Intelligent ship maneuverability deduction improvement method based on physical information neural network

Abstract

The invention relates to the field of ship maneuverability intelligent forecasting, in particular to an intelligent ship maneuverability deduction improvement method based on a physical information neural network, and the method comprises the steps: building a ship planar motion linear mathematical model; constructing a manipulation motion physical model, substituting the manipulation motion physical model into the physical information neural network, and forecasting the ship maneuverability; the ship plane motion linear mathematical model, the control motion physical model and the ship maneuverability deduction physical information neural network jointly form a ship maneuverability model, so that ship maneuverability data are forecasted in real time, the ship maneuverability model is optimized through two-time off-line iteration and one-time on-line iteration, forecasting precision is improved, and the ship maneuverability deduction performance is improved. The system has the advantages of intelligence, high precision and high speed.

Description

Technical Field

[0001] The invention relates to the field of intelligent prediction of ship maneuverability, and in particular to an improved method for deducing maneuverability of an intelligent ship based on a physical information neural network. Background Art

[0002] In recent years, with the increasing complexity and automation of ship design, traditional ship maneuverability prediction methods have gradually exposed their limitations, such as low accuracy of empirical formulas, cumbersome and time-consuming measurements of actual ships, etc. In addition, the accuracy of these traditional methods often relies on experience and simplified assumptions, making it difficult to cope with complex maneuvering conditions and changes in ship design, especially for new ships or nonlinear maneuvering conditions. There is an urgent need for an intelligent prediction technology that has both high calculation speed and high accuracy. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an improved method for intelligent ship maneuverability deduction based on physical information neural network, comprising the following steps: S1. Establish a linear mathematical model of ship planar motion; S2. construct a physical model of maneuvering motion, and bring the physical model of maneuvering motion into a physical information neural network to obtain a physical information neural network for ship maneuverability deduction; S3, the ship's plane motion linear mathematical model, the maneuvering motion physical model and the ship's maneuverability deduction physical information neural network together constitute the ship's maneuvering model, which is optimized using two offline iterations and one online iteration.

[0004] In a preferred embodiment, the linear mathematical model of the ship planar motion is: ; Where m is the mass of the ship, , The correlation coefficient of the force of the ship's lateral and longitudinal motion, is the correlation coefficient of the torque on the yaw motion, represents the environmental inertia force, is the environmental inertia moment, is the force coefficient related to the yaw rate, is the torque coefficient related to the yaw rate, are the lateral force and yaw moment of the ship, respectively, which are controlled by the rudder angle δ. is the change, u is the longitudinal velocity, v is the lateral velocity, and r is the yaw angular velocity; Converting the above equations into dimensionless form, we can obtain the ship's inertial force derivative matrix, viscosity force derivative matrix, and rudder force derivative matrix, where L is the length of the ship: ; The prime in the above matrix represents the first-order derivative.

[0005] In a preferred embodiment, the physical model of the manipulation motion in step S2 is expressed as: ; In the formula, the x-axis points to the bow, the y-axis points to the starboard, the z-axis points to the direction of gravity, and m represents the mass of the ship. represents the longitudinal additional mass of the hull, represents the transverse additional mass of the hull, u represents the longitudinal speed of the ship, v represents the transverse speed of the ship, r represents the bow angular velocity of the ship, is the hull inertia moment, represents the additional moment of inertia of the hull, They represent the longitudinal bare ship force, viscous force, wave force and wind resistance, respectively. They represent the bare ship force, viscous force, wave force and wind resistance in the transverse direction of the hull, respectively. They represent bare ship moment, viscous moment, wave moment and wind resistance moment respectively; The above-mentioned maneuvering motion physical model is added to the loss function of the neural network to obtain a physical information neural network for ship maneuverability deduction.

