Ship control motion intelligent forecasting method based on hybrid fuzzy neural network

Abstract

The invention discloses a hybrid fuzzy neural network-based ship control motion intelligent forecasting method, which relates to the technical field of ships, and comprises the following steps of: constructing a hydrodynamic correction item data set by combining test data of a still water ship control motion test with a still water ship control motion forecasting model; in combination with the characteristic that the input of a hydrodynamic correction term data set contains both uncertain variables and deterministic variables, an improved hybrid fuzzy neural network is constructed to establish a mapping relationship between control motion parameters and hydrodynamic error correction terms in a ship control motion prediction model; according to the method, a mathematical model method, a ship test method and an artificial intelligence technology are combined to comprehensively carry out intelligent forecasting on the ship control motion, and the method has the advantages of being high in precision, high in efficiency, high in generalization ability and the like. And a foundation is laid for on-line forecasting of ship control motion in an actual marine environment.

Description

Technical Field

[0001] This application relates to the technical field of ships, and in particular to an intelligent prediction method for ship maneuvering motion based on a hybrid fuzzy neural network. Background Art

[0002] The ship maneuvering motion is the motion response of a ship under the control of a driver or an autopilot. Rapid and accurate prediction of ship maneuvering motion is of great significance to ship navigation safety and effective control, and has always been a research hotspot.

[0003] Traditional ship maneuvering motion prediction methods include model test methods, Computational Fluid Dynamics (CFD) methods, and mathematical model methods. These three mainstream methods have their own advantages and disadvantages: (1) The model test method is considered the most reliable method, but it has various limitations in test facilities and test costs. (2) The CFD method considers fluid viscosity, can capture the details of the flow field and improve the prediction accuracy. However, it has the problem of low computational efficiency. (3) In the mathematical model method, Taylor series expansion is used to approximate the ship's hydrodynamic force, with a large number of simplifications and insufficient consideration of fluid viscosity. When predicting macroscopic indicators such as turning radius and overshoot angle that characterize ship maneuvering characteristics, it generally can meet the engineering requirements. However, when predicting the maneuvering motion process, the prediction accuracy is relatively poor. To sum up, traditional ship maneuvering motion prediction methods have problems such as high cost, slow efficiency, or poor accuracy. Therefore, these methods are currently mainly applicable to the evaluation of performance indicators in the ship design stage and still cannot meet the requirements of real-time, accurate, and online prediction of motion for safety assistance decision-making and effective control during actual ship navigation. Summary of the Invention

[0004] In view of the above problems and technical requirements, this application proposes an intelligent prediction method for ship maneuvering motion based on a hybrid fuzzy neural network. The technical solution of this application is as follows:

[0005] An intelligent prediction method for ship maneuvering motion based on a hybrid fuzzy neural network, the intelligent prediction method for ship maneuvering motion includes:

[0006] Construct a ship maneuvering motion prediction model in still water including a hydrodynamic force error correction term. The hydrodynamic force error correction term is related to n ship maneuvering motion parameters. Among the n ship maneuvering motion parameters, there are n1 motion parameters belonging to uncertain variables and n2 manipulation parameters belonging to deterministic variables, n = n1 + n2, integer parameter n1 ≥ 1, n2 ≥ 1;

[0007] Obtain the test data of the n ship maneuvering motion parameters in each ship maneuvering motion test in still water;

[0008] Substitute the test data of n maneuvering motion parameters in each ship maneuvering motion test in still water into the ship maneuvering motion prediction model in still water, and calculate the values of the corresponding hydrodynamic error correction terms to form a hydrodynamic correction term data set;

[0009] Take the test data of n maneuvering motion parameters in each ship maneuvering motion test in still water as the input of the hybrid fuzzy neural network, and the calculated values of the hydrodynamic error correction terms as the output of the hybrid fuzzy neural network, and use the hydrodynamic correction term data set to train the model of the hybrid fuzzy neural network; among them, the constructed hybrid fuzzy neural network uses a membership function to map each input uncertain variable, and uses a linear function to map each input deterministic variable;

[0010] Use the trained hybrid fuzzy neural network combined with the ship maneuvering motion prediction model in still water to realize the intelligent prediction of ship maneuvering motion in still water.

[0011] A further technical solution thereof is that the constructed hybrid fuzzy neural network includes an input layer, a mapping layer, a rule generation layer, a normalization layer and an output layer;

[0012] The input layer includes n input nodes. For any integer parameter 1 ≤ i ≤ n, when 1 ≤ i ≤ n1, the i-th input node belongs to the first type of input nodes and is used to obtain a corresponding motion parameter; when n1 + 1 ≤ i ≤ n, the i-th input node belongs to the second type of input nodes and is used to obtain a corresponding maneuvering parameter;

