent neural network data-driven model based on the two-stage optimization algorithm framework of bayesian optimization and whale optimization is established, to predict the error of physical mechanism model and historical ship each fuel consumption data;Each fuel consumption predicted based on physical mechanism model is added to the error output by data-driven model through the physical data hybrid drive model formed by the combination of data-driven model and physical mechanism model, to obtain the final each fuel consumption prediction result, and the ship energy efficiency operation index is calculated.The relative error of the energy efficiency operation index of liquefied natural gas dual-fuel ship obtained by the present application can be smaller, to meet the actual energy efficiency management needs.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship energy efficiency management, and in particular to a dual-fuel ship energy efficiency operation index prediction method based on physical data hybrid driving. BACKGROUND

[0002] In recent years, with the gradual slowdown of the global economy and the continuous rise of fuel prices, the shipping industry is facing unprecedented operating pressure. As one of the largest expenditures in shipping operations, fuel costs have driven the industry's growing concern about ship energy efficiency. Traditional ships generally use high-energy power systems, which not only have high operating costs, but also cause serious greenhouse gas emissions, which have a negative impact on the environment and no longer meet the requirements of sustainable development. Therefore, it is particularly important to find clean fuel. Liquid natural gas is considered the most viable alternative fuel for ships due to its significant carbon reduction effect, mature technology, and well-established supply chain, and is expected to gradually replace traditional low-sulfur fuel oil and marine diesel oil and become the mainstream fuel.

[0003] Although considerable progress has been made in ship energy efficiency prediction, most existing models still focus primarily on traditional single-fuel diesel ships. In contrast, liquefied natural gas dual-fuel ships differ fundamentally from single-fuel ships due to their unique combustion mechanism and operating mode. In liquefied natural gas dual-fuel ships, liquefied natural gas is usually used as the main fuel, supplemented by a small amount of marine light diesel oil as pilot fuel. In some operating conditions, marine heavy oil can also be used as an alternative fuel. In dual-fuel mode, a small amount of marine light diesel oil must be used as an ignition fuel during the combustion of liquefied natural gas. Therefore, current research on liquefied natural gas dual-fuel ship energy efficiency operation index prediction is relatively scarce, and existing research does not use a physical data hybrid driving model to predict dual-fuel ship energy efficiency operation index.

[0004] In addition, most existing traditional ship fuel consumption prediction models use single-layer adaptive parameter adjustment algorithms or fixed parameter settings. In the face of the complex operating conditions of liquefied natural gas dual-fuel ships operating in multiple modes and with multiple fuels, the prediction accuracy of the model is difficult to meet the actual energy efficiency management needs, and there is an urgent need to develop advanced prediction methods that have both high accuracy and complex operating condition adaptability. SUMMARY

[0005] The present application provides a dual-fuel ship energy efficiency operation index prediction method based on physical data hybrid driving to overcome the above technical problems.

[0006] To achieve the above purpose, the technical solution of the present application is:

[0007] A dual-fuel ship energy efficiency operation index prediction method based on physical data hybrid driving, comprising:

[0008] S1: Collect historical energy efficiency data of liquefied natural gas dual-fuel ships, preprocess the data to obtain preprocessed energy efficiency data; the historical energy efficiency data includes ship historical fuel consumption data, ship historical operation data and sailing environment data;

[0009] S2: Feature selection is performed on the preprocessed energy efficiency data to obtain input feature data;

[0010] S3: A physical mechanism model of a dual-fuel ship is constructed according to the input feature data to predict the fuel consumption;

[0011] S4: A two-stage optimization algorithm framework of Bayesian optimization algorithm and whale optimization algorithm is constructed by introducing a bidirectional recurrent neural network, and the hyperparameters of the bidirectional recurrent neural network are optimized to obtain a data-driven model based on the two-stage optimization algorithm framework, the data-driven model based on the two-stage optimization algorithm framework is used to receive the input feature data and the error between the ship historical fuel consumption data and the fuel consumption predicted based on the physical mechanism model, and output the minimum error of the fuel consumption;

[0012] S5: The data-driven model based on the two-stage optimization algorithm framework is combined with the physical mechanism model to form a physical data hybrid driving model; the fuel consumption predicted based on the physical mechanism model and the minimum error of the fuel consumption output by the data-driven model based on the two-stage optimization algorithm framework are added through the physical data hybrid driving model to obtain the fuel consumption prediction result based on the physical data hybrid driving model, and then the energy efficiency operation index prediction result of the dual-fuel ship is calculated according to the fuel consumption prediction result output by the physical data hybrid driving model.

[0013] Further, a two-stage optimization algorithm framework of Bayesian optimization algorithm and whale optimization algorithm is constructed by introducing a bidirectional recurrent neural network, and the hyperparameters of the bidirectional recurrent neural network are optimized to obtain a data-driven model based on the two-stage optimization algorithm framework, the data-driven model based on the two-stage optimization algorithm framework is used to receive the input feature data and the error between the ship historical fuel consumption data and the fuel consumption predicted based on the physical mechanism model, and output the minimum error of the fuel consumption, including:

[0014] S41: The error between the input feature data and the ship historical fuel consumption data and the fuel consumption predicted based on the physical mechanism model is divided into a training set and a test set;

[0015] S42: The whale optimization algorithm parameters are randomly initialized using the Bayesian optimization algorithm, and the whale optimization algorithm parameters include the number of whale population, the linear decreasing factor, the coefficient vector, the random vector, the probability parameter and the spiral parameter;

[0016] S43: initialize a whale population, each individual representing a combination of parameters of a bidirectional recurrent neural network; train the bidirectional recurrent neural network corresponding to the combination of parameters of the bidirectional recurrent neural network based on the initial whale population using the training set, and calculate the root mean square error of the error of each fuel consumption data of the ship history and the fuel consumption predicted based on the physical mechanism model on the test set as the fitness of the initial whale individual;

