Dry bulk cargo ship displacement measuring and calculating method based on bending moment-deflection equal ratio model

Abstract

The invention discloses a method for measuring and calculating the displacement of a dry bulk cargo ship based on a bending moment-deflection equal ratio model. The method comprises the steps that S1, multi-source ship data used for measuring and calculating the displacement of the dry bulk cargo ship are acquired; s2, obtaining a ship longitudinal bending moment differential equation about a ship beam deflection function based on an elastic beam theory; S3, constructing a bending moment-deflection equal ratio model for correcting ship deflection to obtain ship section deflection distribution data; s4, taking the longitudinal bending moment distribution data and the corresponding ship section deflection distribution data as data samples to obtain a data sample set; s5, constructing a network comprehensive loss function with ship physical constraints and deflection offset supervision loss; a ship deflection prediction network model is obtained based on the multi-layer perceptron neural network; and S6, obtaining corrected ship deflection data according to the ship deflection prediction network model, and obtaining a corrected waterline according to the corrected ship deflection data so as to measure and calculate the displacement of the dry bulk ship. The problem that a traditional metering method is difficult to meet the high-precision transportation accounting requirement is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship load measurement and displacement calculation, and particularly to a method for calculating the displacement of dry bulk carriers based on a bending moment-deflection ratio model. Background Art

[0002] In dry bulk cargo transportation, traditional draft survey is generally based on Archimedes' principle and the measured data of six-sided draft marks (bow left, bow right, midship left, midship right, stern left, stern right). The displacement is calculated by referring to the ship's hydrostatic table or load line data, and factors such as trim, hogging, and port water density difference are corrected to a certain extent. Most existing methods use a "1:6:1" weighting or other empirical formulas in order to obtain a more accurate estimate of the displacement. In addition, the loading computer can generally provide longitudinal bending moment information for analyzing the overall force state of the ship. However, in the traditional draft survey process, these bending moment data are often not organically combined with the correction of hull deformation (such as hogging and sagging).

[0003] However, with the continuous deepening of the trend towards larger dry bulk carriers, the hull often exhibits significant hogging or sagging deformation under heavy load conditions. Traditional draft survey is limited by the limited number of measurement points and is difficult to fully reflect the local draft differences caused by longitudinal bending, often resulting in deviations of hundreds of tons or even more. In addition, most existing processes lack a real-time correction mechanism for bending moment data and cannot make full use of the bending moment curve provided by the ship's loading computer to offset the structural deformation error. Coupled with the fact that the ship's bending moment-deflection data accumulated from historical voyages have not been systematically utilized, it further restricts the improvement of the accuracy of displacement and load correction. In summary, traditional measurement methods are difficult to meet the requirements of high-precision transportation accounting, and there is an urgent need for a new solution that can combine bending moment information and dynamically correct hogging and sagging errors. Summary of the Invention

[0004] The present invention provides a method for calculating the displacement of dry bulk carriers based on a bending moment-deflection ratio model to overcome the above technical problems.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] A method for calculating the displacement of dry bulk carriers based on a bending moment-deflection ratio model specifically includes the following steps:

[0007] S1: Obtain multi-source ship data for calculating the displacement of dry bulk carriers;

[0008] And the multi-source ship data includes historical ship measurement data under different working conditions, ship attribute data, and longitudinal bending moment distribution data read from the ship's loading computer;

[0009] The ship measurement data at least includes the deflection data at the longitudinal mid - position of the hull;

[0010] The ship attribute data at least includes the ship's reference design bottom line and the draft survey curve;

[0011] S2: Based on the elastic beam theory, obtain the ship's longitudinal bending moment differential equation about the ship beam deflection function,

[0012] S3: According to the ship's longitudinal bending moment differential equation combined with historical ship measurement data, construct a bending moment - deflection ratio model for correcting the ship's deflection, and obtain the ship cross - section deflection distribution data based on the bending moment - deflection ratio model;

[0013] S4: Take the longitudinal bending moment distribution data and the corresponding ship cross - section deflection distribution data as data samples to obtain a data sample set;

[0014] And randomly divide the data sample set and obtain a training set and a test set;

[0015] S5: Construct a network comprehensive loss function with ship physical constraints and deflection offset supervision loss;

[0016] And based on the multi - layer perceptron neural network, correct and train the multi - layer perceptron neural network according to the training set to obtain the corrected and trained multi - layer perceptron neural network;

[0017] Based on the network comprehensive loss function, evaluate whether the output of the corrected and trained multi - layer perceptron neural network converges according to the test set combined with the ship attribute data, and take the corrected and trained multi - layer perceptron neural network when the output converges as the ship deflection prediction network model;

[0018] S6: Obtain the corrected ship deflection data according to the ship deflection prediction network model, and obtain the corrected draft line according to the corrected ship deflection data to realize the displacement measurement of the dry - bulk carrier.

