Digital twin method for predicting navigation attitude of unmanned ship

Abstract

The invention belongs to the technical field of unmanned ship state prediction. The invention provides a digital twin method for predicting the navigation attitude of an unmanned ship. According to the embodiment of the invention, a highly refined twinborn prediction model is established through a digital twinborn technology, and the motion characteristics and stress conditions of the unmanned ship in a complex marine environment are accurately reflected. And dynamically adjusting and updating the constructed twin prediction model by acquiring sensor fusion data of a real unmanned ship in real time in combination with an improved unscented Kalman filtering algorithm to obtain a highly real unmanned ship model. The identified twin model is used for predicting the navigation state of the unmanned ship in a certain time in the future, support is provided for intelligent navigation decision making of the unmanned ship, and the operation efficiency and safety of the unmanned ship are improved.

Description

Technical Field

[0001] The disclosed embodiments relate to the technical field of unmanned boat state prediction, and in particular to a digital twin method for unmanned boat navigation posture prediction. Background Art

[0002] With the continuous advancement of science and technology, unmanned surface vehicles (USVs) have been widely used in various fields such as ocean exploration, environmental monitoring, shipping and logistics. The ability of USVs to navigate autonomously is crucial to navigation safety and mission completion efficiency, and navigation attitude prediction is one of the key technologies to ensure the safe and stable operation of USVs. The navigation attitude of USVs includes important parameters such as position, speed, and heading angle. These parameters need to be accurately predicted to perform navigation control tasks and ensure the stability and safety of USVs in complex marine environments. Traditional navigation attitude prediction methods are mostly based on model calculations, but in the face of various uncertainties in complex environments, existing methods have significant deficiencies in accuracy and robustness.

[0003] Technical solutions of existing technology: 1. Mathematical modeling: This approach uses dynamic and kinematic models of the unmanned vehicle to predict its navigational state. However, this approach typically requires high modeling standards and struggles to cope with external uncertainties and dynamic changes in the face of environmental disturbances and complex sea conditions, resulting in large prediction errors and poor accuracy.

[0004] 2. Single data source: Using a single data source (such as GPS or IMU) to predict the navigation attitude of an unmanned vehicle has numerous drawbacks. GPS updates infrequently in dynamic environments, and its accuracy decreases under conditions of signal obstruction or significant interference, making it unable to provide high-frequency, real-time attitude change information. While IMUs can provide high-frequency attitude change data, they suffer from cumulative errors and drift, leading to inaccurate predictions over time. Relying solely on either data source is prone to inaccuracy, error accumulation, poor real-time performance, and poor adaptability to environmental changes.

[0005] 3. Machine Learning: With the development of artificial intelligence (AI), some research has attempted to use machine learning or deep learning methods to train models using large amounts of historical data to predict the navigation posture of unmanned vehicles. However, machine learning methods typically rely on large amounts of high-quality historical data, lack interpretability, and are difficult to adjust models to new environments in real time. Furthermore, they are computationally expensive.

[0006] In practical applications, existing technical solutions have problems such as insufficient accuracy, difficulty in real-time model updating, poor interference and noise processing capabilities, lack of robustness and adaptability, etc., making it difficult to achieve efficient and accurate navigation attitude prediction in complex and dynamic ocean environments.

[0007] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0008] It should be noted that this section is intended to provide background or context for the technical solutions of the present disclosure stated in the claims. The description herein is not admitted to be prior art by virtue of being included in this section. Summary of the Invention

[0009] The purpose of the embodiments of the present disclosure is to provide a digital twin method for predicting the navigation posture of an unmanned boat, thereby overcoming one or more problems caused by the limitations and defects of related technologies to at least a certain extent.

[0010] According to a first aspect of an embodiment of the present disclosure, a digital twin method for predicting the navigation posture of an unmanned boat is provided, the method comprising: Build a twin prediction model based on the unmanned boat's kinematic model, dynamic model, and environmental disturbance data; Acquire GPS data and IMU data in real time, and use Kalman filter fusion technology to fuse the data to obtain real-time observation data; Use Kalman filter fusion technology to fuse real-time observation data and twin historical data to obtain fused state data; The fusion state data is dynamically extracted through a sliding window and processed segment by segment, and the state variables and model parameters of the unmanned boat are recursively calculated to optimize the twin prediction model; When the sliding window reaches the data boundary, the optimized twin prediction model is obtained, and the navigation posture of the unmanned boat is predicted using the optimized digital twin model.