[0017] S44: compare the fitness of all initial whale individuals, and select the position of the whale individual with the minimum fitness as the global optimal position of the initial whale population, i.e. the current optimal combination of parameters of the bidirectional recurrent neural network;

[0018] S45: update the coefficient vector, linear decreasing factor, random vector, probability parameter and spiral parameter of the whale optimization algorithm;

[0019] S46: for each whale individual, execute three different position update strategies according to the probability parameter and the modulus of the coefficient vector, and the specific strategies are as follows:

[0020] When the probability parameter is greater than or equal to a set threshold, perform a "spiral position update" operation, i.e. the whale individual moves along a spiral trajectory around the global optimal solution to update the position;

[0021] When the probability parameter is less than the set threshold and the modulus of the coefficient vector is less than a preset value, perform a "surrounding prey" operation, i.e. the whale individual converges inwardly with the current global optimal solution as the center to update the position;

[0022] When the probability parameter is less than the set threshold and the modulus of the coefficient vector is greater than or equal to the preset value, perform a "random search prey" operation, i.e. the whale individual updates the position by randomly selecting a reference individual for global exploration;

[0023] S47: generate a new combination of parameters of the bidirectional recurrent neural network according to the updated position of the whale individual, calculate the new fitness in the manner of step S43, and update the global optimal position of the whale population according to it;

[0024] S48: repeat the steps of S43-S47 until the termination condition of the whale optimization algorithm is reached, i.e. the root mean square error tends to converge or the whale optimization algorithm reaches the maximum number of iterations, and output the global optimal position obtained by the whale optimization algorithm;

[0025] S49: determining whether a termination condition of the Bayesian optimization algorithm is reached, i.e., the Bayesian optimization algorithm reaches a maximum number of iterations or a root mean square error tends to converge, if not, updating the surrogate model in the Bayesian algorithm according to the global optimal position obtained by the whale optimization algorithm at this time, re-sampling a new set of whale optimization algorithm parameters to enter the next iteration, if reached, ending the iteration and outputting the minimum error of each fuel consumption and the global optimal position obtained by the whale optimization algorithm, i.e., the final bidirectional recurrent neural network parameter combination; the minimum error of each fuel consumption includes a minimum error of marine heavy oil consumption, a minimum error of marine light diesel oil consumption and a minimum error of liquefied natural gas consumption.

[0026] Further, the data-driven model based on the two-stage optimization algorithm framework is combined with the physical mechanism model to form a physical data hybrid driving model; the minimum error of each fuel consumption based on the physical mechanism model and the data-driven model based on the two-stage optimization algorithm framework is added through the physical data hybrid driving model to obtain the prediction result of each fuel consumption based on the physical data hybrid driving model, including:

[0027] S51: using the physical mechanism model to predict according to the input feature data to obtain the prediction value of each fuel consumption based on the physical mechanism model;

[0028] S52: inputting the error of each fuel consumption data of the ship history and the input feature data into the data-driven model based on the two-stage optimization algorithm framework to output the minimum error of each fuel consumption;

[0029] S53: combining the data-driven model based on the two-stage optimization algorithm framework with the physical mechanism model to form a physical data hybrid driving model; adding the prediction value of each fuel consumption based on the physical mechanism model obtained in S51 and the minimum error of each fuel consumption output from the data-driven model based on the two-stage optimization algorithm framework in S52 through the physical data hybrid driving model to obtain the final prediction result of each fuel consumption based on the physical data hybrid driving model.

[0030] Further, the physical mechanism model of the dual-fuel ship is constructed according to the input feature data to predict each fuel consumption, including:

[0031] S31, calculating the total resistance of the ship based on the input feature data;

[0032] S32, calculating the effective power of the propeller based on the total resistance of the ship, as shown in formula (1),

[0033] (1)

[0034] wherein, P is the effective power of the propeller, T is the effective thrust generated by the propeller, V is the ship speed, R is the total resistance of the ship;

[0035] S33, the main engine power of the ship is calculated based on the effective power of the propeller and the ship energy efficiency data, as shown in equation (2),

[0036] (2)

[0037] wherein, P is the main engine power; η is the propeller wake efficiency; η is the shafting transmission efficiency; η is the relative rotation efficiency; η represents the hull efficiency;

[0038] The propeller wake efficiency is shown in equation (3),

[0039] (3)

[0040] wherein, T is the thrust of the propeller, M is the torque of the propeller, N is the rotational speed of the propeller, V is the advance speed of the propeller, C is the thrust coefficient, C is the torque coefficient, C is the propeller advance speed coefficient, as shown in equation (4),

[0041] (4)

[0042] wherein, L is the advance of the propeller; D is the propeller diameter, K is the wake coefficient;

[0043] The thrust of the propeller is shown in equation (5),

[0044] (5)

[0045] wherein, ρ is the seawater density, N is the rotational speed of the propeller, D is the propeller diameter;

[0046] The thrust deduction coefficient is calculated based on the thrust of the propeller and the total resistance of the ship, as shown in equation (6),

[0047] (6)

[0048] wherein, is a thrust deduction fraction, is an effective thrust of the propeller;

[0049] The hull efficiency is calculated based on the thrust deduction fraction and the wake fraction, as shown in equation (7),

[0050] (7)

[0051] wherein, is a hull efficiency, is a wake fraction;

[0052] S34, each fuel consumption is calculated based on the main engine power, as shown in equations (8)-(10),

[0053] (8)

[0054] (9)

[0055] (10)

[0056] wherein, is a main engine power; , , are a marine heavy fuel oil consumption rate, a marine light diesel oil consumption rate and a liquefied natural gas consumption rate of the ship main engine, respectively; is a ship sailing time, , , are the marine heavy fuel oil consumption, the marine light diesel oil consumption and the liquefied natural gas consumption predicted by the physical mechanism model.