[0013] The mapping layer includes n mapping groups, and the i-th mapping group includes M i mapping nodes, and the outputs of the i-th input node are respectively connected to the inputs of the M i mapping nodes in the i-th mapping group; for any integer parameter 1 ≤ i ≤ n, when 1 ≤ i ≤ n1, the m-th mapping node in the i-th mapping group maps the input x i of the i-th input node based on the membership function and outputs the corresponding membership degree μ im , 1 ≤ m ≤ M i ; when n1 + 1 ≤ i ≤ n, M i =1 and the mapping node in the i-th mapping group linearly maps the input x i of the i-th input node and outputs the corresponding membership degree μ i1 =x i ;

[0014] The rule generation layer includes A rule generation node, any one of the rule generation nodes takes as input a combination formed by the membership degree values output by randomly selecting one mapping node from each of the n mapping groups, and outputs an activation intensity; the combinations of the membership degree values input to different rule generation nodes are not repeated;

[0015] The normalization layer includes a normalization node, the input of each normalization node is correspondingly connected to the output of a rule generation node, and each normalization node normalizes the activation intensity output by the corresponding rule generation node and then outputs;

[0016] The output layer includes K output nodes, and the number of K is equal to the number of hydrodynamic error correction terms in the static water ship maneuvering motion prediction model; the input of each output node is respectively connected to each normalization node, and each output node weights the outputs of all normalization nodes according to the connection weights and then outputs the result of a hydrodynamic error correction term.

[0017] A further technical solution thereof is that for the m-th mapping group in any i-th mapping group when 1≤i≤n1, it maps the input x of the i-th input node based on a Gaussian function and outputs the corresponding membership degree μ i as: im For:

[0018]

[0019] where, c im is the center point of the Gaussian function used by the m-th mapping group in the i-th mapping group when 1≤i≤n1, σ im is the width vector of the Gaussian function used by the m-th mapping group in the i-th mapping group when 1≤i≤n1, and e represents the natural exponent.

[0020] A further technical solution thereof is that the motion parameters among the n maneuvering motion parameters include the longitudinal speed u, the lateral speed v, and the yaw angular velocity r of the ship, and the maneuvering parameters among the n maneuvering motion parameters include the rudder angle δ of the ship.

[0021] A further technical solution thereof is that the static water ship maneuvering motion prediction model constructed including the hydrodynamic error correction term is:

[0022]

[0023] where, m is the mass of the ship, I zz represents the moment of inertia of the ship about the oz axis of the body-fixed coordinate system; u is the longitudinal speed of the ship, v is the lateral speed of the ship, and r is the yaw angular velocity of the ship; is the longitudinal acceleration of the ship, is the lateral acceleration of the ship, is the yaw angular acceleration of the ship; X1 is the longitudinal inertial hydrodynamic force acting on the ship, Y1 is the lateral inertial hydrodynamic force acting on the ship, and N1 is the yaw inertial moment acting on the ship; X2 is the longitudinal viscous hydrodynamic force acting on the ship, Y2 is the lateral viscous hydrodynamic force acting on the ship, and N2 is the yaw viscous moment acting on the ship; ΔX, ΔY, and ΔN are all hydrodynamic error correction terms, representing the longitudinal hydrodynamic error correction term, the lateral hydrodynamic error correction term, and the yaw hydrodynamic error correction term respectively.

[0024] Its further technical solution is that obtaining the test data of n maneuvering motion parameters in each calm water ship maneuvering motion test includes: obtaining the test data of n maneuvering motion parameters of the actual ship in the calm water ship maneuvering motion test; or, conducting a self-propelled model test of the calm water ship maneuvering motion and obtaining the test data of n maneuvering motion parameters of the ship model in the calm water ship maneuvering motion test; where the calm water ship maneuvering motion test includes course-keeping maneuvering motion, turning maneuvering motion, and Z-shaped maneuvering motion.

[0025] Its further technical solution is that obtaining the test data of n maneuvering motion parameters in each calm water ship maneuvering motion test includes:

[0026] Determine the test data of n2 maneuvering parameters applied in each calm water ship maneuvering motion test, and collect the original measurement data of n1 motion parameters. Filter the high-frequency noise data in the original measurement data of each motion parameter according to the ship scale to obtain the test data of the motion parameter.

[0027] Its further technical solution is that filtering the high-frequency noise data in the original measurement data of each motion parameter according to the ship scale to obtain the test data of the motion parameter includes:

[0028] Use the Fourier transform method to convert the original measurement data sig(t) in the time-domain signal form of each motion parameter into a frequency-domain signal ω represents frequency, t represents time, represents the imaginary number, and e represents the natural exponent;

[0029] Use the critical frequency matched with the ship scale for the frequency-domain signal Do the inverse Fourier transform to obtain the test data after filtering out the high-frequency noise data

[0030] Its further technical solution is that using the trained hybrid fuzzy neural network combined with the calm water ship maneuvering motion prediction model to realize the intelligent prediction of the calm water ship maneuvering motion includes:

[0031] Substitute the trained hybrid fuzzy neural network into the ship maneuvering motion prediction model in still water, and solve it using the fourth-order Runge-Kutta method according to the prediction time step to obtain the intelligent prediction result of the ship maneuvering motion in still water.

