An Online Modeling and Prediction Method for the State Information of Unmanned Vessels Based on Simulation Technology

Abstract

The present invention discloses an online modeling and prediction method for the state information of an unmanned ship based on simulation technology, which relates to the technical field of unmanned ship simulation testing. A database system for the unmanned ship is constructed, including simulation data obtained by simulating an unmanned ship model in a real flow field of star-ccm+ and a Noetic ocean environment, as well as data collected during real sea trials. A nonlinear reservoir computing algorithm based on a sliding window algorithm is designed to perform online modeling and prediction of the state information. By combining the dynamic characteristics of computational fluid dynamics, physical simulation environment, and real ocean environment, the present invention can enable the unmanned ship to realize online modeling and prediction of state information using small sample data in the real ocean environment, provide effective prior information for collision avoidance, planning, and control of the unmanned ship, and safely and effectively complete the navigation task.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned ship simulation testing, and in particular to an online modeling and prediction method for the state information of an unmanned ship based on simulation technology. Background Art

[0002] In order to better explore the ocean, develop and utilize ocean resources, ocean unmanned ships are all developing towards the direction of unmanned and intelligent. Compared with ordinary manned ships, the advantages of small unmanned ships are mainly in the following aspects: (1) small volume, low energy consumption, flexible movement, which can not only go to the open sea to perform tasks, but also carry out operations in special terrains such as shoals and river channels; (2) high autonomy, does not need to rely on crew members, only requires a few people for remote control, can ensure the safety of personnel, and can operate for a long time in harsh environments; (3) high operation accuracy and strong anti-interference ability, which can reduce errors caused by manual operation.

[0003] When an unmanned ship is performing tasks, it may encounter various problems. Real-time monitoring and prediction of the navigation state of the unmanned ship are crucial for the safety and control of the unmanned ship. In the prior art, there is a lack of a reliable prediction method to predict the dynamic characteristics of the unmanned ship. Summary of the Invention

[0004] In order to overcome the above problems existing in the prior art, the present invention proposes an online modeling and prediction method for the state information of an unmanned ship based on simulation technology.

[0005] The technical solution adopted by the present invention to solve its technical problems is: an online modeling and prediction method for the state information of an unmanned ship based on simulation technology, including the following steps:

[0006] Step 1, simulate the unmanned ship model to obtain the navigation data of the unmanned ship in the fluid;

[0007] Step 2, build an ocean environment simulator, simulate the unmanned ship model, and obtain the navigation data of the unmanned ship in the ocean environment;

[0008] Step 3, use the navigation data collected in Step 1 and Step 2 as an offline training set, and use the sea trial data of the unmanned ship as an online training set and an online verification set to build an unmanned ship database;

[0009] Step 4, design a nonlinear reservoir algorithm based on the sliding window method to realize the online modeling and prediction of the state information of the unmanned ship.

[0010] For the above online modeling and prediction method for the state information of an unmanned ship based on simulation technology, the specific content of Step 1 includes:

[0011] Step 1.1, import the unmanned ship model into star-ccm+;

[0012] In Step 1.2, hexahedral cut cell meshes are selected for the overall mesh division of the computational domain, and prism layer meshes and surface reconstruction are adopted near the hull wall surface;

[0013] In Step 1.3, the model is adopted;

[0014] In Step 1.4, the front boundary of the computational domain is the velocity inlet boundary condition, the rear boundary of the computational domain is the pressure outlet boundary condition, the upper, lower and both side boundaries of the computational domain are the velocity inlet boundary conditions, and the damping wave elimination method is adopted at the outlet boundary for wave elimination;

[0015] In Step 1.5, CFD simulation is carried out to collect motion data as the offline training set O 1 .

[0016] For the above online modeling and prediction method of unmanned ship state information based on simulation technology, the specific steps of Step 2 include:

[0017] In Step 2.1, a Noetic ocean environment simulator is built, and the unmanned ship model is imported into the ocean simulation environment;

[0018] In Step 2.2, a hydrodynamic dynamic plug-in is designed based on the hull segmentation method;

[0019] In Step 2.3, the water surface displacement at each grid point is determined based on the sum of component waves, the buoyancy generated is determined based on the position of the ship grid point relative to the water surface, the force condition is directly applied to the corresponding grid point, and the dynamic change of the unmanned ship attitude is realized according to the water surface displacement and force condition of each part;

[0020] In Step 2.4, collect motion data as the offline training set O 2 .