[0057] Further, the total resistance of the ship is calculated based on the input feature data, including:

[0058] S311: the frictional resistance is calculated, as shown in equation (11),

[0059] (11)

[0060] wherein, is a frictional resistance, is a seawater density, is a ship speed, is a frictional resistance coefficient, as shown in equations (12) and (13), is a wet surface area, as shown in equation (14),

[0061] (12)

[0062] (13)

[0063] (14)

[0064] wherein, is the frictional resistance coefficient, is the Reynolds number representing the ratio of inertial force and viscous force, is the ship speed, is the waterline length of the ship, is the kinematic viscosity of seawater, is the ship's length between perpendiculars, is the ship's displacement;

[0065] S312: Calculate the residual resistance of the ship based on the Savitsky-Keller method, as shown in equation (15),

[0066] (15)

[0067] wherein, is the residual resistance, is the residual resistance coefficient;

[0068] S313: Calculate the appendage resistance of the ship, as shown in equation (16),

[0069] (16)

[0070] wherein, is the appendage resistance, is the appendage resistance coefficient;

[0071] S314: Calculate the air resistance of the ship, as shown in equation (17),

[0072] (17)

[0073] wherein, is the air resistance, is the air resistance coefficient, is the air density, is the wind speed, is the projected area in the direction of the ship's waterline above the cross-sectional area;

[0074] S315: Calculate the wave-induced resistance, as shown in equation (18),

[0075] (18)

[0076] wherein, is the wave-induced resistance, is the wave height, is the ship width, is the square coefficient, is the seawater density, is the gravity acceleration, is the ship length;

[0077] S316: calculating the total resistance of the ship, as shown in equation (19),

[0078] (19)

[0079] wherein, is the total resistance of the ship, is the frictional resistance, is the residual resistance, is the appendage resistance, is the air resistance, is the wave resistance.

[0080] Further, the energy efficiency operation index prediction result of the dual-fuel ship is calculated according to the fuel consumption prediction result output by the physical data hybrid driving model, including:

[0081] The energy efficiency operation index prediction result of the dual-fuel ship is calculated based on the fuel consumption prediction result output by the final physical data hybrid driving model, as shown in equation (20),

[0082] (20)

[0083] wherein, is the energy efficiency operation index of the dual-fuel ship, , , respectively represent the carbon dioxide emission factors of marine heavy oil, marine light diesel oil and liquefied natural gas, 、 and represent the prediction results of the consumption of marine heavy oil, marine light diesel oil and liquefied natural gas based on the final physical data hybrid driving model output, represent the mass of the loaded cargo, represent the transportation distance of the loaded cargo.

[0084] Further, the Pearson correlation coefficient is used for feature selection to obtain the input feature data.

[0085] Beneficial effects: the application provides a dual-fuel ship energy efficiency operation index prediction method based on physical data hybrid driving, which provides a physical theoretical basis for fuel consumption prediction of dual-fuel ships through a physical mechanism model, and can efficiently and accurately find the bidirectional recurrent neural network parameter combination suitable for predicting the fuel consumption of the target ship LNG dual-fuel ship through the two-stage optimization algorithm framework based on the Bayesian optimization algorithm and the whale optimization algorithm, so as to correct the error between the physical mechanism model prediction value and the fuel consumption obtained from the actual ship collected historical data, so that the final obtained fuel consumption prediction result of the target ship LNG dual-fuel ship is more accurate, and the relative error of the energy efficiency operation index of the predicted LNG dual-fuel ship is smaller, which meets the actual energy efficiency management demand. BRIEF DESCRIPTION OF DRAWINGS

[0086] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0087] Figure 1 A flowchart of a dual-fuel ship energy efficiency operation index prediction method based on physical data hybrid driving provided by the present application is shown in the figure.

[0088] Figure 2 A flowchart of a physical data hybrid driving model of the present application is shown in the figure.

[0089] Figure 3 A flowchart of hyperparameter optimization of bidirectional recurrent neural network by two-stage optimization algorithm framework based on Bayesian optimization algorithm and whale optimization algorithm of the present application is shown in the figure. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0091] The present embodiment provides a dual-fuel ship energy efficiency operation index prediction method based on physical data hybrid driving, as shown in the figure, which includes: Figure 1

[0092] ​S1: collect historical energy efficiency data of a liquefied natural gas dual-fuel ship, perform data preprocessing to obtain preprocessed energy efficiency data; the historical energy efficiency data includes ship historical fuel consumption data, ship historical operation data and sailing environment data;

[0093] S2: perform feature selection on the preprocessed energy efficiency data to obtain input feature data;

[0094] S3: construct a physical mechanism model of the dual-fuel ship according to the input feature data to predict the fuel consumption;

[0095] S4: introduce a bidirectional recurrent neural network, construct a two-stage optimization algorithm framework of Bayesian optimization algorithm and whale optimization algorithm, optimize the hyperparameters of the bidirectional recurrent neural network, and obtain a data-driven model based on the two-stage optimization algorithm framework, the data-driven model based on the two-stage optimization algorithm framework is used to receive the input feature data and the error between the ship historical fuel consumption data and the fuel consumption predicted based on the physical mechanism model, and output the minimum error of the fuel consumption;

[0096] S5: combine the data-driven model based on the two-stage optimization algorithm framework with the physical mechanism model to form a physical data hybrid driving model; add the fuel consumption predicted based on the physical mechanism model and the minimum error of the fuel consumption output by the data-driven model based on the two-stage optimization algorithm framework through the physical data hybrid driving model to obtain the fuel consumption prediction result based on the physical data hybrid driving model, and then calculate the dual-fuel ship energy efficiency operation index prediction result according to the fuel consumption prediction result output by the physical data hybrid driving model.