[0032] Its further technical solution is that the expression of the inertial hydrodynamic force acting on the ship is:

[0033]

[0034] where, λ 11 、λ 22 、λ 26 、λ 66 are the added mass coefficients of the ship;

[0035] Select the initial constant-speed straight-ahead motion state of the ship as the Taylor series expansion point, and expand the longitudinal viscous hydrodynamic force acting on the ship by Taylor series as:

[0036]

[0037] where, X0, Y0, Z0 represent the longitudinal hydrodynamic force, lateral hydrodynamic force and yaw moment acting on the ship in the initial constant-speed straight-ahead motion state; Δu = u - u0, u0 represents the initial straight-ahead speed of the ship; the first-order hydrodynamic derivative the second-order hydrodynamic derivative the third-order hydrodynamic derivative where, ξ = u, v, r, δ, ε = u, v, rδ.

[0038] The beneficial technical effects of this application are:

[0039] This application discloses an intelligent prediction method for ship maneuvering motion based on a hybrid fuzzy neural network. This method constructs a ship maneuvering motion prediction model in still water that includes a hydrodynamic error correction term, and then uses the experimental data of the ship maneuvering motion in still water to construct a hydrodynamic correction term data set. Combining the characteristics that the input of the hydrodynamic correction term data set includes both uncertain variables and deterministic variables, an improved hybrid fuzzy neural network is constructed to map these two different variables respectively to establish the mapping relationship between the maneuvering motion parameters and the hydrodynamic error correction term. Combining with the ship maneuvering motion prediction model in still water, the real-time and accurate online prediction of the ship maneuvering motion in still water can be realized. This method combines the mathematical model method, the ship test method and the artificial intelligence technology to comprehensively conduct the intelligent prediction of the ship maneuvering motion, and has the advantages of high accuracy, fast efficiency and strong generalization ability, laying a foundation for the online prediction of the ship maneuvering motion in the actual marine environment.

[0040] This method constructs a ship maneuvering motion prediction model in still water based on the traditional maneuvering motion equations in still water, ensuring the generalization ability of the model. The addition of the hydrodynamic error correction term can correct the hydrodynamic expression errors introduced by simplification and approximation, thereby improving the prediction accuracy.

[0041] The improved hybrid fuzzy neural network used in this method is obtained by improving on the network structure of the traditional fuzzy neural network. It can make up for the defect that the traditional fuzzy neural network can only input uncertain variables, enabling the improved hybrid fuzzy neural network to be applicable to the application scenario of this application and to consider the uncertain characteristics of ship maneuvering motion data, thereby further improving the prediction accuracy.

[0042] There are scale effect problems when the traditional model test method, computational fluid dynamics method, and mathematical model method extrapolate the model-scale prediction results to the full scale. The test data used in this application to construct the hydrodynamic correction term dataset can not only adopt the ship model test data collected during the self-propelled model test of ship maneuvering motion, but also adopt the full-scale ship test data during the full-scale ship maneuvering motion. Although it is more difficult to obtain full-scale ship test data than ship model test data, when using the full-scale ship test data to construct the hydrodynamic correction term dataset to train the hybrid fuzzy neural network, the scale effect between the prediction results of small-scale ship models and the motion of full-scale ships can be eliminated, making the method of this application have the advantage of strong scalability to the full scale. Brief Description of the Drawings

[0043] Figure 1 is a flowchart of the intelligent prediction method for ship maneuvering motion according to an embodiment of this application.

[0044] Figure 2 is a schematic diagram of the earth coordinate system and the ship-fixed coordinate system describing ship maneuvering motion.

[0045] Figure 3 is a network structure diagram of the hybrid fuzzy neural network constructed in an embodiment. Detailed Embodiments

[0046] The following further describes the detailed embodiments of this application with reference to the drawings.

[0047] This application discloses an intelligent prediction method for ship maneuvering motion based on a hybrid fuzzy neural network, including the following content. Please refer to Figure 1 the flowchart shown:

[0048] Step 1, construct a ship maneuvering motion prediction model in still water including a hydrodynamic error correction term.

[0049] For convenience of description, first define the earth coordinate system O-x0y0z0 and the ship-fixed coordinate system o-xyz describing ship maneuvering motion asFigure 2 As shown in the figure, both of these two coordinate systems satisfy the right - hand rule. The origin o of the ship - fixed coordinate system o - xyz is located at the center of gravity of the ship. The positive direction of the ox axis points along the ship's centerline towards the bow, and the positive direction of the oy axis points towards the starboard side. The angle ψ between the Ox0 axis of the earth - fixed coordinate system and the ox axis of the ship - fixed coordinate system is defined as the heading angle. It is stipulated that the positive direction of ψ is the direction from the Ox0 axis to the ox axis according to the right - hand rule around the Oz0 axis. The angle β between the direction of the ship speed U and the ox axis is defined as the drift angle. It is stipulated that the positive direction of β is the direction from the velocity vector direction to the ox axis according to the right - hand rule around the Oz0 axis. It is stipulated that the positive direction of the rudder angle δ is the rotation to the starboard side of the hull.