[0021] For the above online modeling and prediction method of unmanned ship state information based on simulation technology, the specific steps of Step 2.2 include: dividing the unmanned ship model into six parts; determining the position of the hull grid points based on the current position and attitude of the ship, calculating the component waves of each part with Gerstner waves and three independent superposition wave functions at the position , and generating the wave height at the position , and the specific calculation formula is:

[0022] ;

[0023] Among them, for each component wave, i is the simulation time, N is the end time, is the steepness, is the amplitude, is the wave vector, is the angular frequency, is the phase; the wave vector is the horizontal vector in the wave propagation direction, and its magnitude is equal to the wave number , where is the wavelength.

[0024] The above-mentioned online modeling and prediction method for the state information of an unmanned ship based on simulation technology, the specific steps of step 3 are as follows:

[0025] Step 3.1, select part of the sea trial data as the online training set , and select part of the sea trial data in the online training set as the online verification set F;

[0026] Step 3.2, select n data samples from the offline training set ; select n data samples from the offline training set ; select 2n data samples from the online training set ;

[0027] The data samples mainly include axial velocity , axial velocity , axial velocity , rudder angle , heading angle ; when each training set moves with the time window, data is continuously supplemented from the database system.

[0028] The above-mentioned online modeling and prediction method for the state information of an unmanned ship based on simulation technology, the specific steps of step 4 are as follows:

[0029] Step 4.1, construct a linear vector according to the input data of the total training set :

[0030] ;

[0031] where is the simulation time, is the time interval, is the number of time delays; , is the dimension of the input data, is the sequential connection of vectors, and the total training set contains ;

[0032] Construct a non-linear vector according to the input data of the total training set :

[0033] ;

[0034] wherein is the order, is an operator for collecting the only monomial in the results of two outer products;

[0035] Construct the total feature vector :

[0036] ;

[0037] wherein is , is the rudder angle, is the strong nonlinear coefficient exhibited in the marine environment, is a hyperparameter; represents the random connection of vectors, demonstrating the randomness of the marine environment;

[0038] The output prediction vector is:

[0039] ;

[0040] wherein, is a combination of one or more of, representing the true output information of the sea trial data; represents the output weight of the reservoir computing network;

[0041] The error MSE is:

[0042] ;

[0043] wherein, represents the number of samples, represents the true value of the sample;

[0044] Step 4.2, before the unmanned ship starts, based on the offline training set and the offline training set complete the offline training of the nonlinear reservoir computing model with the sample data; after the unmanned ship starts, add the online training set samples to the offline training set to continue training, model and predict the state information of the unmanned ship at future moments, and design the sliding of the time window to discard the long - term historical data and select new data.

[0045] For the above - mentioned online modeling and prediction method of the unmanned ship state information based on simulation technology, the discarding of long - term historical data and the selection of new data in the step 4.2 are specifically as follows:

[0046] Set the error threshold , if , without discarding the historical data before the time window position, continue to complete the prediction of the state information at future moments by the non-linear reservoir computing model; if , according to the sliding of the time window, discard the historical data before the time window position, re-select the data after the time window as the training set, add the predicted data at the last position of the time window to the total training set, and retrain the non-linear reservoir computing model by combining the historical information and the current moment information to complete the prediction of the state information at future moments.

[0047] The beneficial effects of the present invention are that, by combining the computational fluid dynamics, the physical simulation environment and the dynamic characteristics of the real ocean environment, the present invention can enable an unmanned ship to realize the online modeling and prediction of state information by using small sample data in the real ocean environment, provide effective prior information for the collision avoidance, planning and control of the unmanned ship, and safely and effectively complete the navigation task. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is the Solidworks model diagram of the embodiment of the present invention;

[0049] Figure 2 is the CFD simulation diagram of the embodiment of the present invention, where (a) represents the overall view in the mesh division, (b) represents the bow view, (c) represents the stern view, and (d) represents the simulation diagram of the model ship running in the flow field;

[0050] Figure 3 is the schematic diagram of the structure of the model propeller of the executable file of the specific model of the embodiment of the present invention;