[0019] Further, S2 specifically includes the following steps:

[0020] S21: Based on the elastic beam theory, simplify the longitudinal part of the hull into an elastic beam with a span of L;

[0021] S22: Assume that the shear deformation and non - linear effect of the elastic beam are ignored, and then construct the ship's longitudinal bending moment differential equation about the ship beam deflection function;

[0022] And the expression of the ship's longitudinal bending moment differential equation is

[0023]

[0024] Where: E represents the elastic modulus; I represents the moment of inertia of the cross-section; x represents the position of any cross-section on the elastic beam; M(x) represents the longitudinal bending moment of the ship at the cross-section position; w(x) represents the deflection function of the ship beam, and the boundary conditions at both ends of the elastic beam are w(0) = 0 and w(L) = 0.

[0025] Furthermore, S3 specifically includes the following steps:

[0026] S31: Assume a reference bending moment distribution of the ship And there exists a scaling ratio coefficient α for any bending moment distribution curve to obtain the scaled bending moment distribution;

[0027] And the expression of the scaled bending moment distribution M α (x) is

[0028]

[0029] S32: According to the longitudinal bending moment differential equation of the ship, the deflection function of the reference bending moment distribution of the ship can be obtained satisfying the longitudinal bending moment differential equation of the ship reference;

[0030] And the longitudinal bending moment differential equation of the ship reference is

[0031]

[0032] S33: According to the longitudinal bending moment differential equation of the ship reference combined with S31, the deflection function w of the scaled bending moment distribution can be obtained α (x) satisfying the longitudinal bending moment differential equation of the ship scaled;

[0033] And the expression of the longitudinal bending moment differential equation of the ship scaled is

[0034]

[0035] S34: Multiply the in the longitudinal bending moment differential equation of the ship reference by the scaling ratio coefficient α, and denote it as Then

[0036]

[0037] And according to the uniqueness principle of common linear boundary value problems, it can be known that:

[0038] Furthermore, a bending moment-deflection ratio model for correcting the ship deflection can be obtained, and the expression of the bending moment-deflection ratio model is

[0039]

[0040] Where: x ref represents the position of any reference section of the ship and x ref ∈(0, L); w α (x ref ) represents the scaled bending moment distribution M α (x ref ) of the deflection function; represents the reference deflection obtained through the ship's draft marks; represents the reference bending moment value obtained based on the reference deflection ;

[0041] S35: Obtain the ship's cross-section deflection distribution data according to the bending moment-deflection ratio model, and the acquisition formula for the ship's cross-section deflection distribution data is

[0042]

[0043] Where: ε(x i ) represents the deflection at the position x i of any longitudinal cross-section of the ship; M(x i ) represents the bending moment value at the position x i of any longitudinal cross-section of the ship; ε mid represents the deflection data at the longitudinal midsection of the ship's hull measured; M mid represents the bending moment value at the longitudinal midsection of the ship's hull read based on the ship's loadometer; d FP represents the draft on the port bow; d FS represents the draft on the starboard bow; d AP represents the draft on the port stern; d AS represents the draft on the starboard stern; d MP represents the draft on the port stern; d MS represents the draft on the starboard stern.

[0044] Further, the specific steps of S4 are as follows:

[0045] S41: Define the ship's cross-section deflection distribution data obtained according to the bending moment-deflection ratio model as the deflection estimation offset

[0046] and obtain the preliminary correction curve of the ship's cross-section deflection according to the deflection estimation offset ;

[0047] And the data point sequence on the preliminary correction curve is

[0048] S42: Perform a smooth distribution process on the preliminary correction curve to obtain the beam deflection correction function for ship deflection correction And take the beam deflection correction function curve as the geometric ratio distribution data of the ship's deflection;

[0049] And the process of smoothing the preliminary correction curve includes:

[0050] Perform spline interpolation on the preliminary correction curve based on the multi-spline interpolation method to obtain the interpolation curve;

[0051] Adopt the method of polynomial fitting, fit according to the interpolation curve and obtain the beam deflection correction function;

[0052] S43: Take the longitudinal bending moment distribution data of the ship as the feature data, and take the corresponding geometric ratio distribution data of the ship's deflection as the label data to obtain the data sample set;

[0053] Randomly divide the data sample set and obtain the training set and the test set.

[0054] Furthermore, the network comprehensive loss function constructed in S5 with ship physical constraints and deflection offset supervision loss, its expression is:

[0055]

[0056]

[0057] In the formula: Represents the network comprehensive loss function; Represents the supervision loss of the measured deflection offset at the longitudinal position of the hull; ε obs,i Represents the longitudinal position x of the hull obs,i The measured deflection at the place; w θ (x obs,i ) Represents the predicted ship deflection at the longitudinal position x of the hull output by the multi-layer perceptron neural network obs,i At the place; Represents the physical residual loss of the deflection at the longitudinal position of the hull; x r Represents the residual sampling point at the longitudinal position of the hull and r(x r ) Represents the physical residual term, M(x r ) Represents the longitudinal bending moment of the ship at the residual sampling point; and w θ″ (x r ) Represents the simplified form obtained by using automatic differentiation Of.