[0006] In a preferred embodiment, in step S3, three rounds of correction are performed on the ship maneuvering model, which are two rounds of offline correction and one round of online correction: ; In the formula, represents the initial model, , represents the correction amount in the first round of offline optimization, , represents the intermediate model after the first round of corrections, represents the correction amount in the second round of offline optimization, represents the intermediate model after the second round of revisions, represents the correction amount in the third round of online optimization, Represents the final model of ship maneuvering.

[0007] In a preferred embodiment, firstly, offline correction is performed by regression iteration, hydrodynamic data is obtained by model test, and the hydrodynamic simulation model is optimized, and the model correction value is obtained by using the optimized simulation model. , realize offline correction of digital model; carry out real ship test, use real ship motion state data to optimize the hydrodynamic simulation model again, and obtain the correction value of digital model , and then obtain a digital model of navigation performance based on offline correction of actual ship test data.

[0008] In a preferred embodiment, the digital model of ship maneuverability after regression iteration correction is corrected offline by an optimization learning method; Afterwards, online correction is carried out, and the ship status information, control information, and marine environment information are used as input to drive the physical model of the control motion. At the same time, the physical model simulation results and the actual status, control, marine environment and other information of the ship are used as input to drive the physical information neural network, thereby realizing the deduction of the ship's control performance.

[0009] Compared with the prior art, the present invention has the following advantages: By establishing a mathematical model of ship maneuvering motion and using relevant physical equations to constrain the neural network, the ship maneuvering performance data can be predicted in real time, and the prediction accuracy can be improved through three iterative optimizations. It has the advantages of intelligence, high precision and high speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic diagram of establishing a coordinate system in the ship maneuvering model of the present invention; Figure 2 It is a physical information neural network structure diagram of the present invention; Figure 3 This is the first offline correction schematic diagram of the present invention Figure 4 This is a schematic diagram of the second offline correction of the present invention; Figure 5 This is a schematic diagram of the first online correction of the present invention; Figure 6 It is a comparison between the route map deduced by the method of the present invention and the actual route map; DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0012] In the drawings of the specific embodiments of the present invention, in order to better and more clearly describe the working principles of the various components in the system, the connection relationship of the various parts in the device is shown, which only clearly distinguishes the relative position relationship between the various components, and cannot constitute a limitation on the signal transmission direction, connection sequence and structural size, size and shape of each part within the component or structure.

[0013] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0014] A method for improving the maneuverability of an intelligent ship based on a physical information neural network comprises the following steps: S1. Establish a linear mathematical model for the ship's plane motion.

[0015] The geodetic coordinate system and the body coordinate system are used together. The geodetic coordinate system is used to determine the specific position and navigation route, and the body coordinate system is used to describe the six-degree-of-freedom motion of the ship. Figure 1 As shown in the figure, the coupling relationship between the two coordinate systems is shown, and the upper right corner shows the ship planar motion model ignoring the vertical motion, where r is the angular velocity of the bow rotation and V is the current propeller speed.

[0016] ; ; ; is the motion evaluation matrix is the heading motion evaluation matrix, is the pitch motion evaluation matrix, is the roll motion evaluation matrix in, Angles used to describe the pitch, roll and yaw of a ship. 0 represents the speed in the z direction, u 0 represents the longitudinal velocity component of the ship's motion, v 0 represents the lateral velocity component of the ship's motion, (x 初 ,y 初 , z 初 ) is the initial position coordinate. Since the focus during normal navigation is the coordinates of the ship on the sea level, the position of the ship to be described is simplified by (x, y), so that the kinematic equation reflecting the real-time position of the ship can be obtained.

[0017] ;

[0018] u represents the longitudinal speed of the ship, v represents the lateral speed of the ship. After obtaining the kinematic model of the actual ship, the dynamic model needs to be established. The two are combined to obtain the mathematical model of the plane motion of the ship. 0 ,y 0 Refers to the horizontal and vertical coordinates of the origin of the body coordinate system, which is the center of mass of the ship.