[0011] Furthermore, based on the kinematic model, dynamic model and environmental disturbance data of the unmanned boat, the steps of constructing a twin prediction model include: Based on the kinematic model, the mathematical model of the unmanned boat is constructed by combining its kinematic model, control model and environmental disturbance data; The viscosity coefficient and propeller thrust coefficient are taken as parameter vectors to be optimized, and the mathematical model of the unmanned vehicle is differentially discretized to construct a twin prediction model.

[0012] Furthermore, the steps of acquiring GPS data and IMU data in real time and fusing the data using Kalman filter fusion technology to obtain sensor fusion data include: Acquire GPS data and IMU data in real time and perform preprocessing; Use Kalman filter to perform state estimation and prediction on the preprocessed GPS data and IMU data to obtain GPS observation data and IMU observation data; The Kalman gains of GPS observation data and IMU observation data are weighted, and the state prediction generated by the fusion is calculated to obtain real-time observation data.

[0013] Furthermore, the fusion state data is dynamically extracted through the sliding window and processed segment by segment, and the state variables and model parameters of the unmanned boat are recursively calculated to optimize the steps of the twin prediction model, including: Introducing a dynamic sliding window to process the fusion state data segment by segment; The parameter vector of the twin prediction model is used as the extended state vector to construct a three-degree-of-freedom state equation containing kinematic parameters and parameters to be identified; The three-degree-of-freedom unmanned boat motion twin model is used to recursively calculate the twin historical data and real-time observation data in the dynamic sliding window, estimate the state variables of the unmanned boat and update the model parameters in real time to optimize the twin prediction model.

[0014] Furthermore, the total length of the sliding window consists of a basic window and an adaptive adjustment window. The basic window length is determined based on the dynamic response time constant of the unmanned boat, and the adaptive adjustment window length is dynamically adjusted according to the posture data characteristics and prediction error. Adopting a periodic adaptive adjustment strategy, combined with the wave spectrum main period and UKF information covariance feedback, the sliding window length is adjusted in real time; When the environmental disturbance parameter exceeds the preset threshold, the window length is forcibly extended to enhance the robustness of the model.

[0015] Furthermore, when the sliding window reaches the data boundary, the optimized twin prediction model is obtained. The steps of using the optimized digital twin model to predict the navigation posture of the unmanned boat include: When the sliding window reaches the data boundary, the final identification result is output, the twin model parameters are updated, and the optimized twin prediction model is obtained; By using the optimized digital twin model, combining real-time observation data with twin historical data, the future navigation status of the unmanned boat is predicted through the rolling time window technology to predict the navigation posture of the unmanned boat.

[0016] Furthermore, the twin model parameters include: State matrix, state transfer matrix, observation matrix, process noise covariance matrix and measurement noise covariance matrix.