[0051] Figure 4 is the schematic diagram of the dynamic display effect of the wave plug-in in the embodiment of the present invention, where (a) represents the simulation effect of the OsgOcean rendering library in the prior art UWsim, (b) represents the simulation effect diagram of the prior art plug-in Freefloating, (c) represents the schematic diagram of the hull segmentation in this embodiment; (d) represents the dynamic display effect obtained by the hull segmentation method of (c) in this embodiment;

[0052] Figure 5 is the embodiment of the present invention Schematic diagram of the time window sliding and modeling prediction process;

[0053] Figure 6 is the embodiment of the present invention Schematic diagram of the time window sliding and modeling prediction process. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] This embodiment discloses an online modeling and prediction method for the state information of an unmanned ship based on simulation technology. This embodiment conducts simulations on the unmanned ship model made in Solidworks. The unmanned ship model in this embodiment is named "Jiuhang 750", and star-ccm+ and CFD technology are used for simulations, with a default sampling frequency of 1 Hz. The Solidworks model diagram is as shown in Figure 1 follows, and the specific online modeling and prediction method includes:

[0056] Step 1: Use star-ccm+ to simulate the unmanned ship soildworks model to obtain the navigation data of the unmanned ship in the fluid.

[0057] Step 1 specifically includes:

[0058] Step 1.1: Import the "Jiuhang 750" solidworks model into star-ccm+;

[0059] Step 1.2: Select hexahedral cut cell meshes for the overall mesh division of the computational domain, and use prism layer meshes and surface reconstruction near the hull wall surface;

[0060] Step 1.3: Adopt the model for the turbulence model; in the numerical simulation process, the water density is taken as 1025 kg / m 3 , and the dynamic viscosity of the water is 1.21*10 -3 kg / (m·s);

[0061] Step 1.4: The front boundary of the computational domain is the velocity inlet boundary condition, the rear boundary of the computational domain is the pressure outlet boundary condition, the upper, lower, and both side boundaries of the computational domain are the velocity inlet boundary conditions, and the damping wave elimination method is used for wave elimination at the outlet boundary;

[0062] Step 1.5: After the configuration is completed, conduct CFD simulations. The CFD simulation diagram is as shown in Figure 2 , and collect the motion data as the offline training set .

[0063] Step 2: Build a Noetic marine environment simulator, design the unmanned ship soildworks model and convert it into an executable file. The executable file includes the parent coordinate system, child coordinate system, parent joint, child joint, axis joint, etc. Taking the Figure 3 structural diagram of the thruster as an example, simulate to obtain the navigation data of the unmanned ship in the marine environment.

[0064] Step 2 specifically includes:

[0065] Step 2.1: Set up the Noetic ocean environment simulator, construct the 3D solidworks model file of "Long Endurance 750" and convert it into a model executable file, then import it into the ocean simulation environment.

[0066] The effects of the wave simulation plug-ins in the prior art are as shown in Figure 4 Figures (a) and (b). In (a), the height of the wave is obtained using the fast Fourier transform, but this information cannot be shared with the simulated ocean environment. Therefore, the shape of the wave is ignored and the simulation of the actual situation is restricted. All ships float on a flat water plane (Z = 0). In (b), the plug-in Freefloating can receive the wave height data of the buoyancy center of the unmanned ship and use it as the input for the buoyancy effect. Although this allows the ship to move up and down with the waves, it is only the data of the hull center. So sometimes the front (bow) or rear (stern) of the ship may float outside the water surface or be completely submerged.

[0067] Step 2.2: To solve the problem of poor dynamic motion visualization effects in scenarios (a) and (b) of the wave simulation plug-in, design a hydrodynamic dynamic plug-in based on the hull segmentation method. First, divide the unmanned ship model into six parts, as shown in Figure 4 Figure (c). Specifically: divide the ship into six equal parts, calculate the wave height and buoyancy of the centers of the six parts respectively, and directly apply them to the unmanned ship in the simulation environment, so as to show the change of the attitude of the unmanned ship. While Figure 4 in Figure (b) only calculates the whole hull and does not consider the unity of each part of the hull. Figure 4 Then, based on the current state (position and attitude) of the ship, determine the positions of the hull grid points. At the position

[0068] , calculate the component waves of each part using a simple Gerstner wave and three independent superimposed wave functions at the position and generate the wave height at the position . The specific calculation formula is as follows:

[0069] ;

[0070] where, for each component wave, i is the simulation time, N is the end time, is the steepness, is the amplitude, is the wave vector, is the angular frequency, is the phase. The wave vector is the horizontal vector in the wave propagation direction, and its magnitude is equal to the wave number , where is the wavelength.