[0019] ;

[0020] X is the longitudinal force on the hull, Y is the transverse force on the hull, and N is the moment on the hull; the actual motion of the ship during navigation is strongly coupled and nonlinear. Although small fluctuations in buoyancy, sinking, or rolling and pitching will interfere with the motion of the vehicle, when vertical motion is ignored, the high-order small variables on the right side of the above equation can be ignored, which can obtain a more accurate linear mathematical model of the ship's planar motion.

[0021] ;

[0022] Where m is the mass of the ship, , Correlation coefficient of the forces of lateral and longitudinal motion of the ship. is the correlation coefficient of the torque on the yaw motion. Indicates the rate change in the horizontal and vertical directions, that is, the environmental inertia force, is the environmental inertia moment. is the force coefficient related to the rotational speed of the ship, i.e. the yaw rate, is the torque coefficient related to the ship's rotational speed, i.e., the yaw rate, are the lateral force and yaw moment of the ship, respectively, which are controlled by the rudder angle δ. is the change, u is the longitudinal velocity, v is the lateral velocity, r is the rotational velocity, that is, the yaw angular velocity. By converting the above equations into dimensionless style, we can get the inertial force derivative matrix, viscosity force derivative matrix and rudder force derivative matrix of the ship, and L is the length of the ship.

[0023] The inertial force derivative matrix is: ;

[0024] The viscosity force derivative matrix is: ;

[0025] The rudder force derivative matrix is: ;

[0026] ;

[0027] in, is the rudder angle, L is the length of the ship, and the primes in the above matrices represent the first-order derivatives.

[0028] S2. Construct a physical model of maneuvering motion, and bring the physical model of maneuvering motion into a physical information neural network to obtain a physical information neural network for ship maneuverability deduction.

[0029] The physical model of maneuvering motion is expressed by the three-degree-of-freedom MMG equation of the ship, which is expressed as follows: ;

[0030] In the formula, the x-axis points to the bow, the y-axis points to the starboard, the z-axis points to the direction of gravity, and m represents the mass of the ship. represents the longitudinal additional mass of the hull, represents the transverse additional mass of the hull, u represents the longitudinal speed of the ship, v represents the transverse speed of the ship, r represents the bow angular velocity of the ship, is the hull inertia moment, represents the additional moment of inertia of the hull, They represent the longitudinal bare ship force, viscous force, wave force and wind resistance, respectively. They represent the bare ship force, viscous force, wave force and wind resistance in the transverse direction of the hull, respectively. They represent the bare ship moment, viscous moment, wave moment and wind resistance moment respectively.

[0031] The above maneuvering motion physical model is added to the loss function of the neural network to obtain the physical information neural network for ship maneuverability deduction. The structure diagram is shown in Figure 2 .

[0032] Relying on traditional computational fluid dynamics methods to obtain training sets and validation sets, and comparing the predicted results with the actual results, the error does not exceed 0.14%.

[0033]

[0034] S3, the ship's plane motion linear mathematical model, the maneuvering motion physical model and the ship's maneuverability deduction physical information neural network together constitute the ship's maneuvering model, which is optimized using two offline iterations and one online iteration.

[0035] For the ship maneuvering model, it needs to go through multiple correction processes to achieve the maximum match with the actual ship. For this reason, this method designs three rounds of correction work, namely two rounds of offline correction and one round of online correction.

[0036] ;

[0037] In the formula, represents the initial model, , represents the correction amount in the first round of offline optimization, , represents the intermediate model after the first round of corrections, (Optimization learning) represents the correction amount in the second round of offline optimization, represents the intermediate model after the second round of revisions, (On line) indicates the correction amount in the third round of online optimization. Represents the final model of ship maneuvering.

[0038] First, offline correction is performed using regression iteration, hydrodynamic data is obtained using model tests, and the hydrodynamic simulation model is optimized. The model correction value is obtained using the optimized simulation model. On the basis of the above, the actual ship test was carried out, and the hydrodynamic simulation model was optimized again using the actual ship motion state data to obtain the digital model correction value. , and then obtain a digital model of navigation performance based on offline correction of actual ship test data.