[0058] Furthermore, the method for obtaining the ship deflection prediction network model in S5 specifically includes:

[0059] S51: Take the longitudinal bending moment distribution data of the ship as the feature input data, and take the corresponding ship cross-section deflection distribution data as the label output data;

[0060] S52: Randomly initialize the network parameter weights of the multi-layer perceptron neural network, and combine with the Adam optimization algorithm to correct and train the multi-layer perceptron neural network according to the training set, and obtain the multi-layer perceptron neural network after the correction training;

[0061] S53: Based on the network comprehensive loss function, evaluate whether the output of the multi-layer perceptron neural network after the correction training converges according to the test set combined with the ship attribute data;

[0062] And the method for evaluating whether the output of the multi-layer perceptron neural network after the correction training converges specifically includes:

[0063] S001: Obtain the ship cross-section deflection distribution data of the output of the multi-layer perceptron neural network evaluated at this time, and based on the ship cross-section deflection distribution data combined with the ship baseline design bottom line, obtain the corrected draft line;

[0064] And the formula for obtaining the corrected draft line is

[0065]

[0066] In the formula: z actual (x) corrected draft line; z design (x) represents the baseline design bottom line; represents the ship cross-section deflection distribution data of the output of the multi-layer perceptron neural network evaluated after the correction training;

[0067] S002: Based on the displacement acquisition strategy, obtain the current displacement according to the corrected draft line;

[0068] And the displacement acquisition strategy includes any one of the water level curve method, the sectional integration method or the molded line data integration method;

[0069] Water level curve method: Perform linear interpolation operation on the preset draft-displacement query table according to the corrected draft line to obtain the current displacement;

[0070] Sectional integration method: Based on the water level measurement curve, query and confirm the underwater area of the ship longitudinal section according to the corrected draft line, and obtain the current displacement according to the underwater area of the ship longitudinal section;

[0071] And the expression for obtaining the current displacement according to the sectional integration method is

[0072]

[0073] In the formula: Δ represents the current displacement; A(T real (x)) represents the underwater area of the ship longitudinal section;

[0074] If so, the multi-layer perceptron neural network after corrected training when the output converges is used as the ship deflection prediction network model;

[0075] Otherwise, adaptively adjust the network weight parameters in the multi-layer perceptron neural network after corrected training based on the backpropagation method, and repeat the execution of S52.

[0076] The present invention provides a method for calculating the displacement of dry bulk carriers based on the bending moment-deflection ratio model, and the beneficial effects are as follows:

[0077] (1) By introducing the bending moment data of the loading instrument on the basis of the traditional six-sided water gauge and making an overall correction of the ship's cross-section deflection through the constructed bending moment-deflection ratio model, it can more truly reflect the ship's draft deformation caused by the hull bending moment and reduce the measurement error at the level of hundreds of tons.

[0078] (2) Construct a network comprehensive loss function with ship physical constraints and deflection offset supervision loss; train and obtain a ship deflection prediction network model based on the network comprehensive loss function; make full use of the longitudinal bending moment distribution data read from the ship's loading instrument and the historical voyage error information, and the correction of the ship's displacement can be completed online or semi-real-time without adding a large number of additional sensors or modifying the main equipment.

[0079] (3) By methods such as physical constraint neural network or polynomial fitting, combine the elastic beam theory with multi-source data (including bending moment, historical measurement error, and measured draft) to form a model framework that can continuously learn and update, so as to maintain a high measurement accuracy of the displacement under different working conditions and ship age conditions. Description of the Drawings

[0080] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0081] Figure 1 It is a flowchart of the method for calculating the displacement of dry bulk carriers based on the bending moment-deflection ratio model of the present invention;

[0082] Figure 2 It is a curve graph of the longitudinal bending moment distribution data read from the ship's loading instrument in this embodiment;

[0083] Figure 3 It is a curve graph of the ship's cross-section deflection distribution data obtained through the bending moment-deflection ratio model in this embodiment;

[0084] Figure 4This is the draft line correction comparison simulation diagram in this embodiment. Specific implementation mode

[0085] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0086] This embodiment provides a method for calculating the displacement of dry bulk carriers based on a bending moment-deflection ratio model, as Figure 1 shown, which specifically includes the following steps:

[0087] S1: Obtain multi-source ship data for calculating the displacement of dry bulk carriers;

[0088] And the multi-source ship data includes historical ship measurement data under different working conditions, ship attribute data, and longitudinal bending moment distribution data read based on a ship stowage factor;

[0089] In this embodiment, the longitudinal bending moment distribution {(x k , M k )} under the current working condition is obtained from a ship stowage factor or historical voyage records, where x k is the coordinate point discretized in the ship length direction, and M k is the corresponding bending moment value; if the bending moment is originally sparsely distributed, a denser longitudinal bending moment distribution data sequence (x, M) can also be obtained through an interpolation method (linear / spline, etc.); if there is multi-condition / multi-voyage data, a richer data set can be formed for subsequent training and verification of the neural network;

[0090] The ship measurement data at least includes deflection data at the longitudinal middle position of the hull;

[0091] In this embodiment, in-situ measurement is carried out at the ship midsection (x = L / 2) to obtain the midsection deflection ε mid or "hog / sag offset"; if the ship only has conventional six-sided draft marks, the midsection offset can also be inferred by calculating the longitudinal line difference; if there are more measurement points (such as at the bow, stern, or other sections) observation data, then ε obs,i is recorded and used for subsequent network training and accuracy verification;