[0050] The ship maneuvering motion prediction model constructed in this application is obtained by adding a hydrodynamic error correction term to the existing maneuvering motion equation in still water. Using the existing maneuvering motion equation in still water to express the known mechanism part of the ship maneuvering motion can ensure the generalization ability of the constructed model. The expression of the existing maneuvering motion equation in still water is as follows:

[0051]

[0052] Among them, m is the mass of the ship, and I zz represents the moment of inertia of the ship about the oz axis of the ship - fixed coordinate system. u is the longitudinal velocity of the ship along the ox axis of the ship - fixed coordinate system, v is the transverse velocity of the ship along the oy axis of the ship - fixed coordinate system, and is the yaw angular velocity of the ship about the oz axis of the ship - fixed coordinate system. is the longitudinal acceleration of the ship along the ox axis of the ship - fixed coordinate system, is the transverse acceleration of the ship along the oy axis of the ship - fixed coordinate system, is the yaw angular acceleration of the ship about the oz axis of the ship - fixed coordinate system.

[0053] X1 is the longitudinal inertial hydrodynamic force acting on the ship, Y1 is the transverse inertial hydrodynamic force acting on the ship, and N1 is the yaw inertial moment acting on the ship. The expressions are:

[0054]

[0055] Among them, λ 11 、λ 22 、λ 26 、λ 66 are the added mass coefficients of the ship, which can be obtained through constrained model tests, empirical formulas, or computational fluid dynamics methods.

[0056] X2 is the longitudinal viscous hydrodynamic force acting on the ship, Y2 is the transverse viscous hydrodynamic force acting on the ship, and N2 is the yaw viscous moment acting on the ship. Considering the symmetry of the ship shape and choosing the initial straight - line motion state of the ship at a constant speed as the Taylor series expansion point, the Taylor series expansion is expressed as follows:

[0057]

[0058] Among them, X0, Y0, and Z0 represent the longitudinal hydrodynamic force, lateral hydrodynamic force, and yaw moment acting on the ship in the initial steady straight-ahead motion state. Δu = u - u0, where u0 represents the initial straight-ahead speed of the ship. The first-order hydrodynamic derivatives The second-order hydrodynamic derivatives The third-order hydrodynamic derivatives Among them, ξ = u, v, r, δ, ε = u, v, r, δ. The hydrodynamic derivatives of each order can be obtained through constrained model tests, empirical formulas, or computational fluid dynamics methods.

[0059] Since there are errors in the traditional maneuvering motion equations in still water shown in equations (1) - (3) where the hydrodynamic forces of the ship are approximately expanded as a Taylor series, therefore, on this basis, a hydrodynamic force error correction term is added to the maneuvering motion equation in still water in this application to correct the hydrodynamic force expression errors introduced by simplification and approximation, etc., in order to improve the prediction accuracy of ship maneuvering motion. Thus, the constructed prediction model of ship maneuvering motion in still water containing the hydrodynamic force error correction term is as follows:

[0060]

[0061] Compared with equation (1), equation (4) adds three hydrodynamic force error correction terms ΔX, ΔY, and ΔN. ΔX, ΔY, and ΔN represent the longitudinal hydrodynamic force error correction term, lateral hydrodynamic force error correction term, and yaw hydrodynamic force error correction term respectively.

[0062] Step 2: Obtain the experimental data of n maneuvering motion parameters in each ship maneuvering motion test in still water.

[0063] These n maneuvering motion parameters are the maneuvering motion parameters related to the hydrodynamic force error correction terms in equation (4) during the ship maneuvering motion process, and these n maneuvering motion parameters include n1 motion parameters and n2 maneuvering parameters, n = n1 + n2, where the integer parameter n1 ≥ 1 and n2 ≥ 1. The ship maneuvering motion tests in still water carried out include course-keeping maneuvering motion, turning maneuvering motion, and Z-shaped maneuvering motion, and record the measured values of the n1 motion parameters and n2 maneuvering parameters of the ship in each ship maneuvering motion test in still water.

[0064] In one embodiment, if the ship maneuvering motion test in still water in this step is a full-scale ship test, this step is actually to obtain the test data of n maneuvering motion parameters of the full-scale ship in the ship maneuvering motion test in still water. However, considering the great difficulty of conducting the ship maneuvering motion test in still water using a full-scale ship, in another embodiment, a self-propelled model test of the ship maneuvering motion in still water is carried out, and the test data of n maneuvering motion parameters of the ship model in the ship maneuvering motion test in still water is obtained.