[0039] It should be explained here that the ship maneuvering model consists of two parts, one is the mathematical model of ship maneuvering motion, and the other is the physical model of ship maneuvering motion. The establishment of the physical model of ship maneuvering motion requires an evolutionary process. It is necessary to establish an initial model called a hydrodynamic simulation model through model testing. After adding data and correcting it, a digital model of navigation performance will be obtained. After the digital model of navigation performance is corrected online by adding new physical information and data, the above-mentioned final model of ship maneuvering will be obtained. The physical model of ship maneuvering motion is also the MMG equation.

[0040] Figure 3 It represents the first offline correction, which is performed by regression iteration. The hydrodynamic data is obtained by model test, and the hydrodynamic simulation model is optimized. The model correction value is obtained by using the optimized simulation model. On the basis of the above, the actual ship test was carried out, and the hydrodynamic simulation model was optimized again using the actual ship motion state data to obtain the digital model correction value. , and then obtain a digital model of navigation performance based on offline correction of actual ship test data.

[0041] For the digital model of navigation performance after regression iteration correction, we also need to further correct it offline through the method of optimal learning. Before the simulation is initialized, the static parameters, working condition loading and other information are input once, which are the invariants of the entire simulation process; the simulation model reserves the model dynamic parameter adjustment interface, and iterates and evolves through optimal learning methods such as genetic algorithms, and dynamically adjusts the model parameters within a certain range of changes through correction coefficients, so that the digital model can continuously have the ability of accurate self-learning during operation.

[0042] Figure 4It represents the second offline correction: In the second round of offline correction, new data needs to be added to the digital model of navigation performance for updating. The new data includes fixed configuration parameters, such as ship mass, geometric parameters, etc., and also includes variable dynamic parameters, such as propeller wake component and thrust reduction, ship viscous hydrodynamic derivatives under different states, ship inertial hydrodynamic derivatives under different states, etc. The addition of these dynamic parameters requires the use of genetic algorithms and optimization algorithms. The operating mechanism of the digital model of navigation performance is that the front-end input parameters are ship speed, wind speed and wind direction, wave speed, direction, wavelength, propeller speed, rudder angle, etc., and the output parameters are ship position, ship main engine power, bow turning speed, ship swaying speed, etc.

[0043] Before the simulation is initialized, static parameters, working condition loading and other information are input once, which are the invariants of the entire simulation process. The simulation model reserves a dynamic parameter adjustment interface for the model, and iteratively evolves through optimization learning methods such as genetic algorithms. The correction coefficient dynamically adjusts the model parameters within a certain range of variation, so that the digital model can continuously have the ability of accurate self-learning during operation.

[0044] get Afterwards, online correction is performed, and the ship's state information (position, speed, etc.), control information (propeller speed, rudder angle, etc.), and ocean environment information (wind, waves, etc.) are used as input to drive the control motion physical model (MMG equation); at the same time, the physical model simulation results and the ship's actual state, control, ocean environment and other information are used as input to drive the physical information neural network, thereby achieving more accurate deduction of the ship's control performance.

[0045] Figure 5 It means that the online correction takes the ship status information (position, speed, etc.), control information (propeller speed, rudder angle, etc.), and marine environment information (wind, waves, etc.) as input to drive the maneuvering motion physical model (MMG equation); at the same time, the physical model simulation results and the actual state, control, marine environment and other information of the ship are used as input to drive the physical information neural network model, thereby achieving more accurate ship maneuvering performance deduction. The hull model, propeller model, wind model, and wave force model are input into the equation as independent variables of the MMG equation. When the entire network is running, the data of the real-time operation status of the ship is needed to prevent network calculation deviations through supervised learning. The ship's operational physical equations use relevant physical equations to constrain the neural network.

[0046] Figure 6 This is a real ship verification of this method. The trajectory predicted by this method is consistent with the actual ship's navigation trajectory.