[0092] The ship attribute data at least includes the ship's reference design baseline and the draft survey curve;

[0093] Among them, the method for obtaining multi-source ship data for calculating the displacement of dry bulk ships in this embodiment is a well-known prior art method, and will not be elaborated here;

[0094] S2: Based on the elastic beam theory, obtain the longitudinal bending moment differential equation of the ship regarding the beam deflection function, specifically including the following steps:

[0095] S21: Based on the elastic beam theory, simplify the longitudinal hull into an elastic beam with a span length of L;

[0096] S22: Assume that the influence of shear deformation and non-linear effects of the elastic beam is ignored, and then construct the longitudinal bending moment differential equation of the ship regarding the beam deflection function;

[0097] And in the coordinate x ∈ [0, L], the expression of the longitudinal bending moment differential equation of the ship is

[0098]

[0099] In the formula: E represents the elastic modulus; I represents the moment of inertia of the cross-section, and the product EI of the two is regarded as a constant (and the beam cross-section and material do not change with x under this assumption); x represents the position of any cross-section on the elastic beam; M(x) represents the longitudinal bending moment of the ship at the cross-section position; w(x) represents the beam deflection function of the ship, and the boundary conditions at both ends of the elastic beam are w(0) = 0, w(L) = 0.

[0100] S3: According to the longitudinal bending moment differential equation of the ship combined with historical ship measurement data, construct a bending moment-deflection ratio model for correcting the ship deflection, and obtain the ship cross-section deflection distribution data based on the bending moment-deflection ratio model;

[0101] Specifically including the following steps:

[0102] S31: Assume a reference bending moment distribution of the ship And there exists a scaling ratio coefficient α of any bending moment distribution curve to obtain the scaled bending moment distribution; where the scaling of the bending moment distribution curve in this embodiment specifically means that when the load intensity changes but the "distribution shape" remains unchanged, the scaled bending moment distribution can be obtained, which means physically that the "shape" of the longitudinal load of the ship remains unchanged, but is only obtained by multiplying by a certain magnification (or reduction) coefficient α as a whole;

[0103] And the expression of the scaled bending moment distribution M α (x) is

[0104]

[0105] S32: According to the longitudinal bending moment differential equation of the ship, the deflection function of the reference bending moment distribution of the ship can be obtained of The satisfied differential equation of the reference bending moment of the ship;

[0106] And the differential equation of the reference bending moment of the ship is

[0107]

[0108] In this embodiment, the differential equation of the reference bending moment of the ship satisfies specific boundary conditions (such as w(0) = 0, etc.), and taking it as the only solution of the assumed differential equation of the reference bending moment of the ship is the problem model;

[0109] S33: According to the differential equation of the reference bending moment of the ship combined with S31, the scaled bending moment distribution of the deflection function w α (x) satisfies the differential equation of the scaled bending moment of the ship;

[0110] And the expression of the differential equation of the scaled bending moment of the ship is

[0111]

[0112] S34: Multiply in the differential equation of the reference bending moment of the ship by the scaling factor α, and denote it as Then we can get

[0113]

[0114] In this embodiment satisfies the same differential equation and boundary values as Equation (4) (if the linear boundary conditions are also satisfied, that is And according to the uniqueness principle of common linear boundary value problems, it can be known that:

[0115] Generally speaking, this embodiment can extend the above principle to any longitudinal section of the ship. In fact, for any reference section x ref ∈(0, L), the same conclusion can be drawn: as long as it is assumed that the longitudinal bending moment distribution of the ship is a pure amplitude scaling, and the stiffness of the elastic beam and the boundary conditions remain unchanged, then the reference deflection obtained from the draft position and the reference bending moment value can both construct a geometric ratio relationship with the deflection w(x ref ) and the bending moment M(x ref ) at any other position, that is, a bending moment-deflection geometric ratio model for correcting the ship's deflection can be obtained;

[0116] And the expression of the bending moment-deflection geometric ratio model is

[0117]

[0118] Where: x ref represents the position of an arbitrary reference section of the ship, and x ref ∈(0, L); w α (x ref ) represents the scaled bending moment distribution M α (x ref ) of the deflection function; represents the reference deflection obtained from the ship's draft position; represents the reference bending moment value obtained based on the reference deflection ;

[0119] S35: In the actual loading process of this embodiment, the measured deflection at the longitudinal middle position of the hull is more easily obtained. The measured deflection ε mid at the longitudinal middle position of the hull can be estimated using the drafts on the left and right sides of the bow (d FP , d FS ), the drafts on the left and right sides of the midship (d MP , d MS ), and the drafts on the left and right sides of the stern (d AP , d AS ). Then, according to the bending moment-deflection ratio model, the deflection distribution data of the ship's cross-section can be obtained;

[0120] And the acquisition formula for the deflection distribution data of the ship's cross-section is

[0121]