[0065] Whether a full-scale ship test or a self-propelled model test is adopted, when obtaining the test data of n maneuvering motion parameters in each ship maneuvering motion test in still water, since the values of the maneuvering parameters are determined, the values of the n2 maneuvering parameters applied to the full-scale ship / ship model in each ship maneuvering motion test in still water are directly determined as the test data. The original measurement data of the n1 motion parameters of the full-scale ship / ship model directly collected is obtained by internal differentiation of the test system, and the colored noise generated by the differentiation will interfere with the establishment of the subsequent hydrodynamic correction intelligent model. Therefore, the high-frequency noise data in the original measurement data of each motion parameter is filtered according to the ship scale first to obtain the test data of this motion parameter. In one embodiment, the Fourier transform method is used to filter the high-frequency noise data, including:

[0066] First, the Fourier transform method is used to convert the original measurement data sig(t) in the time-domain signal form of each motion parameter collected into a frequency-domain signal as:

[0067]

[0068] where w represents frequency, t represents time, j represents an imaginary number, and e represents the natural exponent.

[0069] Then, the critical frequency matched with the ship scale is used to perform the inverse Fourier transform on the frequency-domain signal to obtain the test data sig′(t) after filtering out the high-frequency noise data as:

[0070]

[0071] When a full-scale ship test is adopted, the critical frequency matched with the full-scale ship scale When a ship model test is adopted, since the ship model is generally less than 10 meters, the critical frequency matched with the ship model scale is taken

[0072] According to the actual construction situation, among these n maneuvering motion parameters, there are n1 = 3 motion parameters and n2 = 1 maneuvering parameter. These three motion parameters are respectively the longitudinal speed u of the ship, the lateral speed v, and the yaw angular velocity r, and the maneuvering parameter is the rudder angle δ. Then the test data obtained in each ship maneuvering motion test in still water includes the test data of the longitudinal speed u, the lateral speed v, the yaw angular velocity r, and the rudder angle δ of the actual ship / model at different times t during the test. In fact, the heading angle ψ, heave speed w, roll angular velocity p, and pitch angular velocity q at different times t in each ship maneuvering motion test in still water are also obtained.

[0073] Step 3: Substitute the test data of the n maneuvering motion parameters in each ship maneuvering motion test in still water into the still water ship maneuvering motion prediction model of Equation (4), and calculate the values of the corresponding hydrodynamic error correction terms to form a hydrodynamic correction term data set.

[0074] Based on the premise that the n maneuvering motion parameters include the longitudinal speed u, the lateral speed v, the yaw angular velocity r, and the rudder angle, the three hydrodynamic error correction terms related to these n maneuvering motion parameters can actually be expressed as ΔX(δ, u, v, r), ΔY(δ, u, v, r), and ΔN(δ, u, v, r). Then, substitute the test data of the longitudinal speed u, the lateral speed v, the yaw angular velocity r, and the rudder angle δ in each ship maneuvering motion test in still water into the still water ship maneuvering motion prediction model of Equation (4), and the hydrodynamic error correction terms ΔX(δ, u, v, r), ΔY(δ, u, v, r), and ΔN(δ, u, v, r) related to the longitudinal speed u, the lateral speed v, the yaw angular velocity r, and the rudder angle δ can be calculated according to the following formula:

[0075]

[0076] Then, a hydrodynamic correction term data set can be constructed. The hydrodynamic correction term data set includes the test parameters of the longitudinal speed u, the lateral speed v, the yaw angular velocity r, and the rudder angle δ in each ship maneuvering motion test in still water, as well as the calculated longitudinal hydrodynamic error correction term ΔX(δ, u, v, r), lateral hydrodynamic error correction term ΔY(δ, u, v, r), and yaw hydrodynamic error correction term ΔN(δ, u, v, r).

[0077] Step 4: Use the test data of the n maneuvering motion parameters in each ship maneuvering motion test in still water as the input of the hybrid fuzzy neural network, and the calculated values of the hydrodynamic error correction terms as the output of the hybrid fuzzy neural network, and use the hydrodynamic correction term data set to train the model of the hybrid fuzzy neural network.

[0078] Then, the relationship between the maneuvering motion parameters and the hydrodynamic error correction terms is established based on a large amount of data using artificial intelligence technology. Common artificial intelligence technologies such as the Extended Kalman Filter (EKF), maximum likelihood function method, least squares method, neural network method, and support vector machine cannot consider the uncertain characteristics of ship maneuvering motion data, so they cannot accurately predict. The fuzzy neural network is a hybrid technology that combines fuzzy logic and neural networks and can consider data uncertainty to establish a mapping relationship between multiple input variables and multiple output variables. Therefore, this application is based on a fuzzy neural network to predict the relationship between the maneuvering motion parameters and the hydrodynamic error correction terms. The fuzzy neural network is good at processing data with fuzziness or uncertainty in the input or output layer and shows strong capabilities in dealing with complex uncertain real-world problems. However, the limitation of the fuzzy neural network is that it can only input uncertain variables.

[0079] In this application, considering that the motion parameters of the ship are uncertain and thus belong to uncertain variables, while the maneuvering parameters of the ship are definite values and thus belong to definite variables. Therefore, the n maneuvering motion parameters as inputs include both uncertain variables and definite variables. Specifically, n1 of the motion parameters belong to uncertain variables, and the remaining n2 maneuvering parameters belong to definite variables.

