Ship control motion nonparametric identification modeling method based on Gaussian mixture model

Abstract

The invention relates to a ship manipulation nonparametric identification modeling method based on a Gaussian mixture model, which comprises the following steps of: acquiring a rudder angle, a ship motion state and a ship motion force and moment of force when a ship sails, and constructing a data set; a linear regression model is determined according to the force and torque borne by the ship in the ship control motion mathematical model and the ship motion state; taking a rudder angle, a force and a moment as training samples, taking a linear regression model as a basis, segmenting training data according to change characteristics of the rudder angle to construct a plurality of sub-models, and obtaining an improved Gaussian mixture model for ship control motion; training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model; based on the motion state of the ship at the previous moment, the trained improved Gaussian mixture model is called in a regression mode, and prediction of fluid force of ship force at the next moment is achieved; based on the predicted fluid force of the ship force at the next moment, the time derivative of the ship is solved, and the motion state of the ship at the next moment is obtained.

Description

Technical Field

[0001] The present invention belongs to the field of ship technology, and relates to a non-parametric identification modeling method for ship maneuvering motion based on a Gaussian mixture model, and in particular to a non-parametric identification modeling method for ship maneuvering motion based on a Gaussian mixture regression of an expectation maximization algorithm. Background Art

[0002] The mathematical model of ship motion is the core foundation for studying ship maneuverability and control problems. Since the 1970s, with the emergence of supertankers and the development of advanced ship motion controllers, the progress of measurement technology and system identification theory has further promoted the research on the mathematical model of ship motion. From a methodological point of view, the system modeling method is mainly divided into mechanism modeling, identification modeling and hybrid modeling combining the two. Mechanism modeling has good stability and generalization ability, but it is limited by the current knowledge understanding. It is difficult to establish an accurate motion model for complex ship types and complex navigation environments. Therefore, a system identification method that does not rely on the mathematical model of the research object itself and has algorithm stability and accuracy has emerged. The system identification method only requires the state information and inertia of the ship, and the method has good transferability and generalization.

[0003] From the literature [1][2][3][4][5][6] It can be seen that various ship motion identification models have been proposed. Traditional identification methods mainly include the least squares method and the Kalman filter method. The least squares method is widely used, but there are problems such as overfitting and matrix inversion. For this reason, the recursive least squares method is used. [1] The extended Kalman filter is also widely used in the identification of ship models, but the classical Kalman filter algorithm has poor adaptability to nonlinear systems, which greatly limits its application. [5] and unscented Kalman filter [6] Algorithms such as have been proposed, which have good identification effects on nonlinear systems, but they also have problems of parameter drift and dynamic cancellation.

[0004] Traditional system identification methods rely on the mathematical mechanism model of the research object, and the convergence and accuracy of the algorithm are also poor. With the development of computers and artificial intelligence, identification methods are gradually developing towards intelligent algorithms. [2] A highly robust system identification method is proposed to alleviate the impact of parameter drift and overfitting on the reliability of model identification. [3][4]By adopting a local weighted learning algorithm and re-dimensioning the data structure, an offline black box identification modeling method is proposed, which effectively overcomes the problems of parameter drift, uncertainty, modeling dynamics, etc. in traditional mechanism modeling and parameter identification models. At present, most ship modeling methods rely on parameterized models, but non-parametric identification does not require a priori assumptions and can adapt to complex, nonlinear and time-varying systems more flexibly. With the development of intelligent ship technology, the requirements for model accuracy and real-time performance have increased. Non-parametric identification can automatically optimize the model through data-driven, and effectively respond to external disturbances and improve robustness. Its flexibility and adaptability make it have important application value in intelligent ship motion control, significantly improving the safety and accuracy of the system. Summary of the invention

[0005] In order to solve the problems of parameter drift, uncertainty, modeling dynamics and insufficient model accuracy in traditional system identification modeling methods including least squares method and Kalman filter algorithm and their variants, the technical solution adopted by the present invention is: a non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model, comprising the following steps:

[0006] S1: Obtain the rudder angle, ship motion state, ship motion force and torque during navigation, and construct a data set;

[0007] S2: Taking the three-degree-of-freedom ship motion mathematical model as the identification object, the linear regression model is determined according to the forces and moments exerted on the ship and the ship motion state in the ship maneuvering motion mathematical model;

[0008] S3: Taking the rudder angle, force and torque as training samples and based on the linear regression model, the training data is segmented to construct multiple sub-models according to the change characteristics of the rudder angle, and an improved Gaussian mixture model for ship maneuvering motion is obtained;

[0009] S4: training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model;

[0010] S5: Based on the motion state of the ship at the previous moment, the trained improved Gaussian mixture model is regressed to predict the fluid force of the ship at the next moment;

[0011] S6: Based on the predicted fluid force of the ship force at the next moment, the time derivative of the ship is solved to obtain the motion state of the ship at the next moment.

[0012] Furthermore: the linear regression model is expressed as follows:

[0013]

[0014] Among them; the input variables are The output variable is y = [X, Y, N] T; The parameter matrix is ​​β = [β 1 ,β 2 ,β 3 ] T .

[0015] Further: the process of segmenting the training data and constructing multiple sub-models according to the changing characteristics of the rudder angle is as follows:

[0016] By capturing the change of the rudder angle, segmenting is performed when the rudder angle change rate is ≥ 5° / s, and the longer data is divided into several segments.

[0017] Further: the step of training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model comprises the following steps:

[0018] Step S31: Give a local model of the linear regression model: the training data includes the steering angle, force and torque [δ, X, Y, N];

[0019] Step S32: Initialize mean μ k , covariance Σ k and the mixing coefficient α k , calculate the initial value of the log-likelihood function;

[0020] Step S33: maximizing the log-likelihood function as the objective function by using the expectation maximization algorithm, and outputting the trained improved Gaussian mixture model.

[0021] Further: the process of solving the maximization problem by using the expectation maximization algorithm and taking the log-likelihood function as the objective function and outputting the trained improved Gaussian mixture model is as follows:

[0022] First, through the expectation step, the posterior probability γ(z nk ), update the parameter posterior probability;

[0023] Then, through the maximization step, the parameters are re-estimated using the current posterior probability Update parameter mean Covariance matrix Mixing coefficient

[0024] Finally, the log-likelihood function lnp(A|μ,Σ,α) is calculated to check the convergence of the parameters or log-likelihood function. When the change of the model parameters between two iterations is less than 0.0001, the expected step is returned; otherwise, the trained improved Gaussian mixture model is output.

[0025] Further: The expression of the log-likelihood function is as follows:

[0026]

[0027] Where: lnp(A|μ,Σ,α) represents the log-likelihood function.

[0028] Further: based on the predicted fluid force of the ship force at the next moment, solving the time derivative of the ship to obtain the motion state of the ship at the next moment includes the following steps:

[0029] Step 51: Based on the trained improved Gaussian mixture model, the ship system input x at time s s =δ;

[0030] Step 52: Calculate input x s The posterior probability of each Gaussian component corresponding to the point

[0031] Step 53: Calculate the input x through the posterior probability s The corresponding target element

[0032] Step 54: By solving the improved Gaussian mixture model parameter matrix β, the prediction value corresponding to the query point is calculated through β

[0033] Step 55: Pass Solve the ship dynamics equation and calculate the spatial motion posture of the ship at the next moment s+1 and trajectory (x G ,y G );

[0034] Step 56: Calculate until δ c The corresponding ship motion state and the ship's position trajectory (x G ,y G ).