[0071] Step 2.3: Determine the water surface displacement at each grid point based on the sum of the component waves. Meanwhile, determine the buoyancy generated based on the position of the ship's grid points relative to the water surface, and directly apply the force conditions to the corresponding grid points. Implement the dynamic change of the unmanned ship's attitude according to the water surface displacement and force conditions of each part, such as shown in (d) of Figure 4 as shown in

[0072] Step 2.4: After completion of the configuration, conduct a simulation and collect the motion data as an offline training set .

[0073] Step 3: Construct a database system for the unmanned ship, including an offline training set, an online training set, and a validation set.

[0074] Step 3 specifically includes:[[]]

[0075] Step 3.1: Select part of the sea trial data of "Jiuhang 750" as the online training set , and select part of the sea trial data from it as the online validation set ;

[0076] Step 3.2: Select 50 data samples from the offline training set ; select 50 data samples from the offline training set ; select 100 data samples from the online training set ; the data samples mainly include axial velocity , axial velocity , axial velocity , rudder angle , heading angle and other data. When each training set moves with the time window, continuously supplement data from the database system.

[0077] Step 4: Design a nonlinear reservoir algorithm based on the sliding window method to achieve online modeling and prediction of the state information of the unmanned ship.

[0078] Step 4 specifically includes:[[]]

[0079] Step 4.1: Construct a linear vector from the input data of the total training set :

[0080] ;

[0081] where is the simulation time,[[]] is the time interval,[[]] is the number of time delays; , is the dimension of the input data, is the sequential connection of vectors, the total training set contains ;

[0082] According to the total training set construct a non - linear vector from the input data :

[0083] ;

[0084] where is the order, is the operator that collects the unique monomials in the result of the outer product of two terms;

[0085] Construct the total feature vector :

[0086] ;

[0087] where is , is the rudder angle, is the strong non - linear coefficient exhibited in the ocean environment, is the hyper - parameter; represents the random connection of vectors, demonstrating the randomness of the ocean environment;

[0088] The output prediction vector is:

[0089] ;

[0090] where, is a combination of one or more of, representing the true output information of the sea trial data; represents the output weight of the reservoir computing network;

[0091] The error MSE is:

[0092] ;

[0093] where, represents the number of samples, represents the true value of the sample.

[0094] Step 4.2, Before the unmanned ship starts, complete the offline training of the non - linear reservoir computing model based on 100 sample data of the offline training set. After the unmanned ship starts, add 100 sample data after startup to the training set and continue training. Then, start modeling and forecasting the state information of the unmanned ship at future moments. Design the sliding of the time window to discard the long - ago historical data and select new data.

[0095] Step 4.3, set the error threshold , if , do not discard the historical data before the time window position, and continue to complete the prediction of the state information at future moments by the non-linear reservoir computing model. The time window sliding and modeling prediction process is as Figure 5 shown.

[0096] Step 4.4, if , according to the sliding of the time window, discard the historical data before the time window position, and re-select the data after the time window as the training set. Add the prediction data at the last position of the time window to the training set, and retrain the non-linear reservoir computing model by combining the historical information and the current moment information to complete the prediction of the state information at future moments. The time window sliding and modeling prediction process is as Figure 6 shown.

[0097] The above embodiments are only exemplary embodiments of the present invention and are not used to limit the present invention. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present invention, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present invention.