[0080] Since the n maneuvering motion parameters as inputs also include definite variables, the conventional fuzzy neural network cannot handle them. Therefore, the improved hybrid fuzzy neural network applied in this application mainly modifies the second layer compared with the traditional fuzzy neural network to map the input uncertain variables and definite variables in different ways, so that the hybrid fuzzy neural network can handle both uncertain variables and definite variables, thus adapting to the application scenario of this application.

[0081] Specifically, the constructed hybrid fuzzy neural network uses the membership function to map each input uncertain variable in the traditional method and adds a linear function to map each input definite variable. Under this design concept, the network structure of the constructed hybrid fuzzy neural network is as Figure 3 shown. Similar to the classical fuzzy neural network, it needs to go through five network structures, including an input layer, a mapping layer, a rule generation layer, a normalization layer, and an output layer in sequence from input to output:

[0082] (1) The input layer includes n input nodes, and the n input nodes respectively obtain the n maneuvering motion parameters.

[0083] For any integer parameter \(1\leq i\leq n\), when \(1\leq i\leq n_1\), the \(i\)-th input node belongs to the first type of input nodes and is used to obtain a corresponding motion parameter. When \(n_1 + 1\leq i\leq n\), the \(i\)-th input node belongs to the second type of input nodes and is used to obtain a corresponding manipulation parameter.

[0084] Such as Figure 3 Taking the manipulation motion parameters including longitudinal speed \(u\), lateral speed \(v\), yaw angular velocity \(r\), and rudder angle as an example, the input layer includes 4 input nodes. The 1st to 3rd input nodes respectively obtain the longitudinal speed \(u\), lateral speed \(v\), and yaw angular velocity \(r\), and the 4th input node obtains the rudder angle \(\delta\).

[0085] (2) The mapping layer includes \(n\) mapping groups. The \(i\)-th mapping group includes \(M\) i mapping nodes. The outputs of the \(i\)-th input node are respectively connected to the inputs of the \(M\) i mapping nodes in the \(i\)-th mapping group. For any integer parameter \(1\leq i\leq n\), mapping is performed in two cases:

[0086] When \(1\leq i\leq n_1\), the mapping nodes \(M\) i contained in the \(i\)-th mapping group usually have multiple. The value of the mapping node \(M\) i is the number of fuzzy classification levels of the motion parameter obtained by the \(i\)-th input node, which can be set according to specific problems and relevant experience. And the number of mapping nodes contained in any two of the 1st to \(n_1\) mapping groups may be equal or not equal. In this case, the \(m\)-th mapping node in the \(i\)-th mapping group maps the input \(x\) i of the \(i\)-th input node based on the membership function and outputs the corresponding membership degree \(\mu\) im , \(1\leq m\leq M\) i . In one embodiment, when \(1\leq i\leq n_1\), the \(m\)-th mapping group in the \(i\)-th mapping group maps the input \(x\) i of the \(i\)-th input node based on the Gaussian function and outputs the corresponding membership degree \(\mu\) im as:

[0087]

[0088] In formula (8), \(c\) im is the center point of the Gaussian function used by the \(m\)-th mapping group in the \(i\)-th mapping group when \(1\leq i\leq n_1\), \(\sigma\) im is the width vector of the Gaussian function used by the \(m\)-th mapping group in the \(i\)-th mapping group when \(1\leq i\leq n_1\), and \(e\) represents the natural exponent.

[0089] When any integer \(n_1 + 1\leq i\leq n\), the mapping node \(M\) i contained in the \(i\)-th mapping group is equal to 1. The only mapping node in the \(i\)-th mapping group maps the input \(x\) of the \(i\)-th input nodei Perform a linear mapping and output the corresponding membership degree μ i1 = x i .

[0090] (3) The rule generation layer includes rule generation nodes. Any one rule generation node takes as input a combination formed by the membership degree values output by randomly selecting one mapping node from each mapping group of the mapping groups, and outputs an activation strength. When randomly selecting one mapping node from each mapping group to output the combination of membership degree values, the total number of permutation and combination methods formed is types. These types of permutation and combination methods are respectively input into rule generation nodes, and the combinations of membership degree values input into different rule generation nodes do not repeat.

[0091] (4) The normalization layer includes normalization nodes. The input of each normalization node is correspondingly connected to the output of a rule generation node, and each normalization node performs normalization processing on the activation strength output by the corresponding rule generation node and then outputs.

[0092] (5) The output layer includes K output nodes. The input of each output node is respectively connected to each normalization node, and each output node performs weighted processing on the outputs of all normalization nodes according to the connection weights and then outputs the result of a hydrodynamic error correction term. The number of K is equal to the number of hydrodynamic error correction terms in the ship maneuvering motion prediction model in still water. Based on the form of Equation (4), K = 3.