[0035] The present invention provides a non-parametric identification modeling method for ship maneuvering motion based on a Gaussian mixture model. Non-parametric identification is used to dynamically simulate the ship dynamics model. In order to reduce the error in the area where the data changes greatly and improve the recognition efficiency, the present application improves the traditional Gaussian mixture model and proposes a non-parametric identification model based on the improved Gaussian mixture model to simulate and predict the motion state of the ship. The improved Gaussian mixture model is trained using multi-dimensional data such as hydrodynamics and rudder angles, the traditional initial model data is improved in sections, and the expectation maximization algorithm is used to optimize the parameters. The improved model parameters are obtained by calculating the improved Gaussian mixture coefficients. Using the trained model and the predicted rudder angle, the hydrodynamics of the ship are calculated by Gaussian mixture regression, and the maneuvering motion state is solved by integration. The motion state of different rudder angles is tested by cyclotron test and Z-type test, and the feasibility and effectiveness of this method are verified to have the following advantages:

[0036] 1. The present invention can be closer to actual engineering application scenarios. Based on the identified ship motion mathematical model, the corresponding ship motion state can be calculated by inputting query variables.

[0037] 2. The present invention can predict the motion state of a ship in real time by identifying the relationship between forces and moments in the motion of the ship.

[0038] 3. Compared with the traditional Gaussian mixture model, the improved Gaussian mixture model used in this article has higher accuracy, faster operation efficiency, and greater engineering significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 This is a flow chart of a non-parametric identification modeling method for ship maneuvering based on a Gaussian mixture model of the present invention;

[0041] Figure 2 It is a ship motion dynamic prediction diagram based on an improved Gaussian mixture model of the present invention;

[0042] Figure 3 The results of the force and moment identification of the ship during the Z-type test, where (a) is under the forward force state, (b) is under the lateral force state, and (c) is under the bow moment state;

[0043] Figure 4 The results of the ship's force and moment (forward force, lateral force, bow moment) identified during the ship's turning test, where (a) is under forward force, (b) under lateral force, and (c) under bow moment;

[0044] Figure 5 The results of solving the dynamic motion state of the ship for the Z-type test, where (a) is under the forward speed state, (b) is under the lateral speed state, and (c) is under the bow angular velocity state;

[0045] Figure 6 The results of solving the dynamic motion state of the ship for the ship turning test, where (a) is under the forward speed state, (b) is under the lateral speed state, and (c) is under the bow angular velocity state;

[0046] Figure 7The results of the ship dynamic motion trajectory and attitude example solved by the present invention, where (a) is the Z-type test ship motion trajectory, (b) is the Z-type test ship rudder angle and heading;

[0047] Figure 8 The present invention provides a motion track of a gyration test ship. DETAILED DESCRIPTION

[0048] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0049] In order to make the purpose, 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 in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The present invention provides a non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model. Figure 1 As shown, it mainly includes the following steps:

[0051] A non-parametric identification modeling method for ship maneuvering based on a Gaussian mixture model, characterized by comprising the following steps:

[0052] S1: Obtain the rudder angle, ship motion state, ship motion force and torque during navigation, and construct a data set;

[0053] S2: Taking the three-degree-of-freedom ship motion mathematical model as the identification object, the linear regression model is determined according to the forces and moments exerted on the ship and the ship motion state in the ship maneuvering motion mathematical model;

[0054] S3: Taking the rudder angle, force and torque as training samples and based on the linear regression model, the training data is segmented to construct multiple sub-models according to the change characteristics of the rudder angle, and an improved Gaussian mixture model for ship maneuvering motion is obtained;

[0055] S4: training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model;

[0056] S5: Based on the motion state of the ship at the previous moment, the trained improved Gaussian mixture model is regressed to predict the fluid force of the ship at the next moment; the regression call adopts the Gaussian mixture regression method;

[0057] S6: Based on the predicted fluid force of the ship force at the next moment, the time derivative of the ship is solved to obtain the motion state of the ship at the next moment.

[0058] The steps S1 / S2 / S3 / S4 / S5 / S6 are performed sequentially;

[0059] The identified mathematical model of ship motion is a three-degree-of-freedom ship dynamics equation, which describes the relationship between the forces and moments on the ship in the three degrees of freedom of forward movement, transverse movement and bow turning and its motion state. The specific model is as follows:

[0060]

[0061] Among them, u, v, and r represent the forward speed, lateral speed, and bow angular speed respectively; They represent forward acceleration, lateral acceleration, and bow angular acceleration respectively; X, Y, and N represent the forward force, lateral force, and bow turning moment of the ship respectively; Represent three acceleration hydrodynamic derivatives respectively; m represents the mass of the ship; I ZZ represents the mass moment of inertia (rotational inertia) of the hull around the vertical axis passing through the center of gravity; δ represents the rudder angle of the ship.