[0047] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0048] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for improving the maneuverability of intelligent ships based on physical information neural network, characterized in that: The following steps are involved: S1. Establish a linear mathematical model of ship planar motion; S2. construct a physical model of maneuvering motion, and bring the physical model of maneuvering motion into a physical information neural network to obtain a physical information neural network for ship maneuverability deduction; S3, the ship's plane motion linear mathematical model, the maneuvering motion physical model and the ship's maneuverability deduction physical information neural network together constitute the ship's maneuvering model, which is optimized using two offline iterations and one online iteration.

2. The method for improving the maneuverability of an intelligent ship based on a physical information neural network according to claim 1 is characterized in that: The linear mathematical model of the ship's plane motion is: ； Where m is the mass of the ship, , The correlation coefficient of the force of the ship's lateral and longitudinal motion, is the correlation coefficient of the torque on the yaw motion, represents the environmental inertia force, is the environmental inertia moment, is the force coefficient related to the yaw rate, is the torque coefficient related to the yaw rate, are the lateral force and yaw moment of the ship, respectively, which are controlled by the rudder angle δ. is the change, u is the longitudinal velocity, v is the lateral velocity, and r is the yaw angular velocity; Converting the above equations into dimensionless form, we can obtain the ship's inertial force derivative matrix, viscosity force derivative matrix, and rudder force derivative matrix, where L is the length of the ship: ； The prime in the above matrix represents the first-order derivative.

3. The method for improving the maneuverability of an intelligent ship based on a physical information neural network according to claim 2 is characterized in that: The physical model of the manipulation motion in step S2 is expressed as: ； In the formula, the x-axis points to the bow, the y-axis points to the starboard, the z-axis points to the direction of gravity, and m represents the mass of the ship. represents the longitudinal additional mass of the hull, represents the transverse additional mass of the hull, u represents the longitudinal speed of the ship, v represents the transverse speed of the ship, r represents the bow angular velocity of the ship, is the hull inertia moment, represents the additional moment of inertia of the hull, They represent the longitudinal bare ship force, viscous force, wave force and wind resistance, respectively. They represent the bare ship force, viscous force, wave force and wind resistance in the transverse direction of the hull, respectively. They represent bare ship moment, viscous moment, wave moment and wind resistance moment respectively; The above-mentioned maneuvering motion physical model is added to the loss function of the neural network to obtain a physical information neural network for ship maneuverability deduction.

4. The method for improving the maneuverability of an intelligent ship based on a physical information neural network according to claim 1 is characterized in that: In step S3, three rounds of corrections are performed on the ship maneuvering model, including two rounds of offline corrections and one round of online correction: ； In the formula, represents the initial model, , represents the correction amount in the first round of offline optimization, , represents the intermediate model after the first round of corrections, represents the correction amount in the second round of offline optimization, represents the intermediate model after the second round of revisions, represents the correction amount in the third round of online optimization, Represents the final model of ship maneuvering.

5. The method for improving the maneuverability of an intelligent ship based on a physical information neural network according to claim 4 is characterized in that: First, offline correction is performed using regression iteration, hydrodynamic data is obtained using model tests, and the hydrodynamic simulation model is optimized. The model correction value is obtained using the optimized simulation model. , realize offline correction of digital model; carry out real ship test, use real ship motion state data to optimize the hydrodynamic simulation model again, and obtain the correction value of digital model , and then obtain a digital model of navigation performance based on offline correction of actual ship test data.

6. The method for improving the maneuverability of an intelligent ship based on a physical information neural network according to claim 5 is characterized in that: For the digital model of ship maneuverability after regression iteration correction, offline correction is performed through the optimization learning method; Afterwards, online correction is performed, and the ship status information, control information, and marine environment information are used as input to drive the physical model of the control motion. At the same time, the physical model simulation results and the actual status, control, and marine environment information of the ship are used as input to drive the physical information neural network, thereby realizing the deduction of the ship's control performance.

Images