[0122] Where: ε(x i ) represents the deflection at the position x i of any longitudinal cross-section of the ship; M(x i ) represents the bending moment value at the position x i of any longitudinal cross-section of the ship; ε mid represents the deflection data at the longitudinal middle position of the hull measured; M mid represents the bending moment value at the longitudinal middle position of the hull read from the ship's loadometer; d FP represents the draft on the left side of the bow; d FS represents the draft on the right side of the bow; d AP represents the draft on the left side of the stern; d AS represents the draft on the right side of the stern; d MP represents the draft on the left side of the stern; d MS represents the draft on the right side of the stern;

[0123] S4: Use the longitudinal bending moment distribution data and the corresponding ship cross-section deflection distribution data as data samples to obtain a data sample set, and randomly divide the data sample set to obtain a training set and a test set;

[0124] Specifically, the following steps are included:

[0125] S41: In this embodiment, in the known bending moment distribution, find the bending moment corresponding to the longitudinal middle position of the hull For any cross-section x The ship cross-section deflection distribution data obtained according to the bending moment-deflection ratio model is defined as the deflection estimation offset i And according to the deflection estimation offset

[0126] Obtain the preliminary correction curve of the ship cross-section deflection;

[0127] And the data point sequence on the preliminary correction curve is And denoted as Indicating the corrected deflection;

[0128] S42: Perform a smoothing distribution process on the preliminary correction curve to obtain the beam deflection correction function for ship deflection correction And take the beam deflection correction function curve as the "zero-order approximation solution" or "initial value solution" of the geometric distribution data of the ship deflection;

[0129] And the smoothing distribution process on the preliminary correction curve includes:

[0130] Perform spline interpolation on the preliminary correction curve based on the multi-spline interpolation method to obtain the interpolation curve;

[0131] Adopt the method of polynomial fitting, fit according to the interpolation curve and obtain the beam deflection correction function;

[0132] S43: Use the ship longitudinal bending moment distribution data as the feature data, and use the corresponding geometric distribution data of the ship deflection as the label data to obtain the data sample set;

[0133] Randomly divide the data sample set and obtain the training set and the test set;

[0134] S5: Construct a network comprehensive loss function with ship physical constraints and deflection offset supervision loss;

[0135] And based on the multi-layer perceptron neural network, perform model correction training on the multi-layer perceptron neural network according to the training set to obtain the multi-layer perceptron neural network after correction training;

[0136] Specifically, the traditional multi-layer perceptron neural network includes several hidden layers, and the dimension of each layer ranges from 32 to 128. The activation function can be any one of Tanh, ReLU, Sigmoid, etc., and the input layer receives x and optional operating conditions parameters (that is, the ship longitudinal bending moment distribution data); the output layer gives the scalar w θ ​(x) (Geometric progression distribution data of ship deflection); The network can be initialized as: Taking the geometric interpolation curve as the initial value distribution to accelerate convergence; in this embodiment, the network structure of the multi-layer perceptron neural network is an existing network structure, and this embodiment only processes data based on its original network structure;

[0137] Based on the network comprehensive loss function, evaluate whether the output of the trained multi-layer perceptron neural network converges according to the test set combined with ship attribute data, and use the trained multi-layer perceptron neural network when the output converges as the ship deflection prediction network model;

[0138] Specifically, the constructed network comprehensive loss function with ship physical constraints and deflection offset supervision loss has the following expression:

[0139]

[0140]

[0141] In the formula: Represents the network comprehensive loss function; Represents the supervision loss of the measured deflection offset at the longitudinal position of the hull, that is, if there is a measured ship offset (such as bow, midship, and stern measurement points), it is recorded as ε obs,i At position x obs,i The existence of offset can increase the supervision term, where the longitudinal middle position of the hull is one of the measurement points and occupies an important weight; ε obs,i Represents the longitudinal position x of the hull obs,i The measured deflection at this position; w θ (x obs,i ) Represents the predicted ship deflection at the longitudinal position x of the hull output by the multi-layer perceptron neural network obs,i At this position; Represents the physical residual loss of the longitudinal position deflection of the hull; x r Represents the residual sampling point at the longitudinal position of the hull, and r(x r ) Represents the physical residual term, M(x r ) Represents the longitudinal bending moment of the ship at the residual sampling point; and w θ (x r ) Represents the simplified form obtained by using automatic differentiation ; λ phys Represents the learnable loss balance weight parameter; in this embodiment, the residual sampling points are uniformly distributed or randomly distributed in [0, L], for example, one point every 5 - 10 meters;

[0142] In a specific implementation, the method for obtaining the ship deflection prediction network model in S5 specifically includes:

[0143] S51: Use the longitudinal bending moment distribution data of the ship as the characteristic input data, and use the corresponding ship cross-section deflection distribution data as the label output data;

[0144] S52: Randomly initialize the network parameter weights of the multi-layer perceptron neural network, and combine the Adam optimization algorithm or the LBFGS algorithm to correct and train the multi-layer perceptron neural network according to the training set to obtain the corrected and trained multi-layer perceptron neural network;

[0145] S53: Based on the network comprehensive loss function, evaluate whether the output of the corrected and trained multi-layer perceptron neural network converges according to the test set combined with the ship attribute data;