[0097] Specifically, in the sailing process of the target ship liquefied natural gas dual-fuel ship, three types of fuels are used: marine heavy oil, marine light diesel oil and liquefied natural gas. The ship can run in dual-fuel mode. The first mode is fuel mode, which only uses marine heavy oil. The second mode is gas mode, which mainly uses liquefied natural gas, and cooperates with a small amount of marine light diesel oil as the ignition fuel to ignite liquefied natural gas, and the specific prediction process is as follows Figure 1 and Figure 2As shown, first, historical energy efficiency data of the liquefied natural gas dual-fuel ship is collected, data preprocessing and feature selection are performed, and input feature data is obtained; second, a physical mechanism model and a data-driven model based on a two-stage optimization algorithm framework are constructed, the physical mechanism model is based on a physical mechanism mechanism, the error data is obtained by subtracting the fuel consumption results predicted by the physical mechanism model from the historical fuel consumption data of the ship, and the input feature data is input into the data-driven model together, the error value predicted by the data-driven model is obtained, and finally, the predicted value of the physical mechanism model and the error value predicted by the data-driven model are added through the physical data hybrid driving model to obtain the prediction result of the finally constructed physical data hybrid driving model.

[0098] In specific embodiments, historical energy efficiency data of the liquefied natural gas dual-fuel ship is collected, data preprocessing is performed, and preprocessed energy efficiency data is obtained; the historical energy efficiency data includes ship historical fuel consumption data, ship historical operation data and sailing environment data, and the scheme is as follows:

[0099] The historical energy efficiency data of a voyage of the liquefied natural gas dual-fuel ship is collected, wherein the ship historical fuel consumption data includes marine heavy oil consumption, marine light diesel oil consumption and liquefied natural gas gas consumption, the ship historical operation data includes trim, average draft, speed, heading, shaft power and speed, and the sailing environment data includes wind direction, wind speed, wave height, current speed and current direction;

[0100] Specifically, the ship operation data used in this embodiment is collected by various sensors installed on the ship, including fuel flow meters, shaft power sensors, inclinometers and global positioning systems, and the sampling frequency is once every 10 minutes. The wind direction, wind speed and wave height data in the sailing environment data used in this embodiment come from the fifth generation reanalysis data set of the European Centre for Medium-Range Weather Forecasts, and the current speed and current direction data in the sailing environment data come from the global ocean physical analysis and forecast data set of Copernicus Marine Environmental Monitoring Service. The sampling frequency of all sailing environment data is once every hour, and the data is distributed on a latitude and longitude grid. The fifth generation reanalysis data set stores data using two different grid sizes: the grid size of the wind direction and wind speed data is 0.25°x0.25°, and the grid size of the wave height data is 0.5°x0.5°. The grid size of the current speed and current direction data of the global ocean physical analysis and forecast data set of Copernicus Marine Environmental Monitoring Service is 0.083°x0.083°;

[0101] In order to obtain the actual sailing environment state of the target ship, this embodiment performs vector synthesis calculation based on the east-west (u component) and north-south (v component) data of the wind speed and current speed provided in the fifth generation reanalysis data set and the global ocean physical analysis and forecast data set, so as to obtain the true wind speed and wind direction as well as the current speed and current direction;

[0102] The energy efficiency data is cleaned and matched in time and space by spatio-temporal interpolation technology in this embodiment:

[0103] The data cleaning technology refers to cleaning the abnormal data of the speed and the rotating speed of the ship operation data beyond the specified range. During the collection of the ship operation data, abnormal values such as zero value, negative value or outlier value may be introduced due to environmental factors, sensor failure and the like. Therefore, during the data preprocessing stage, only the samples with the speed between 5 and 16 knots and the propeller rotating speed within the range of 25 to 62.5 revolutions per minute are reserved through data screening. The samples below this range usually represent the low-speed maneuvering or acceleration stage, which shows obvious transient characteristics and sensor fluctuations. The samples exceeding the upper limit exceed the maximum continuous power rating of the main engine and cannot represent the normal navigation condition. Therefore, all the samples exceeding this operation range are regarded as abnormal data violating the physical assumption and need to be deleted.

[0104] Since the sampling frequencies of the ship operation data and the navigation environment data are different, and the ship navigation position cannot completely fall on the grid points of the fifth generation reanalysis data set and, the spatio-temporal interpolation method of cubic B-spline interpolation is adopted for spatio-temporal interpolation in this embodiment:

[0105] Firstly, the originally hourly sampled environment data is interpolated to 10-minute intervals; secondly, the spatial interpolation is performed based on the longitude and latitude coordinates of the target ship position. In this way, the ship operation data and the navigation environment data can be aligned in time and space dimensions.

[0106] In specific embodiments, the scheme for performing feature selection on the preprocessed energy efficiency data to obtain input feature data is:

[0107] In order to reduce the calculation cost and remove the redundant features irrelevant to the fuel consumptions, the Pearson correlation coefficient is used as an analysis tool to evaluate the relationship between the fuel consumptions and the input features in this embodiment. According to the analysis of the Pearson correlation coefficient, the ship speed, the trim, the average draft, the wind direction, the wind speed, the flow rate, the flow direction and the wave height have high correlation with the ship heavy fuel oil consumption, the ship light diesel oil consumption and the liquefied natural gas consumption, so these are used as input features.