[0017] The technical solutions provided by the embodiments of the present disclosure may have the following beneficial effects: In the embodiment of the present disclosure, through the above-mentioned digital twin method for predicting the navigation posture of an unmanned boat, first, a highly refined twin prediction model is established through digital twin technology to accurately reflect the motion characteristics and force conditions of the unmanned boat in a complex marine environment. Secondly, the constructed twin prediction model is dynamically adjusted and updated by obtaining the sensor fusion data of the real unmanned boat in real time and combining it with the improved unscented Kalman filter algorithm to obtain a highly realistic unmanned boat model. Finally, the identified twin model is used to predict the navigation status of the unmanned boat within a certain period of time in the future, to provide support for the intelligent navigation decision-making of the unmanned boat, and to improve the operating efficiency and safety of the unmanned boat. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0019] Figure 1 A step diagram showing a digital twin method for predicting the navigation posture of an unmanned boat in an exemplary embodiment of the present disclosure is shown; Figure 2 A specific flow chart of a digital twin method for predicting the navigation posture of an unmanned boat in an exemplary embodiment of the present disclosure is shown; Figure 3 A schematic diagram of a twin prediction model architecture in an exemplary embodiment of the present disclosure is shown; Figure 4 A schematic diagram of a motion coordinate system of an unmanned boat in an exemplary embodiment of the present disclosure is shown; Figure 5 A schematic diagram of a three-dimensional geometric model of a twin unmanned boat in an exemplary embodiment of the present disclosure is shown; Figure 6 A flow chart showing the principle of sensor fusion in an exemplary embodiment of the present disclosure is shown; Figure 7 A flow chart of twin model parameter identification in an exemplary embodiment of the present disclosure is shown; Figure 8 A schematic diagram showing an offshore test environment in an exemplary embodiment of the present disclosure; Figure 9 A schematic diagram showing a process of sailing an unmanned boat at sea in an exemplary embodiment of the present disclosure; Figure 10 The 20° right turn test in the wind and wave environment in the exemplary embodiment of the present disclosure is shown. Parameter identification process; Figure 11 The 20° right turn test in the wind and wave environment in the exemplary embodiment of the present disclosure is shown. Parameter identification process; Figure 12 The 20° right turn test in the wind and wave environment in the exemplary embodiment of the present disclosure is shown. Parameter identification process; Figure 13 A motion prediction diagram of a 20° right turn test in a wind and wave environment in an exemplary embodiment of the present disclosure is shown; Figure 14 Shows the Z-type maneuvering test in a wind and wave environment of 20° / -20° in an exemplary embodiment of the present disclosure Parameter identification process; Figure 15 Shows the Z-type maneuvering test in a wind and wave environment of 20° / -20° in an exemplary embodiment of the present disclosure Parameter identification process; Figure 16 Shows the Z-type maneuvering test in a wind and wave environment of 20° / -20° in an exemplary embodiment of the present disclosure Parameter identification process; Figure 17 A motion prediction diagram of a 20° / -20° Z-type maneuvering test in a wind and wave environment in an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0020] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0021] In addition, the accompanying drawings are merely schematic illustrations of embodiments of the present disclosure and are not necessarily drawn to scale. Like reference numerals in the figures represent like or similar parts, and thus repeated descriptions thereof will be omitted. Some of the blocks shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically separate entities.

[0022] This example embodiment provides a digital twin method for predicting the navigation posture of an unmanned boat. Figure 1 As shown in , the digital twin method for unmanned boat navigation posture prediction may include: Step S101: Building a twin prediction model based on the kinematic model, dynamic model and environmental disturbance data of the unmanned boat; Step S102: acquiring GPS data and IMU data in real time, and fusing the data using Kalman filter fusion technology to obtain real-time observation data; Step S103: Using Kalman filter fusion technology to fuse the real-time observation data and the twin historical data to obtain fused state data; Step S104: Dynamically extract the fusion state data through the sliding window, process it section by section, and recursively calculate the state variables and model parameters of the unmanned boat to optimize the twin prediction model; Step S105: When the sliding window reaches the data boundary, the optimized twin prediction model is obtained, and the navigation posture of the unmanned boat is predicted using the optimized digital twin model.

[0023] The digital twin method for predicting the navigation posture of an unmanned vehicle (UV) first establishes a highly refined twin prediction model using digital twin technology, accurately reflecting the motion characteristics and stress conditions of the UV in complex marine environments. Secondly, by acquiring real-time sensor fusion data from a real UV, combined with an improved unscented Kalman filter algorithm, the constructed twin prediction model is dynamically adjusted and updated to produce a highly realistic UV model. Finally, the identified twin model is used to predict the UV's navigation status within a certain timeframe, providing support for intelligent navigation decision-making and improving its operational efficiency and safety.

[0024] Below, we will refer to Figures 1 to 17 Each step of the digital twin method for predicting the navigation posture of an unmanned boat in this example embodiment is described in more detail.

[0025] In step S101, Figure 2 The figure shows a specific flow chart of the digital twin method for predicting the navigation posture of the unmanned boat. First, a digital twin model of the unmanned boat is established.

[0026] Based on the kinematic and dynamic models of the UAV, a fully parameterized digital twin model is constructed to simulate the UAV's motion and interaction with the external environment. Based on the UAV's dynamic model, the fully parameterized digital twin model is combined with its kinematic, dynamic, and control models, as well as environmental disturbances such as wind and waves. This model can reflect the UAV's dynamic state in real time, enabling accurate state prediction.