[0146] And the method for evaluating whether the output of the corrected and trained multi-layer perceptron neural network converges specifically includes:

[0147] S001: Obtain the ship cross-section deflection distribution data output by evaluating the corrected and trained multi-layer perceptron neural network at this time, and obtain the corrected draft line based on the ship cross-section deflection distribution data combined with the ship's baseline design bottom line;

[0148] And the formula for obtaining the corrected draft line is

[0149]

[0150] In the formula: z actual (x) corrected draft line; z design (x) represents the baseline design bottom line; represents the ship cross-section deflection distribution data output by evaluating the corrected and trained multi-layer perceptron neural network;

[0151] S002: Based on the displacement acquisition strategy, obtain the current displacement according to the corrected draft line;

[0152] And the displacement acquisition strategy includes any one of the draft curve method, the sectional integration method or the molded line data integration method;

[0153] Draft curve method: Perform linear interpolation on the preset draft-displacement query table according to the corrected draft line to obtain the current displacement;

[0154] Sectional integration method: Based on the draft measurement curve, query and confirm the underwater area of the ship's longitudinal section according to the corrected draft line, and obtain the current displacement according to the underwater area of the ship's longitudinal section;

[0155] And the expression for obtaining the current displacement according to the sectional integration method is

[0156]

[0157] Where: Δ represents the current displacement; A(T real (x)) represents the underwater area of the longitudinal section of the ship;

[0158] If so, the trained multi-layer perceptron neural network after correction during output convergence is used as the ship deflection prediction network model;

[0159] Otherwise, the network weight parameters in the trained multi-layer perceptron neural network after correction are adaptively adjusted based on the backpropagation method, and S52 is repeatedly executed;

[0160] According to the current displacement, the deadweight of the dry bulk carrier can also be obtained: Deadweight = Corrected displacement - Light ship weight. If there are more operating weights (such as fuel, fresh water, etc.), they can be deducted together to obtain the "net cargo capacity";

[0161] S6: Obtain the corrected ship deflection data according to the ship deflection prediction network model, and obtain the corrected draft line according to the corrected ship deflection data, so as to realize the measurement of the displacement of the dry bulk carrier.

[0162] This embodiment also includes the online / offline deployment of the ship deflection prediction network model:

[0163] Online application: During each loading or voyage interval, collect the current bending moment distribution and interruption measurement values, and quickly execute the "geometric ratio + small amount of iteration" steps; Since the initialization is already close to the real situation (geometric ratio correction), the physical residual iteration convergence speed is relatively fast, and the correction can be completed in a short time (even at the minute level);

[0164] Offline batch training: Uniformly train the neural network with a large amount of data of {M(x), wobs(x)} measured under multiple voyages and multiple working conditions, so that it "remembers" various bending moment distributions and deflection shapes in a wider range, and can be faster and more accurate during online inference.

[0165] Incremental learning: During the operation of the ship, continuously add measured draft / deflection data. These new data can be periodically incorporated into the training set to update the weights or polynomial coefficients of the neural network, and gradually improve the adaptability to hull aging and structural deformation.

[0166] The beneficial effects of the method of this embodiment are as follows:

[0167] (1) On the basis of the traditional six-sided water gauge, the loadometer bending moment data is introduced, and the overall correction of the ship section deflection is carried out through the constructed bending moment-deflection geometric ratio model, which can more realistically reflect the ship draft deformation caused by the hull bending moment and reduce the measurement error at the level of hundreds of tons.

[0168] (2) Construct a network comprehensive loss function with ship physical constraints and deflection offset supervision loss; train and obtain a ship deflection prediction network model based on the network comprehensive loss function; make full use of the longitudinal bending moment distribution data read by the ship's loadometer and the historical voyage error information, and the correction of the ship's displacement can be completed online or semi-real-time without adding a large number of additional sensors or modifying the main equipment.

[0169] (3) Combine the elastic beam theory with multi-source data (including bending moment, historical measurement error, and measured draft) through methods such as physical constraint neural network or polynomial fitting to form a model framework that can continuously learn and update, so as to maintain a high displacement calculation accuracy under different working conditions and ship ages.

[0170] The simulation experiment of this embodiment: The data of this example comes from the measured bending moment and the ship bottom line record of a ship under typical loading conditions. The bending moment curve data is as Figure 2 shown; First, combine the measured deflection at the longitudinal middle position of the hull with the longitudinal bending moment distribution, and perform a first correction on the deflection of each section through the "bending moment-deflection ratio model" to generate a more realistic hull deformation curve, and its deflection is as Figure 3 shown by the blue line. The first correction result is further optimized through the ship deflection prediction network model, that is, neural network or physical constraint solution, to further improve the accuracy at the connection of the ship's bow and stern, as Figure 3 shown by the green line in the figure. As Figure 4 shown, the comparison of the ship bottom (draft) lines after uncorrected (black solid line), first correction (red dashed line), and second correction (blue dash-dotted line) is plotted respectively. Compared with the uncorrected curve, the first correction significantly corrects the deviation in the middle section and the bow and stern by performing an equal ratio scaling on the measured deflection and bending moment at the longitudinal middle position of the hull; the second correction further combines additional observations or physical constraints on this basis, making the ship bottom line smoother at the interruption and the connection of the bow and stern. In summary, the differences among the three curves show that the method proposed in this embodiment can effectively capture the draft error caused by load or structural deformation and lay a foundation for more accurate displacement calculation.