[0062] The rudder angle, the motion state of the ship, and the motion force and torque of the ship are obtained during the navigation of the ship to construct a data set;

[0063] Firstly, a data set is constructed. To stimulate the ship motion characteristics, the ship navigation parameter data (rudder angle) δ in the ship maneuvering test and the corresponding ship motion state (forward speed, lateral speed, bow angular velocity) [u, v, r] data are taken as the main collection objects. The ship motion data with the same time interval Δs and different maneuvering angles are collected, and the collected data are segmented based on the change of rudder angle.

[0064] The collected data is processed in segments based on the change of the rudder angle; the improved Gaussian mixture model is initialized, and the parameters are optimized through the expectation maximization algorithm; the mixing coefficients in the optimized parameters are readjusted; the corresponding output (the forward force, lateral force, and bow turning moment corresponding to the rudder angle) is calculated by Gaussian mixture regression with the rudder angle as input; the forward force, lateral force, and bow turning moment output are used to calculate the ship's motion state at that moment through the ship motion mathematical model. Finally, the ship dynamic simulation is realized.

[0065] By capturing the change of the rudder angle, segmenting at the place where the rudder angle changes greatly, dividing the longer data into several segments of shorter data to reduce the complexity of the model, and then initializing each sub-model of the model and then training the model. The segmentation processing method reduces the computational burden by reducing the amount of data for each sub-model, significantly shortens the training time and improves the training efficiency. The longer data is data of 200 data points, and the shorter data is data of less than 200 data points;

[0066] The acceleration data in differential form of the input data is obtained by forward difference:

[0067]

[0068] Substitute the differential acceleration data into the above dynamic equation to obtain the output data y s =[X,Y,N] T , where Δs represents the time interval; s and s+1 represent a certain time node and the next time node, They represent the forward acceleration, lateral acceleration and bow angular acceleration at the sth moment respectively.

[0069] For the nonlinear mapping relationship in ship maneuvering, multiple sub-models (Gaussian mixture sub-models) are used to identify this correspondence. Here, the local model adopts a linear regression model:

[0070]

[0071] The input is: The output is: y = [X, Y, N] T The matrix β represents the mapping function between the input matrix and the output matrix.

[0072] The expression of the improved Gaussian mixture model is as follows:

[0073]

[0074] Where: μ k Mean, Σ k is the covariance matrix, α k is the mixing coefficient, x represents the observed data vector, which includes the ship's rudder angle and hydrodynamic matrix; p(x) is the total probability density of the mixed model; K represents the number of Gaussian components;

[0075] Figure 2 It is a ship motion dynamic prediction diagram based on Gaussian mixture model of the present invention;

[0076] Furthermore, the training of the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model includes the following steps:

[0077] Step S31: Give a local model of the linear regression model: the training data includes the steering angle, force and torque [δ, X, Y, N];

[0078] Step S32: Initialize mean μ k , covariance Σ k and the mixing coefficient α k , calculate the initial value of the log-likelihood function;

[0079] Step S33: maximizing the log-likelihood function as the objective function by using the expectation maximization algorithm. The specific process is as follows:

[0080] First, through the expectation step, the posterior probability γ(z nk ), update the posterior probability of the parameters, and then re-estimate the parameters using the current posterior probability through the maximization step Update parameter mean Covariance matrix Mixing coefficient

[0081] Finally, the log-likelihood function lnp(A|μ,Σ,α) is calculated to check the convergence of the parameters or log-likelihood function. If the convergence criteria are not met, the expected step is returned.

[0082] The convergence criterion is: when the change of the Gaussian mixture model parameters between two iterations is less than a threshold value of 0.0001.