[0027] This application selects the mechanism model as the basis, considering the longitudinal, heaving and bow pitch motions of the unmanned boat, and the constructed twin model is as follows:

[0028] The dimensioned viscosity coefficient and propeller thrust coefficient are selected as identification targets and constitute the parameter vector to be optimized. , such as the formula:

[0029] definition As input matrix:

[0030] Constructing a twin prediction model based on first-order differences:

[0031] in 、 、 、 、 、 is the navigation state of the unmanned boat at time k from sensor fusion, and the coefficient is the parameter optimized at time k. The prediction model of the future rudder angle is:

[0032] like Figure 3 As shown in the figure, it is a schematic diagram of the twin prediction model architecture. Figure 4 As shown in the figure, it is a schematic diagram of the motion coordinate system of the unmanned boat. Figure 5 Shown is the three-dimensional geometric model of the twin unmanned boat.

[0033] In steps S102 and S103 , sensor data (ie, real-time observation data) are fused to perform state estimation.

[0034] This application uses Kalman filter fusion technology to fuse data from multiple sensors such as GPS and IMU. The fusion principle is as follows: Figure 6 As shown in the figure, GPS provides global position information, and IMU provides high-frequency local acceleration and angular velocity information. The Kalman filter optimizes these sensor data to obtain high-precision state estimation, improving the prediction accuracy of navigation attitude.

[0035] The state prediction produced by the sensor fusion method is calculated by weighting the Kalman gains of the two models as follows:

[0036] in, and are the fusion weights, which correspond to the weighting coefficients of GPS and IMU (Inertial Measurement Unit) data. These weights are used to calculate the fusion result of the data provided by the two sensors; and is the covariance matrix of IMU and GPS, which represents the uncertainty of the data of each sensor. A larger covariance value means that the measurement accuracy of the sensor is higher. and are the state estimates obtained by GPS and IMU respectively, representing the estimated value of the target state (such as position, velocity, acceleration, etc.) of each sensor at time k+1; It is the final fused state estimate, which combines the information provided by GPS and IMU to obtain a more accurate state estimate.

[0037] In step S104, the improved UKF algorithm is applied to perform dynamic optimization.

[0038] An improved UKF algorithm, combined with a dynamic sliding window mechanism, updates the parameters of the digital twin model in real time, enabling efficient estimation and correction of unknown disturbances. The specific steps include: using the improved UKF algorithm, combined with dynamic sliding window technology, to collect state data, control instructions, and sensor measurements. Based on the system model and observation data, the system recursively calculates and estimates the UAV's motion state and model parameters (such as the noise covariance matrix) in real time, enabling real-time parameter identification. The UKF optimizes the model based on the state prediction and actual observations at each moment, making the UAV's navigation attitude prediction more accurate in dynamic environments.

[0039] This application defines the total length of the dynamic sliding window By the basic window With adaptive adjustment window constitute:

[0040] in, The length of the fixed part is determined according to the dynamic response time constant of the unmanned boat (such as Corresponding to 3 seconds of historical data, sampling frequency 5Hz); To dynamically adjust the length of the part, its size is determined by the characteristics of the posture data and the fluctuation of the prediction error.

[0041] This application uses a periodic adaptive adjustment strategy to dynamically adjust the sliding window size , the specific adjustment rules are as follows: Set the adjustment period, and determine the basic period based on the main period of the wave spectrum (3-15 seconds) and the control response time of the unmanned boat. , and then according to the UKF information covariance Real-time feedback for adjustments:

[0042] where is the upper bound of the information covariance (like Corresponding to the 3σ boundary), the adjustment period is shortened when the filtering error increases.

[0043] Calculate the mean fluctuation of ocean environmental parameters within the window (Fusion wave height , wind speed ):

[0044] The Lyapunov exponent is used to evaluate the nonlinearity of the motion state, and the small data method (optimized for short time series) is used to calculate the maximum Lyapunov exponent:

[0045] The nonlinearity index within the window is calculated by the formula:

[0046] Reflecting the divergent trend of the data, It indicates that the local chaotic characteristics are enhanced.

[0047] Then perform fusion weighting on the model:

[0048] Based on the Zhuhai sea trial data (wave height 1-4), a multivariate regression analysis was conducted to determine , , reflecting the dominant characteristics of environmental disturbances. When the IMU lateral acceleration When forced , the window length is doubled to , reflecting the dominant characteristics of environmental disturbance. and , set , saving computing power.

[0049] Remove old data at each sampling moment , new real-time data ,when When reducing, the redundant data at the end of the window is removed synchronously to maintain the total length , adapt to dynamic system changes.