[0093] Based on the constructed network structure of the hybrid fuzzy neural network, the hybrid fuzzy neural network can be trained using the hydrodynamic correction term data set. During the training process, the fuzzy neural network will continuously adjust the model parameters to make the output result closer to the expected result. The adjusted model parameters include the center point c im and width vector σ im of the Gaussian function used in the m-th mapping group in the i-th mapping group when 1 ≤ i ≤ n1, and the connection weights adopted by the output layer, which can be learned and determined using the backpropagation algorithm. When the hybrid fuzzy neural network reaches a certain accuracy, the training process stops.

[0094] During actual training, the hydrodynamic correction term data set is divided into a training set, a validation set, and a test set. First, the hybrid fuzzy neural network is trained based on the training set, and the weighted sum of squared errors of each hydrodynamic error correction term is used as the performance index, and the backpropagation algorithm is used to determine the model parameters. Then, the validation set data is applied to prevent overfitting during the training process. Finally, the accuracy of the hybrid fuzzy neural network is tested by the test set data.

[0095] Step 5: Using the trained hybrid fuzzy neural network in combination with the ship maneuvering motion prediction model in still water, intelligent prediction of ship maneuvering motion in still water is achieved.

[0096] After the training is completed and a hybrid fuzzy neural network that meets the accuracy requirements is obtained, the trained hybrid fuzzy neural network is substituted into the prediction model of ship maneuvering motion in still water, and the fourth-order Longo Kutta method is applied according to the prediction time step to obtain the intelligent prediction result of ship maneuvering motion in still water.

[0097] The above description is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and variations directly derived or imagined by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the scope of protection of the present application.

Claims

1. An intelligent prediction method for ship maneuvering motion based on a hybrid fuzzy neural network, characterized in that, The intelligent prediction method for ship maneuvering motion includes: Constructing a prediction model for ship maneuvering motion in still water that includes a hydrodynamic error correction term. The hydrodynamic error correction term is related to n ship maneuvering motion parameters. Among the n ship maneuvering motion parameters, there are n1 motion parameters belonging to uncertain variables and n2 maneuvering parameters belonging to deterministic variables, where n = n1 + n2, the integer parameter n1 ≥ 1, and n2 ≥ 1; Obtaining the test data of the n ship maneuvering motion parameters in each ship maneuvering motion test in still water; Substituting the test data of the n ship maneuvering motion parameters in each ship maneuvering motion test in still water into the prediction model for ship maneuvering motion in still water, and calculating the obtained values of the corresponding hydrodynamic error correction term to form a hydrodynamic correction term data set; Using the test data of the n ship maneuvering motion parameters in each ship maneuvering motion test in still water as the input of the hybrid fuzzy neural network, and the obtained values of the hydrodynamic error correction term as the output of the hybrid fuzzy neural network, and using the hydrodynamic correction term data set to train the model of the hybrid fuzzy neural network; among them, the constructed hybrid fuzzy neural network uses a membership function to map each input uncertain variable, and uses a linear function to map each input deterministic variable; Using the trained hybrid fuzzy neural network combined with the prediction model for ship maneuvering motion in still water to realize the intelligent prediction of ship maneuvering motion in still water.

2. The intelligent prediction method for ship maneuvering motion according to claim 1, wherein The constructed hybrid fuzzy neural network includes an input layer, a mapping layer, a rule generation layer, a normalization layer, and an output layer; The input layer includes n input nodes. For any integer parameter 1 ≤ i ≤ n, when 1 ≤ i ≤ n1, the i-th input node belongs to the first type of input nodes and is used to obtain a corresponding motion parameter; when n1 + 1 ≤ i ≤ n, the i-th input node belongs to the second type of input nodes and is used to obtain a corresponding maneuvering parameter; The mapping layer includes n mapping groups, and the i-th mapping group includes M i mapping nodes. The output of the i-th input node is respectively connected to the inputs of M i mapping nodes in the i-th mapping group; For any integer parameter \(1\leq i\leq n\), when \(1\leq i\leq n_1\), the \(m\)-th mapping node in the \(i\)-th mapping group maps the input \(x\) of the \(i\)-th input node based on the membership function and outputs the corresponding membership degree \(\mu\) i , where \(1\leq m\leq M\) im ; when \(n_1 + 1\leq i\leq n\), \(M\) i = 1 and the mapping node in the \(i\)-th mapping group linearly maps the input \(x\) of the \(i\)-th input node and outputs the corresponding membership degree \(\mu\) i = \(x\) i ; i1 i ​​ The rule generation layer includes rule generation nodes. Any one of the rule generation nodes takes as input a combination formed by membership degree values randomly taken from each mapping group of n mapping groups and outputs an activation strength; The combinations of membership values input to different rule generation nodes are not repeated; The normalization layer includes normalization nodes. The input of each normalization node is correspondingly connected to the output of a rule generation node, and each normalization node outputs after normalizing the activation intensity of the output of the corresponding rule generation node; The output layer includes K output nodes, and the number of K is equal to the number of hydrodynamic error correction terms in the prediction model for ship maneuvering motion in still water; the input of each output node is respectively connected to each normalization node, and each output node outputs the result of a hydrodynamic error correction term after weighted processing of the outputs of all normalization nodes according to the connection weights.