[0171] The method proposed in this embodiment significantly improves the accuracy of ship displacement calculation by introducing a bending moment correction mechanism, namely, the bending moment-deflection ratio model. Practical application data shows that the displacement calculated by the traditional uncorrected method is 13,997.56 tons, while the results after the first correction and the second correction of this patent are 14,919.97 tons and 14,846.61 tons respectively, which are about 6.6% and 6.1% higher than the uncorrected value. This significant numerical difference fully proves that ignoring the hogging / sagging effect of the hull will lead to a systematic underestimation of the displacement, and the correction method proposed in this patent can effectively compensate for the measurement error caused by the bending deformation of the hull. Further comparative analysis shows that the corrected displacement calculation result is closer to the actual cargo volume, verifying the practical value of the method proposed in this embodiment in the field of accurate ship measurement.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some 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 invention.

Claims

1. A method for calculating the displacement of dry bulk carriers based on the moment-deflection ratio model, characterized in that Specifically, it includes the following steps: S1: Obtain multi-source ship data for calculating the displacement of dry bulk carriers; And the multi-source ship data includes historical ship measurement data, ship attribute data, and longitudinal bending moment distribution data read based on a ship's loadometer under different working conditions; The ship measurement data includes at least the deflection data at the longitudinal mid-position of the hull; The ship attribute data includes at least the ship's reference design baseline and the draft survey curve; S2: Based on the elastic beam theory, obtain the ship's longitudinal bending moment differential equation for the ship beam deflection function; S3: According to the ship's longitudinal bending moment differential equation combined with historical ship measurement data, construct a bending moment-deflection ratio model for correcting the ship's deflection, and obtain the ship's cross-section deflection distribution data based on the bending moment-deflection ratio model; S4: Use the longitudinal bending moment distribution data and the corresponding ship cross-section deflection distribution data as data samples to obtain a data sample set; And randomly divide the data sample set and obtain a training set and a test set; S5: Construct a network comprehensive loss function with ship physical constraints and deflection offset supervision loss; And based on the multi-layer perceptron neural network, correct and train the multi-layer perceptron neural network according to the training set to obtain the corrected and trained multi-layer perceptron neural network; Based on the network comprehensive loss function, evaluate whether the output of the corrected and trained multi-layer perceptron neural network converges according to the test set combined with the ship attribute data, and use the corrected and trained multi-layer perceptron neural network when the output converges as the ship deflection prediction network model; S6: Obtain the corrected ship deflection data according to the ship deflection prediction network model, and obtain the corrected draft line according to the corrected ship deflection data to realize the calculation of the displacement of dry bulk carriers.

2. The method for calculating the displacement of a dry bulk carrier based on the moment-deflection ratio model according to claim 1, wherein Specifically, S2 includes the following steps: S21: Based on the elastic beam theory, simplify the longitudinal part of the hull into an elastic beam with a span of L; S22: Assume that the influence of the shear deformation and nonlinear effect of the elastic beam is ignored, and then construct the ship's longitudinal bending moment differential equation for the ship beam deflection function; And the expression of the ship's longitudinal bending moment differential equation is In the formula: E represents the elastic modulus; I represents the moment of inertia of the cross-section; x represents the position of any cross-section on the elastic beam; M(x) represents the longitudinal bending moment of the ship at the cross-section position; w(x) represents the deflection function of the ship beam, and the boundary conditions at both ends of the elastic beam are w(0) = 0, w(L) = 0.

3. A method for calculating the displacement of dry bulk carriers based on the moment-deflection ratio model according to claim 2, characterized in that, Specifically, S3 includes the following steps: S31: Assume a reference bending moment distribution of a ship and there exists a scaling factor α for any bending moment distribution curve such that a scaled bending moment distribution can be obtained; And the expression of the scaled bending moment distribution M α (x) is S32: According to the longitudinal bending moment differential equation of the ship, the reference bending moment distribution of the ship can be obtained deflection function satisfying the reference bending moment differential equation of the ship; And the ship's reference bending moment differential equation is S33: Based on the ship's reference bending moment differential equation and combined with S31, the scaled bending moment distribution can be obtained. The deflection function w α (x) of the ship's scaled bending moment differential equation that is satisfied; And the expression of the ship's scaled bending moment differential equation is S34: Multiply the in the ship's reference bending moment differential equation by the scaling factor α, and denote it as Then we can obtain Moreover, according to the uniqueness principle of common linear boundary value problems, it can be known that: Furthermore, a bending moment-deflection ratio model for correcting the ship's deflection can be obtained, and the expression of the bending moment-deflection ratio model is where: x ref represents the position of any reference section of the ship, and x ref ∈(0, L); w α (x ref ) represents the deflected moment distribution M α (x ref ) of the deflection function; represents the reference deflection obtained from the draft position of the ship; represents the reference bending moment value obtained based on the reference deflection ; S35: Obtain the ship's cross-section deflection distribution data according to the bending moment-deflection ratio model, and the acquisition formula for the ship's cross-section deflection distribution data is Where: ε(x i ) represents the deflection at the position x i of any longitudinal section of the ship; M(x i ) represents the bending moment value at the position x i of any longitudinal section of the ship; ε mid represents the deflection data at the longitudinal midship of the ship's hull measured; M mid represents the bending moment value at the longitudinal midship of the ship's hull read based on the ship's loading instrument; d FP represents the draft on the port side of the bow; d FS represents the draft on the starboard side of the bow; d AP represents the draft on the port side of the stern; d AS represents the draft on the starboard side of the stern; d MP represents the draft on the port side of the stern; d MS represents the draft on the starboard side of the stern.