[0108] In specific embodiments, the scheme for constructing a physical mechanism model of the dual-fuel ship based on the input feature data to predict the fuel consumptions is:

[0109] S31, calculating the total resistance of the ship based on the input feature data, the specific steps are as follows:

[0110] S311: calculating the frictional resistance, as shown in formula (21),

[0111] (21)

[0112] wherein, is the frictional resistance, is the seawater density, is the ship speed, is the frictional resistance coefficient, as shown in equations (22) and (23), is the wet surface area, as shown in equation (24),

[0113] (22)

[0114] (23)

[0115] (24)

[0116] wherein, is the frictional resistance coefficient, is the Reynolds number representing the ratio of inertial force to viscous force, is the ship speed, is the waterline length of the ship, is the kinematic viscosity of seawater, is the length between perpendiculars of the ship, is the displacement of the ship;

[0117] S312: calculating the residual resistance of the ship based on the Savitsky-Keller method, as shown in equation (25),

[0118] (25)

[0119] wherein, is the residual resistance, is the residual resistance coefficient determined by the Lamb-Keller diagram;

[0120] S313: calculating the appendage resistance of the ship, as shown in equation (26),

[0121] (26)

[0122] wherein, is the appendage resistance, is the appendage resistance coefficient;

[0123] S314: calculating the air resistance of the ship, as shown in equation (27),

[0124] (27)

[0125] wherein, is the air resistance, is the air resistance coefficient determined by the wind tunnel experiment, is the air density, is the wind speed, is the projected area in the direction of the ship's waterline above the cross-sectional area;

[0126] S315: Calculate the wave-induced resistance, as shown in equation (28),

[0127] (28)

[0128] where, is the wave-induced resistance, is the wave height, is the ship width, is the square coefficient, is the seawater density, is the acceleration of gravity, is the ship length;

[0129] S316: Calculate the total resistance of the ship, as shown in equation (29),

[0130] (29)

[0131] where, is the total resistance of the ship, is the frictional resistance, is the residual resistance, is the appendage resistance, is the air resistance, is the wave-induced resistance;

[0132] S32, calculate the effective power of the propeller based on the total resistance of the ship, as shown in equation (30),

[0133] (30)

[0134] where, is the effective power of the propeller, is the effective thrust generated by the propeller, is the ship speed, is the total resistance of the ship;

[0135] S33, the propulsion system of the ship is essentially an energy conversion system composed of the ship body, shafting, engine and propeller. The propulsion system of the dual-fuel ship is driven by the main engine through the shafting to drive the propeller to rotate and generate thrust to push the ship forward, so the main engine power of the ship is calculated based on the effective power of the propeller, as shown in equation (31),

[0136] (31)

[0137] where, is the main engine power, is the propeller efficiency; is the shafting transmission efficiency, which can be obtained from the target ship trial report; is the relative rotation efficiency, which can be obtained from the target ship trial report; denotes the hull efficiency;

[0138] The propeller slip efficiency is shown in equation (32),

[0139] (32)

[0140] where, is the propeller thrust, is the propeller torque, is the propeller rotational speed, is the propeller advance speed, is the thrust coefficient, is the torque coefficient, is the propeller advance speed coefficient, which is shown in equation (33),

[0141] (33)

[0142] where, is the propeller advance, i.e., the distance advanced in the axial direction per revolution of the propeller; is the propeller diameter, is the wake coefficient;

[0143] The propeller thrust is shown in equation (34),

[0144] (34)

[0145] where, is the seawater density, is the propeller rotational speed, is the propeller diameter;

[0146] When the ship sails at a constant speed, due to the wake effect and the energy loss of the propeller itself, and other factors, the total thrust generated by the propeller cannot be completely converted into the effective thrust of the propeller. The thrust loss can be represented by the thrust deduction coefficient, which is calculated based on the propeller thrust and the total resistance of the ship, as shown in equation (35),

[0147] (35)

[0148] where, is the thrust deduction coefficient, is the effective thrust of the propeller;

[0149] The hull efficiency is calculated based on the thrust deduction coefficient and the wake coefficient, as shown in equation (36),

[0150] (36)

[0151] wherein, is the hull efficiency, is the wake fraction;

[0152] S34, the consumption of each fuel is calculated based on the main engine power, as shown in equations (37)-(39),

[0153] (37)

[0154] (38)

[0155] (39)

[0156] wherein, is the main engine power; , , are the marine heavy fuel oil consumption rate, marine light diesel oil consumption rate and liquefied natural gas consumption rate of the ship's main engine, respectively, which change with the main engine power, wherein, , and the relationship between the fuel consumption rate curve provided in the main engine bench test report; is the ship navigation time, , , is the marine heavy fuel oil consumption, marine light diesel oil consumption and liquefied natural gas consumption predicted by the physical mechanism model.

[0157] In this scheme, the physical mechanism model is based on physical principles such as resistance formula and propulsion power equation, which provides a solid theoretical basis for the prediction results.

[0158] In specific embodiments, a bidirectional recurrent neural network is introduced, a two-stage optimization algorithm framework of Bayesian optimization algorithm and whale optimization algorithm is constructed, the hyperparameters of the bidirectional recurrent neural network are optimized, and a data-driven model based on the two-stage optimization algorithm framework is obtained, the data-driven model based on the two-stage optimization algorithm framework is used to receive input feature data and ship historical fuel consumption data and the error of each fuel consumption based on the physical mechanism model prediction, and the scheme with the minimum error of each fuel consumption is output:

[0159] The optimization process is shown in Figure 3 , and the specific steps are as follows:

[0160] S41: divide the error of the input feature data and the ship historical fuel consumption data and the fuel consumption based on the physical mechanism model prediction into a training set and a test set;

[0161] In this embodiment, the errors between the historical fuel consumption data of ships and the fuel consumption predicted based on the physical mechanism model, namely the error of marine heavy oil consumption, marine light diesel oil consumption, and liquefied natural gas consumption, are divided into three training sets and test sets with the historical energy efficiency data corresponding to the input features in an 8:2 ratio.

[0162] S42: Use the Bayesian optimization algorithm to randomly initialize the parameters of the whale optimization algorithm. The parameters of the whale optimization algorithm include whale population size, linear decrease factor, coefficient vector, random vector, probability parameter and spiral parameter.

[0163] S43: Initialize the whale population, with each individual representing a combination of bidirectional recurrent neural network parameters; train the bidirectional recurrent neural network corresponding to the parameter combination of the initial whale population using the training set, and calculate the root mean square error of the error between the historical fuel consumption data of the ship and the fuel consumption predicted based on the physical mechanism model on the test set, as the fitness of the initial whale individual.