3. The intelligent prediction method for ship maneuvering motion according to claim 1, characterized in that When \(1\leq i\leq n_1\), for any \(m\)-th mapping group in the \(i\)-th mapping group, the input \(x\) of the \(i\)-th input node is mapped based on the Gaussian function and the corresponding membership degree \(\mu\) is output i as follows im : where c im is the center point of the Gaussian function used in the m-th mapping group of the i-th mapping group when 1 ≤ i ≤ n1, σ im is the width vector of the Gaussian function used in the m-th mapping group of the i-th mapping group when 1 ≤ i ≤ n1, and e represents the natural exponent.

4. The intelligent prediction method for ship maneuvering motion according to claim 1, wherein The motion parameters among the n ship maneuvering motion parameters include the longitudinal speed u, the lateral speed v, and the yaw angular velocity of the ship, and the maneuvering parameters among the n ship maneuvering motion parameters include the rudder angle δ of the ship.

5. The intelligent prediction method for ship maneuvering motion according to claim 1, characterized in that The constructed prediction model for ship maneuvering motion in still water that includes a hydrodynamic error correction term is: where m is the mass of the ship, and I zz represents the moment of inertia of the ship about the oz axis of the body-fixed coordinate system; u is the longitudinal velocity of the ship, v is the lateral velocity of the ship, and r is the yaw angular velocity of the ship; is the longitudinal acceleration of the ship, is the lateral acceleration of the ship, is the yaw angular acceleration of the ship; X1 is the longitudinal inertial hydrodynamic force acting on the ship, Y1 is the lateral inertial hydrodynamic force acting on the ship, and N1 is the yaw inertial moment acting on the ship; X2 is the longitudinal viscous hydrodynamic force acting on the ship, Y2 is the lateral viscous hydrodynamic force acting on the ship, and N2 is the yaw viscous moment acting on the ship; ΔX, ΔY, and ΔN are all hydrodynamic error correction terms and represent the longitudinal hydrodynamic error correction term, the lateral hydrodynamic error correction term, and the yaw hydrodynamic error correction term, respectively.

6. The intelligent prediction method for ship maneuvering motion according to claim 1, characterized in that, Obtaining the test data of the n ship maneuvering motion parameters in each ship maneuvering motion test in still water includes: Obtaining the test data of the n ship maneuvering motion parameters in the ship maneuvering motion test of the actual ship in still water; or, conducting a self-propelled model test of ship maneuvering motion in still water, and obtaining the test data of the n ship maneuvering motion parameters in the ship maneuvering motion test of the ship model in still water; Among them, the ship maneuvering motion test in still water includes a course-keeping maneuvering motion, a turning maneuvering motion, and a zigzag maneuvering motion.

7. The intelligent prediction method for ship maneuvering motion according to claim 1, characterized in that The test data for obtaining n maneuvering motion parameters in each ship maneuvering motion test in still water includes: Determine the test data of n2 maneuvering parameters applied in each ship maneuvering motion test in still water, collect the original measurement data of n1 motion parameters, and filter the high-frequency noise data in the original measurement data of each motion parameter according to the ship scale to obtain the test data of the motion parameters.

8. The intelligent prediction method for ship maneuvering motion according to claim 7, characterized in that, Filtering the high-frequency noise data in the original measurement data of each motion parameter according to the ship scale to obtain the test data of the motion parameters includes: The original measurement data sig(t) in the form of a time-domain signal of each motion parameter is converted into a frequency-domain signal by using the Fourier transform method ω represents frequency, t represents time, j represents an imaginary number, and e represents the natural exponent; Critical frequency matching the ship size For the frequency-domain signal Perform the inverse Fourier transform to obtain the test data after filtering out the high-frequency noise data 9. The intelligent prediction method for ship maneuvering motion according to claim 1, characterized in that Using the trained hybrid fuzzy neural network combined with the ship maneuvering motion prediction model in still water to realize the intelligent prediction of ship maneuvering motion in still water includes: Substitute the trained hybrid fuzzy neural network into the ship maneuvering motion prediction model in still water, and solve it using the fourth-order Runge-Kutta method according to the prediction time step to obtain the intelligent prediction result of ship maneuvering motion in still water.

10. The intelligent prediction method for ship maneuvering motion according to claim 6, wherein The expression of the inertial hydrodynamic force acting on the ship is: Among them, λ 11 , λ 22 , λ 26 , λ 66 are the added mass coefficients of the ship; Select the initial constant-speed straight-line motion state of the ship as the Taylor series expansion point, and expand the longitudinal viscous hydrodynamic force acting on the ship in terms of Taylor series as: Among them, X0, Y0, and Z0 represent the longitudinal hydrodynamic force, lateral hydrodynamic force, and yaw moment acting on the ship in the initial steady straight-line motion state; Δu = u - u0, where u0 represents the initial straight-line speed of the ship; the first-order hydrodynamic derivative Second-order hydrodynamic derivative Third-order hydrodynamic derivative Among them, ξ = u, v, r, δ, ε = u, v, rδ.

Images