4. A method for calculating the displacement of dry bulk carriers based on the moment-deflection geometric similarity model according to claim 3, characterized in that Specifically, S4 includes the following steps: S41: Define the ship section deflection distribution data obtained according to the bending moment-deflection equivalent ratio model as the deflection estimation offset And estimate the offset according to the deflection Obtain the preliminary correction curve of the deflection of the ship's cross-section; and the data point sequence on the preliminary correction curve is S42: Smoothly distribute the preliminary correction curve to obtain the beam deflection correction function for ship deflection correction And use the beam deflection correction function curve as the geometric distribution data of ship deflection; And the smoothing distribution process for the preliminary correction curve includes: Perform spline interpolation on the preliminary correction curve based on the multi-spline interpolation method to obtain an interpolation curve; Use the method of polynomial fitting to fit according to the interpolation curve and obtain the beam deflection correction function; S43: Use the ship's longitudinal bending moment distribution data as feature data, and use the corresponding ratio distribution data of the ship's deflection as label data to obtain a data sample set; Randomly divide the data sample set and obtain a training set and a test set.

5. A method for calculating the displacement of a dry bulk carrier based on a bending moment-deflection ratio model according to claim 4, characterized in that, The network comprehensive loss function with ship physical constraints and deflection offset supervision loss constructed in S5 has the following expression: In the formula: represents the network comprehensive loss function; represents the supervision loss of the measured deflection offset of the longitudinal position of the hull; ε obs,i represents the measured deflection at the longitudinal position x of the hull obs,i ; w θ (x obs,i ) represents the predicted ship deflection at the longitudinal position x of the hull output by the multi-layer perceptron neural network obs,i ; represents the physical residual loss of the deflection at the longitudinal position of the hull; x r represents the residual sampling point of the deflection at the longitudinal position of the hull and r(x r ) represents the physical residual term M(x r ) represents the longitudinal bending moment of the ship at the residual sampling point; and w θ″ (x r ) represents the simplified form obtained by automatic differentiation ; λ phys represents the learnable loss balance weight parameter 6. A method for calculating the displacement of a dry bulk carrier based on the bending moment-deflection ratio model according to claim 5, characterized in that The method for obtaining the ship deflection prediction network model in S5 specifically includes: S51: Use the ship longitudinal bending moment distribution data as the feature input data, and use the corresponding ship cross-section deflection distribution data as the label output data; S52: Randomly initialize the network parameter weights of the multi-layer perceptron neural network, and combine with the Adam optimization algorithm to correct and train the multi-layer perceptron neural network according to the training set to obtain the corrected and trained multi-layer perceptron neural network; S53: Based on the network comprehensive loss function, evaluate whether the output of the corrected and trained multi-layer perceptron neural network converges according to the test set combined with the ship attribute data; And the method for evaluating whether the output of the corrected and trained multi-layer perceptron neural network converges specifically includes: S001: Obtain the ship cross-section deflection distribution data output by the corrected and trained multi-layer perceptron neural network at this time, and based on the ship cross-section deflection distribution data combined with the ship baseline design bottom line, obtain the corrected draft line; And the formula for obtaining the corrected draft line is Where: z actual (x) corrected draft line; z design (x) represents the baseline design bottom line; represents the ship cross-section deflection distribution data obtained by evaluating the output of the multi-layer perceptron neural network after correction training; S002: Based on the displacement acquisition strategy, obtain the current displacement according to the corrected draft line; And the displacement acquisition strategy includes any one of the draft curve method, the sectional integration method or the molded line data integration method; Draft curve method: Perform linear interpolation on the preset draft-displacement query table according to the corrected draft line to obtain the current displacement; Sectional integration method: Based on the draft measurement curve, query and confirm the underwater area of the ship longitudinal section according to the corrected draft line, and obtain the current displacement according to the underwater area of the ship longitudinal section; And the expression for obtaining the current displacement according to the sectional integration method is Where: Δ represents the current displacement; A(T real (x)) represents the underwater area of the longitudinal section of the ship; If so, use the corrected and trained multi-layer perceptron neural network when the output converges as the ship deflection prediction network model; Otherwise, adaptively adjust the network weight parameters in the corrected and trained multi-layer perceptron neural network based on the backpropagation method, and repeat S52.

Images