[0164] Specifically, the three training sets are input into the bidirectional recurrent neural network corresponding to each combination of bidirectional recurrent neural network parameters for training, and the root mean square error of the corresponding error is calculated on the three test sets respectively.

[0165] S44: Compare the fitness of all initial whale individuals, and select the position of the whale individual with the lowest fitness as the global optimal position of the initial whale population, which is the current optimal combination of bidirectional recurrent neural network parameters.

[0166] S45: Update the coefficient vector of the whale optimization algorithm and Linear decreasing factor Random vectors and probability parameters And spiral parameters, to control the search direction and stride of an individual whale;

[0167] S46: For each individual whale, based on probability parameters With coefficient vector Length of the module Three different location update strategies are implemented, as follows:

[0168] When the probability parameter is greater than or equal to the set threshold, the "spiral position update" operation is executed, that is, the individual whale moves along the spiral trajectory around the global optimal solution to update its position;

[0169] When the probability parameter is less than the set threshold value and the modulus of the coefficient vector is less than the preset value, a "surround prey" operation is performed, i.e., the whale individual converges inwardly with the current global optimal solution as the center to update the position;

[0170] When the probability parameter is less than the set threshold value and the modulus of the coefficient vector is greater than or equal to the preset value, a "random search prey" operation is performed, i.e., the whale individual updates the position by randomly selecting a reference individual for global exploration;

[0171] In this embodiment, the threshold value of the probability parameter is 0.5, and the preset value of the modulus is 1.

[0172] S47: A new bidirectional recurrent neural network parameter combination is generated according to the updated whale individual position, a new fitness is calculated in the manner of step S43, and the global optimal position of the whale population is updated according to the new fitness;

[0173] S48: The steps of S43-S47 are repeated until the termination condition of the whale optimization algorithm is reached, i.e., the root mean square error tends to converge or the whale optimization algorithm reaches the maximum number of iterations, and the global optimal position obtained by the whale optimization algorithm is output;

[0174] S49: It is judged whether the termination condition of the Bayesian optimization algorithm is reached, i.e., the Bayesian optimization algorithm reaches the maximum number of iterations or the root mean square error tends to converge. If not, the proxy model based on the tree structure Parzen estimation in the Bayesian algorithm is updated according to the global optimal position obtained by the whale optimization algorithm at this time, a new set of whale optimization algorithm parameters is resampled for the next iteration. If yes, the iteration is ended, and the minimum error of each fuel consumption and the global optimal position finally obtained by the whale optimization algorithm, i.e., the final bidirectional recurrent neural network parameter combination, are output. The minimum error of each fuel consumption includes the minimum error of marine heavy oil consumption, the minimum error of marine light diesel oil consumption, and the minimum error of liquefied natural gas consumption.

[0175] The physical mechanism model predicts the fuel consumption through known physical mechanisms such as resistance formula and propulsion power equation, and provides clear physical basis for fuel consumption prediction. However, due to the variability of ship operation and environmental conditions, the physical mechanism model cannot capture the complex nonlinear relationship in the ship operation data and navigation environment data. Therefore, the prediction value of the physical mechanism model often has certain error. The data-driven model can capture the complex nonlinear relationship in the ship operation data and navigation environment data that the physical mechanism model cannot capture through learning a large amount of historical data, thereby correcting the output result of the physical mechanism model by minimizing the error between the prediction value of the physical mechanism model and the actual measured value, so that the final prediction result of each fuel of the target liquefied natural gas dual-fuel ship is closer to the actual situation.

[0176] In a specific embodiment, a data-driven model based on a two-level optimization algorithm framework is combined with a physical mechanism model to form a physical-data hybrid driving model. The minimum error between the fuel consumption predicted by the physical mechanism model and the fuel consumption output by the data-driven model based on the two-level optimization algorithm framework is added through the physical-data hybrid driving model to obtain the fuel consumption prediction results based on the physical-data hybrid driving model. The scheme for calculating the predicted energy efficiency operation index of a dual-fuel ship based on the fuel consumption prediction results output by the physical-data hybrid driving model is as follows:

[0177] S51: Using a physical mechanism model to predict the fuel consumption of marine heavy fuel oil, marine light diesel oil, and liquefied natural gas based on the input feature data, the predicted values ​​are obtained. , , ;

[0178] S52: The error between the predicted values ​​of marine heavy fuel oil consumption, marine light diesel oil consumption, and liquefied natural gas consumption obtained from S51 based on the physical mechanism model and the historical marine heavy fuel oil consumption, marine light diesel oil consumption, and historical liquefied natural gas consumption of ships. , , The input feature data, along with the input data, are fed into a data-driven model based on a two-level optimization algorithm framework. The output is the minimum error of each fuel consumption value. , , ;

[0179] S53: Combine the data-driven model based on the two-level optimization algorithm framework with the physical mechanism model to form a physical-data hybrid driving model; add the predicted values ​​of marine heavy fuel oil consumption, marine light diesel oil consumption, and liquefied natural gas consumption obtained from the physical mechanism model in S51 to the minimum error of each fuel consumption obtained from the output of the data-driven model based on the two-level optimization algorithm framework in S52 through the physical-data hybrid driving model to obtain the final prediction results of each fuel consumption based on the physical-data hybrid driving model. , , .

[0180] The predicted results of the energy efficiency operation index of the dual-fuel ship are calculated based on the fuel consumption prediction results output by the hybrid drive model of the final physical data, as shown in formula (40).

[0181] (40)

[0182] in, For dual-fuel ship energy efficiency operation indicators, , , respectively represent the carbon dioxide emission factors of marine heavy oil, marine light diesel oil and liquefied natural gas, 、 and represent the prediction results of the marine heavy oil, marine light diesel oil and liquefied natural gas consumptions finally based on the physical data hybrid driving model outputs, represents the mass of the carried cargo, represents the transportation distance of the carried cargo.