[0083] The parameters in the expected step satisfy:

[0084]

[0085] The parameters in the maximization step satisfy:

[0086]

[0087] in:

[0088]

[0089] The x n represents the nth sample data point, K represents the number of Gaussian components, k represents the kth Gaussian component, N k denotes that the kth Gaussian distribution is “responsible” for generating the sum of all data points, N denotes the number of data points in the observed data, and NEW denotes the updated parameters.

[0090] Step S34: output the trained improved Gaussian mixture model.

[0091] (5) Re-adjust the mixing coefficient and reduce the mixing coefficient of the Gaussian component of the covariance whose first covariance parameter is greater than 150 to 0 to improve the recognition effect of the model.

[0092] The improved Gaussian mixture model improves the traditional Gaussian mixture model by re-adjusting the mixing coefficient, reducing the mixing coefficient of the Gaussian component with larger covariance, and considering the impact of data complexity on the modeling effect, segmenting the training data to improve the recognition effect of the model. The improved Gaussian mixture model has the ability to model high-dimensional complex data sets, especially when the data changes greatly, to achieve more accurate and robust model performance.

[0093] Furthermore, the expression of the log-likelihood function is as follows:

[0094]

[0095] Where: lnp(A|μ,Σ,α) represents the log-likelihood function.

[0096] Further: based on the predicted fluid force of the ship force at the next moment, solving the time derivative of the ship to obtain the motion state of the ship at the next moment includes the following steps:

[0097] (6-1) Step S51: Based on the trained improved Gaussian mixture model and the ship system input x at time s s =δ;

[0098] (6-2) Step S52: Calculate input x s The posterior probability of each Gaussian component corresponding to the point

[0099]

[0100] Step 53: Calculate the input x through the posterior probability s The corresponding target element

[0101]

[0102] (6-3) Step S54: By solving the improved Gaussian mixture model parameter matrix β, the prediction value corresponding to the query point is calculated using β

[0103] (6-4) Step S55: By Solve the ship dynamics equation and calculate the spatial motion posture of the ship at the next moment s+1 and trajectory (x G ,y G). First, calculate the time derivative and position derivative by the following formula (15), and then calculate the ship's motion posture and trajectory by integral solution;

[0104] (6-5) Step S56: Calculate until δ c The corresponding ship motion state and the ship's position trajectory (x G ,y G ).

[0105] (6-6) According to step S55, the ship motion state attitude derivative at time s can be obtained by solving the ship dynamics equation by [X, Y, N]. and the ship trajectory derivative Specifically:

[0106]

[0107] Substitute [X, Y, N] obtained in the above steps into the above formula and calculate to find the posture [δ, u, v, r] and trajectory (x G ,y G ).

[0108] Figure 3 The results of the force and moment identification of the ship during the Z-type test, where (a) is under the forward force state, (b) is under the lateral force state, and (c) is under the bow moment state;

[0109] Figure 4 The results of the ship's force and moment (forward force, lateral force, bow moment) identified during the ship's turning test, where (a) is under forward force, (b) under lateral force, and (c) under bow moment;

[0110] (6-7) According to step S56, until reaching δ c The corresponding ship motion state [δ c ,u,v,r],(x G ,y G ). Specifically, repeat (6-1)-(6-6) to make the rudder angle δ of the ship's motion state at a certain time d d =δ c , to achieve dynamic simulation of ship motion, such as Figure 5 shown.

[0111] Among them: the time setting is 350s for Z-type experiment and 560s for cyclotron test.

[0112] Figure 5The results of solving the dynamic motion state of the ship for the Z-type test, where (a) is under the forward speed state, (b) is under the lateral speed state, and (c) is under the bow angular velocity state;

[0113] Figure 6 The results of solving the dynamic motion state of the ship for the ship turning test, where (a) is under the forward speed state, (b) is under the lateral speed state, and (c) is under the bow angular velocity state;

[0114] Figure 7 The results of the ship dynamic motion trajectory and attitude example solved by the present invention, where (a) is the Z-type test ship motion trajectory, (b) is the Z-type test ship rudder angle and heading;

[0115] Figure 8 The present invention provides a motion track of a gyration test ship.