Claims

1. An online modeling and prediction method for unmanned ship status information based on simulation technology, characterized in that: The steps include: Step 1, simulating the unmanned ship model to obtain the navigation data of the unmanned ship in the fluid; Step 2: Build a marine environment simulator, simulate the unmanned ship model, and obtain the navigation data of the unmanned ship in the marine environment; Step 3, using the navigation data collected in steps 1 and 2 as an offline training set, and using the sea trial data of the unmanned ship as an online training set and an online verification set to build an unmanned ship database; Step 4, design a nonlinear reserve pool algorithm based on the sliding window method to realize online modeling and prediction of the state information of the unmanned ship; The step 4 specifically includes: Step 4.1, based on the total training set Input data to construct linear vector in, It's simulation time. is the time interval, is the number of time delays; , is the dimension of the input data, is the sequential connection of the vectors, the total training set Include ; According to the total training set Input data to construct nonlinear vector in is the order, is an operator that collects the unique monomials in the result of the outer product of two terms; Constructing the total feature vector in yes , is the rudder angle, is the strong nonlinear coefficient in the marine environment, is a hyperparameter; Represents the random connection of vectors, showing the randomness of the ocean environment; The output prediction vector is in, yes A combination of one or more of represents the real output information of the sea trial data; Represents the output weight of the reserve pool calculation network; The error MSE is ； in, represents the number of samples, represents the true value of the sample; Step 4.2: Before the unmanned ship starts, based on the offline training set and offline training set The sample data of the unmanned ship is used to complete the offline training of the nonlinear reserve pool calculation model; after the unmanned ship is started, the online training set samples are added to the offline training set to continue the training, and the state information of the unmanned ship at the future moment is modeled and predicted. The sliding of the time window is designed to complete the discarding of long-term historical data and the selection of new data; The specific steps of discarding long-term historical data and selecting new data in step 4.2 are as follows: Setting the error threshold ,like , without discarding the historical data before the time window position, continue to complete the nonlinear reserve pool calculation model to predict the state information at future times; if According to the sliding of the time window, the historical data before the time window position is discarded, and the data after the time window is reselected as the training set. The forecast data at the last position of the time window is added to the total training set. The nonlinear reserve pool calculation model is retrained in combination with the historical information and the current moment information to complete the forecast of the state information at the future moment.

2. The method for online modeling and prediction of unmanned ship status information based on simulation technology according to claim 1 is characterized in that: The step 1 specifically includes: Step 1.1, import the unmanned ship model into star-ccm+; Step 1.2, the whole computational domain is meshed using a hexagonal cut volume mesh, and near the hull wall, a prismatic layer mesh and surface reconstruction are used; Step 1.3, the turbulence model is adopted Model; Step 1.4, the front boundary of the computational domain is the velocity inlet boundary condition, the rear boundary of the computational domain is the pressure outlet boundary condition, the upper, lower and both sides of the computational domain are the velocity inlet boundary conditions, and the damping wave elimination method is used to eliminate waves at the outlet boundary; Step 1.5, perform CFD simulation and collect motion data as offline training set O1.

3. The method for online modeling and prediction of unmanned ship status information based on simulation technology according to claim 1 is characterized in that: The step 2 specifically includes: Step 2.1, build the Noetic ocean environment simulator and import the unmanned ship model into the ocean simulation environment; Step 2.2, designing a hydrodynamic dynamic plug-in based on the hull segmentation method; Step 2.3, determine the water surface displacement at each grid point based on the sum of the component waves, determine the buoyancy generated based on the position of the ship grid point relative to the water surface, apply the force directly to the corresponding grid point, and realize the dynamic change of the unmanned ship's posture according to the water surface displacement and force of each part; In step 2.4, motion data is collected as an offline training set O2.

4. The method for online modeling and prediction of unmanned ship status information based on simulation technology according to claim 3 is characterized in that: The step 2.2 specifically includes: dividing the unmanned ship model into six parts; determining the position of the hull grid points based on the current position and attitude of the ship, The component waves of each part are calculated using Gerstner waves and three independent superposition wave functions, and Generated wave height , the specific calculation formula is: where, for each component wave, is the simulation time, N is the end time, is the steepness, is the amplitude, is the wave vector, is the angular frequency, is the phase; the wave vector is the horizontal vector in the direction of wave propagation, and its amplitude is equal to the wave number ,in is the wavelength.

5. The method for online modeling and prediction of unmanned ship status information based on simulation technology according to claim 3 is characterized in that: The step 3 specifically includes: Step 3.1: Select part of the sea trial data as the online training set , in the online training set Select part of the sea trial data as the online verification set F; Step 3.2, from the offline training set Select n data samples from the offline training set Select n data samples from the online training set Select 2n data samples from ; The data sample mainly includes Axis speed , Axis speed , Axis speed , rudder angle , heading angle ; Data is continuously supplemented from the database system as each training set moves along the time window.

Images