[0183] In summary, the present application can make the energy efficiency operation index prediction accuracy of the target liquefied natural gas dual-fuel ship high, that is, the relative error is small, which benefits from the effective combination of the physical mechanism model and the data driven model, and also benefits from the accuracy of the two-stage optimization algorithm in the hyperparameter optimization process. The physical mechanism model provides a solid physical theoretical basis for fuel consumption prediction of the dual-fuel ship by using physical mechanisms such as resistance formula and propulsion power equation, while the two-stage optimization algorithm in the data driven model can efficiently and accurately find the bidirectional recurrent neural network parameter combination suitable for predicting the fuel consumption of the target liquefied natural gas dual-fuel ship, thereby correcting the error between the physical mechanism model prediction value and the actual measured value, making the final fuel prediction results of the target liquefied natural gas dual-fuel ship based on the physical data hybrid driving model more accurate, and further making the relative error of the energy efficiency operation index prediction of the target dual-fuel ship smaller.

[0184] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting energy efficiency operation indicators of dual-fuel ships based on physical data hybrid driving, characterized in that, include: S1: Collect historical energy efficiency data of liquefied natural gas dual-fuel vessels, perform data preprocessing, and obtain preprocessed energy efficiency data; the historical energy efficiency data includes historical fuel consumption data of the vessel, historical operating data of the vessel, and navigation environment data. S2: Perform feature selection on the preprocessed energy efficiency data to obtain input feature data; S3: Construct a physical mechanism model of dual-fuel ships based on the input feature data to predict the consumption of each fuel; S4: Introduce a bidirectional recurrent neural network to construct a two-level optimization algorithm framework of Bayesian optimization algorithm and whale optimization algorithm. Perform hyperparameter optimization on the bidirectional recurrent neural network to obtain a data-driven model based on the two-level optimization algorithm framework. The data-driven model based on the two-level optimization algorithm framework is used to receive the input feature data and the error between the historical fuel consumption data of the ship and the fuel consumption predicted based on the physical mechanism model, and output the minimum error of each fuel consumption. S5: The data-driven model based on the two-level optimization algorithm framework is combined with the physical mechanism model to form a physical-data hybrid-driven model. The minimum error between the fuel consumption predicted by the physical mechanism model and the fuel consumption output by the data-driven model based on the two-level optimization algorithm framework is added through the physical-data hybrid-driven model to obtain the fuel consumption prediction results based on the physical-data hybrid-driven model. Then, based on the fuel consumption prediction results output by the physical-data hybrid-driven model, the predicted results of the dual-fuel ship energy efficiency operation indicators are calculated, including: S51: Use the physical mechanism model to make predictions based on the input feature data to obtain the predicted values ​​of each fuel consumption based on the physical mechanism model; S52: Input the predicted values ​​of each fuel consumption obtained by S51 based on the physical mechanism model and the error and input characteristic data of each historical fuel consumption data of the ship into the data-driven model based on the two-level optimization algorithm framework, and output the minimum error of each fuel consumption. S53: Combine the data-driven model based on the two-level optimization algorithm framework with the physical mechanism model to form a physical-data hybrid driving model; add the predicted values ​​of each fuel consumption obtained from the physical mechanism model in S51 with the minimum error of each fuel consumption obtained from the output of the data-driven model based on the two-level optimization algorithm framework in S52 through the physical-data hybrid driving model to obtain the final prediction results of each fuel consumption based on the physical-data hybrid driving model.

2. The method for predicting energy efficiency operation indicators of dual-fuel ships based on physical data hybrid driving as described in claim 1, characterized in that, A bidirectional recurrent neural network is introduced to construct a two-level optimization algorithm framework combining Bayesian optimization and whale optimization algorithms. Hyperparameter optimization is performed on the bidirectional recurrent neural network to obtain a data-driven model based on this two-level optimization framework. This data-driven model receives the input feature data and the error between historical fuel consumption data of the ship and the fuel consumption predicted by the physical mechanism model, and outputs the minimum error for each fuel consumption, including: S41: Divide the input feature data and the error between the historical fuel consumption data of the ship and the fuel consumption predicted based on the physical mechanism model into a training set and a test set; S42: Use the Bayesian optimization algorithm to randomly initialize the parameters of the whale optimization algorithm. The parameters of the whale optimization algorithm include whale population size, linear decrease factor, coefficient vector, random vector, probability parameter and spiral parameter. S43: Initialize the whale population, with each individual representing a combination of bidirectional recurrent neural network parameters; train the bidirectional recurrent neural network corresponding to the parameter combination of the initial whale population using the training set, and calculate the root mean square error of the error between the historical fuel consumption data of the ship and the fuel consumption predicted based on the physical mechanism model on the test set, as the fitness of the initial whale individual. S44: Compare the fitness of all initial whale individuals, and select the position of the whale individual with the lowest fitness as the global optimal position of the initial whale population, which is the current optimal combination of bidirectional recurrent neural network parameters. S45: Update the coefficient vector, linear decrease factor, random vector, probability parameter, and spiral parameter of the whale optimization algorithm; S46: For each individual whale, three different position update strategies are executed based on the probability parameters and the magnitude of the coefficient vector. The specific strategies are as follows: When the probability parameter is greater than or equal to the set threshold, the "spiral position update" operation is executed, that is, the individual whale moves along the spiral trajectory around the global optimal solution to update its position; When the probability parameter is less than the set threshold and the magnitude of the coefficient vector is less than the preset value, the "surround the prey" operation is executed, that is, the whale individual converges inward with the current global optimal solution as the center to update its position; When the probability parameter is less than the set threshold and the magnitude of the coefficient vector is greater than or equal to the preset value, the "random prey search" operation is executed, that is, the whale updates its position by randomly selecting a reference individual to conduct a global exploration. S47: Generate a new combination of bidirectional recurrent neural network parameters based on the updated individual whale positions, calculate the new fitness in the manner of step S43, and update the global optimal position of the whale population based on it. S48: Repeat steps S43-S47 until the termination condition of the whale optimization algorithm is met, i.e., the root mean square error tends to converge or the whale optimization algorithm reaches the maximum number of iterations, and output the global optimal position obtained by the whale optimization algorithm. S49: Determine whether the termination condition of the Bayesian optimization algorithm has been met, i.e., the Bayesian optimization algorithm has reached the maximum number of iterations or the root mean square error has converged. If not, update the surrogate model in the Bayesian algorithm according to the global optimal position obtained by the whale optimization algorithm at this time, and resample a new set of whale optimization algorithm parameters to enter the next round of iteration. If the condition is met, end the iteration, output the minimum error of each fuel consumption and the global optimal position finally obtained by the whale optimization algorithm, i.e., the final combination of bidirectional recurrent neural network parameters; the minimum error of each fuel consumption includes the minimum error of marine heavy oil consumption, the minimum error of marine light diesel oil consumption, and the minimum error of liquefied natural gas consumption.