[0116] A non-parametric identification modeling device for ship maneuvering based on a Gaussian mixture model, comprising:

[0117] Acquisition module: used to obtain the rudder angle, ship motion state, and ship motion force and torque during navigation, and construct a data set;

[0118] Constructing a linear regression model: It is used to determine the linear regression model based on the forces and moments on the ship and the ship's motion state in the mathematical model of ship maneuvering motion, taking the three-degree-of-freedom ship motion mathematical model as the identification object;

[0119] Constructing an improved Gaussian mixture model module: It is used to construct multiple sub-models for segmenting the training data according to the changing characteristics of the rudder angle, using the rudder angle, force and torque as training samples and based on the linear regression model, to obtain an improved Gaussian mixture model for ship maneuvering motion;

[0120] Training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model;

[0121] Regression module: It is used to regress and call the trained improved Gaussian mixture model based on the motion state of the ship at the previous moment, so as to predict the fluid force of the ship at the next moment;

[0122] State module for solving: It is used to solve the time derivative of the ship based on the predicted fluid force at the next moment of the ship force to obtain the motion state of the ship at the next moment.

[0123] The above-mentioned embodiment numbers of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0124] In the above embodiments, each embodiment focuses on different aspects. For parts not described in detail in a certain embodiment, reference may be made to the relevant descriptions in other embodiments.

[0125] In various embodiments of the present application, the disclosed technical content can be implemented in a variety of ways. The above-mentioned device embodiments are only examples. For example, the division of units can be based on logical functions, but other division methods can also be used in actual implementation. Multiple units or components can be combined, integrated into other systems or ignore certain features. In addition, coupling, direct coupling or communication connection can be an indirect coupling or communication connection of an interface, unit or module, and the connection form can be an electrical connection or other forms. When a unit is described as an independent component, it can be physically separated or non-physically separated, and the unit can be concentrated or distributed in multiple locations. Some or all of the units can be selected according to actual needs to achieve the purpose of this embodiment.

[0126] 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 it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

[0127] [1] KJ, C G.Identification of ship steering dynamics[J].Au tomatica,1976,12(1):9-22.

[0128] [2]Wang Z, Zou Z, Soares C G.Identification of ship manoeuvring motionbased on nu-support vector machine[J].Ocean Engineering, 2019,183:270-281.

[0129] [3]Bai W,Ren J,Li T.Modified genetic optimization-based locallyweighted learning identification modeling of ship maneuvering with full scaletrial[J].Future generation computer systems,2019,93:1036-1045.

[0130] [4] Zhang Z, Ren J, Bai W. MIMO non-parametric modeling of ship maneuvering motion for marine simulator using adaptive moment estimation locally weighted learning [J]. Ocean Engineering, 2022, 261: 112103.

[0131] [5] Cao Weinan. Ship heading model identification and L_2 closed-loop shaping filter control based on EKF[D]. Dalian Maritime University, 2020.

[0132] [6] Qin Cao. Identification of mathematical model of ship motion based on unscented Kalman filter[J]. Ship Science and Technology, 2021, 43(01): 89-94.

Claims

1. A non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model, characterized in that: The following steps are involved: S1: Obtain the rudder angle, ship motion state, ship motion force and torque during navigation, and construct a data set; S2: Taking the three-degree-of-freedom ship motion mathematical model as the identification object, the linear regression model is determined according to the forces and moments exerted on the ship and the ship motion state in the ship maneuvering motion mathematical model; S3: Taking the rudder angle, force and torque as training samples and based on the linear regression model, the training data is segmented to construct multiple sub-models according to the change characteristics of the rudder angle, and an improved Gaussian mixture model for ship maneuvering motion is obtained; S4: training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model; S5: Based on the motion state of the ship at the previous moment, the trained improved Gaussian mixture model is regressed to predict the fluid force of the ship at the next moment; S6: Based on the predicted fluid force of the ship force at the next moment, the time derivative of the ship is solved to obtain the motion state of the ship at the next moment.