3. The method for predicting energy efficiency operation indicators of dual-fuel ships based on physical data hybrid driving as described in claim 1, characterized in that, A physical mechanism model of a dual-fuel ship is constructed based on the input feature data to predict the consumption of each fuel, including: S31. Calculate the total resistance of the ship based on the input feature data; S32. Calculate the effective power of the propeller based on the total resistance of the ship, as shown in formula (1). （1） in, The effective power of the propeller, The effective thrust generated by the propeller For ship speed, This represents the total resistance of the ship. S33. Calculate the ship's main engine power based on the propeller's effective power and the ship's energy efficiency data, as shown in formula (2). （2） in, This refers to the power of the main unit; For propeller water flow efficiency; For shaft transmission efficiency; For relative rotational efficiency; Indicates hull efficiency; The propeller's water flow efficiency is shown in formula (3). （3） in, For the thrust of the propeller, For the propeller torque, The rotational speed of the propeller. The propeller's advance speed. For thrust coefficient, The torque coefficient, Let be the propeller advance coefficient, as shown in formula (4). （4） in, For the propeller's progress; The diameter of the propeller. Here is the wake coefficient; The thrust of the propeller is shown in formula (5). （5） in, The density of seawater, The rotational speed of the propeller. The diameter of the propeller; The thrust reduction factor is calculated based on the propeller thrust and the total ship resistance, as shown in formula (6). （6） in, This is the thrust reduction factor. This is the effective thrust of the propeller; The hull efficiency is calculated based on the thrust reduction factor and the wake factor, as shown in formula (7). （7） in, For hull efficiency, Here is the wake coefficient; S34. Calculate the fuel consumption based on the main engine power, as shown in formulas (8)-(10). （8） （9） （10） in, This refers to the power of the main unit; , , These are the marine heavy fuel oil consumption rate, marine light diesel oil consumption rate, and liquefied natural gas consumption rate of the ship's main engine, respectively. For the ship's sailing time, , , The fuel consumption of marine heavy oil, marine light diesel oil, and liquefied natural gas are predicted by the physical mechanism model.

4. The method for predicting energy efficiency operation indicators of dual-fuel ships based on physical data hybrid driving as described in claim 3, characterized in that, The total resistance of a ship is calculated based on the input feature data, including: S311: Calculate the frictional resistance, as shown in formula (11). （11） in, For frictional resistance, The density of seawater, For ship speed, Let be the coefficient of frictional resistance, as shown in formulas (12) and (13). Let be the wetted surface area, as shown in formula (14). （12） （13） （14） in, The coefficient of frictional resistance. The Reynolds number represents the ratio of inertial force to viscous force. For ship speed, Because of the ship's waterline length, The kinematic viscosity of seawater, The length between the perpendiculars of the ship. The ship's displacement; S312: The residual resistance of the ship is calculated based on the Savitsky-Keller method, as shown in formula (15). （15） in, For the remaining resistance, This is the residual drag coefficient; S313: Calculate the appendage resistance of the ship, as shown in formula (16). （16） in, For attached resistance, This refers to the drag coefficient of the attached structure; S314: Calculate the air resistance of the ship, as shown in formula (17). （17） in, For air resistance, The air drag coefficient, air density, For wind speed, It is the projected area of ​​the transverse section above the waterline of the ship. S315: Calculate wave resistance as shown in formula (18). （18） in, To increase resistance to waves, For wave height, For the width of the boat, The square coefficient, The density of seawater, It is the acceleration due to gravity. Captain; S316: Calculate the total resistance of the ship, as shown in formula (19). （19） in, For the total resistance of the ship, For frictional resistance, For the remaining resistance, For attached resistance, For air resistance, To increase resistance to waves.

5. The method for predicting energy efficiency operation indicators of dual-fuel ships based on physical data hybrid driving as described in claim 1, characterized in that, Based on the fuel consumption prediction results output by the physical data hybrid drive model, the predicted results of the dual-fuel ship energy efficiency operation indicators are calculated, including: The predicted results of the energy efficiency operation index of the dual-fuel ship are calculated based on the fuel consumption prediction results output by the hybrid drive model of the final physical data, as shown in formula (20). （20） in, For dual-fuel ship energy efficiency operation indicators, , , These represent the carbon dioxide emission factors for marine heavy fuel oil, marine light diesel oil, and liquefied natural gas, respectively. 、 and This represents the final predicted consumption results of marine heavy fuel oil, marine light diesel oil, and liquefied natural gas, based on the physical data-driven hybrid model. Indicates the mass of the goods carried. This indicates the transportation distance of the goods being transported.

6. The method for predicting energy efficiency operation indicators of dual-fuel ships based on physical data hybrid driving according to claim 1, characterized in that, The Pearson correlation coefficient was used for feature selection to obtain the input feature data.

Images