2. The non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model according to claim 1 is further characterized by: The linear regression model is expressed as follows: in; The input variables are The output variable is y = [X, Y, N] T ; The parameter matrix is ​​β = [β1, β2, β3] T .

3. The non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model according to claim 1 is further characterized by: The process of segmenting the training data and constructing multiple sub-models according to the changing characteristics of the rudder angle is as follows: By capturing the change of the rudder angle, segmenting is performed when the rudder angle change rate is ≥ 5° / s, and the longer data is divided into several segments.

4. The non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model according to claim 1 is further characterized by: The training of the improved Gaussian mixture model to obtain the trained improved Gaussian mixture model comprises the following steps: Step S31: Give a local model of the linear regression model: the training data includes the steering angle, force and torque [δ, X, Y, N]; Step S32: Initialize mean μ k , covariance Σ k and the mixing coefficient α k , calculate the initial value of the log-likelihood function; Step S33: maximizing the log-likelihood function as the objective function by using the expectation maximization algorithm, and outputting the trained improved Gaussian mixture model.

5. The non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model according to claim 1 is further characterized by: The process of solving the maximization problem by using the expectation maximization algorithm and taking the log-likelihood function as the objective function and outputting the trained improved Gaussian mixture model is as follows: First, through the expectation step, the posterior probability γ(z nk ), update the parameter posterior probability; Then, through the maximization step, the parameters are re-estimated using the current posterior probability Update parameter mean Covariance matrix Mixing coefficient Finally, the log-likelihood function ln p(A|μ,Σ,α) is calculated to check the convergence of the parameters or log-likelihood function. When the change of the model parameters between two iterations is less than 0.0001, the expected step is returned; otherwise, the trained improved Gaussian mixture model is output.

6. The non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model according to claim 1 is further characterized by: The expression of the log-likelihood function is as follows: Where: lnp(A|μ,Σ,α) represents the log-likelihood function.

7. The non-parametric identification modeling method for ship maneuvering based on Gaussian mixture model according to claim 1 is further characterized by: Based on the predicted fluid force of the ship force at the next moment, solving the time derivative of the ship to obtain the motion state of the ship at the next moment includes the following steps: Step 51: Based on the trained improved Gaussian mixture model, the ship system input x at time s s =δ; Step 52: Calculate the input x s The posterior probability of each Gaussian component corresponding to the point Step 53: Calculate the input x through the posterior probability s The corresponding target element Step 54: By solving the improved Gaussian mixture model parameter matrix β, the prediction value corresponding to the query point is calculated through β Step 55: Pass Solve the ship dynamics equation and calculate the spatial motion posture of the ship at the next moment s+1 and trajectory (x G ,y G ); Step 56: Calculate until δ c The corresponding ship motion state and the ship's position trajectory (x G ,y G ).

8. A non-parametric identification modeling device for ship maneuvering based on Gaussian mixture model, characterized in that: include: Acquisition module: used to obtain the rudder angle, ship motion state, and ship motion force and torque during navigation, and construct a data set; Constructing a linear regression model: It is used to determine the linear regression model based on the forces and moments on the ship and the ship's motion state in the mathematical model of ship maneuvering motion, taking the three-degree-of-freedom ship motion mathematical model as the identification object; Constructing an improved Gaussian mixture model module: It is used to construct multiple sub-models for segmenting the training data according to the changing characteristics of the rudder angle, using the rudder angle, force and torque as training samples and based on the linear regression model, to obtain an improved Gaussian mixture model for ship maneuvering motion; Training the improved Gaussian mixture model to obtain a trained improved Gaussian mixture model; Regression module: It is used to regress and call the trained improved Gaussian mixture model based on the motion state of the ship at the previous moment, so as to predict the fluid force of the ship at the next moment; State module for solving: It is used to solve the time derivative of the ship based on the predicted fluid force at the next moment of the ship force to obtain the motion state of the ship at the next moment.

Images