[0050] The parameter vector of the twin model is regarded as a state, and the state vector and parameter vector are estimated at the same time. According to the twin prediction model formula, the three-degree-of-freedom unmanned boat motion twin model equation based on unscented Kalman filtering is constructed as follows:

[0051] State vector Describes the dynamic state of the unmanned boat, which includes the kinematic parameters of the unmanned boat (longitudinal velocity , lateral velocity , bow angular velocity , the true north coordinate in the geodetic coordinate system , the east coordinate in the geodetic coordinate system , heading angle By combining this information with information about the parameters to be identified (dimensional viscous force coefficients and propeller thrust coefficients), UKF can make a comprehensive estimate of the state and adjust these estimates based on sensor data.

[0052]

[0053] Input Control is the vector of the left and right propeller speeds, with the initial rudder angle being 0°:

[0054] The observation vector of the system Usually contains the actual measurement value obtained by the sensor, in this application yes The measurement values ​​of the first six dimensions, that is, the fusion data of sensor data and twin data, are as follows:

[0055] Initial value of the state matrix:

[0056] State transition matrix:

[0057] Observation matrix:

[0058] Process noise covariance matrix :

[0059] Measurement noise covariance matrix :

[0060] The iterative optimization process of the twin model based on the dynamic sliding window improved unscented Kalman filter (DSW-UKF) is as follows: Figure 7As shown in the figure. First, a sliding window is used to dynamically extract the twin data from the system and the actual boat data. The USV data, after fusion using a Kalman filter, includes state variables (such as position and velocity), inputs (control instructions), and observations (sensor measurements). This data is passed to the filter step by step through the sliding window for processing. The unscented Kalman filter uses the system model and measurement data to perform real-time estimation of state variables and model parameters (such as the noise covariance matrix) through recursive calculation. When the sliding window reaches the data boundary, the final identification result is output, thus updating the twin model parameters.

[0061] In step S105, the navigation attitude is predicted.

[0062] The digital twin model optimized through UKF is used to perform rolling predictions of future navigation states. Using sliding time window technology, combined with real-time sensor data and fused data from the twin unmanned boat, the future navigation state of the unmanned boat (such as position, speed, heading angle, etc.) is predicted.

[0063] In a specific embodiment, accurate prediction of the navigation attitude of an unmanned boat is crucial for improving navigation efficiency and safety. However, accurate prediction in actual operation is challenging. In practical applications, data is often affected by noise pollution and the navigation environment. To evaluate the effectiveness of the proposed navigation state prediction method based on improved unscented Kalman filtering and digital twins, this embodiment uses the method of this application combined with measurement data from actual ship experiments to iteratively optimize the unmanned boat motion twin model, and provides the process of iterative optimization of model parameters. The optimized twin model is then used to predict the state of the unmanned boat.

[0064] In this example, a twin model optimized with the unmodified UKF identification algorithm and a commonly used black-box model prediction method constructed with LSTM were selected as comparison models for predicting the navigation state of unmanned vehicles. The test was conducted in complex sea conditions to simulate the challenges faced by unmanned vehicles in real-world environments.

[0065] To ensure a reliable comparison, this embodiment conducts a simulation test on the modeling of an unmanned boat in a university's digital twin laboratory. The detailed data of the unmanned boat is shown in Table 1.

[0066] Table 1 Main parameters of the unmanned boat hull

[0067] In order to obtain the observation data sequence required for the twin unmanned boat parameter identification, an unmanned boat real ship test was carried out in a sea area of ​​Zhuhai. The offshore test environment is as follows: Figure 8 As shown, Figure 9This is a video of an unmanned boat navigating at sea. The boat is equipped with GPS sensors and an IMU (Inertial Measurement Unit) that provide real-time position, velocity, and attitude information. The target boat is controlled by a ground station on shore, with control signals transmitted via Wi-Fi to the onboard motor drivers to control the speed of the two propellers. Multiple sets of target boat turning maneuvering tests were conducted, followed by multiple sets of target boat Z-turn maneuvering tests. After filtering and fusion, the multi-sensor observation data was input into the unmanned boat's twin model in real time for online parameter identification.

[0068] In order to realize the online identification of the twin model, the frequency of the real unmanned boat sending messages is set to 5Hz by using the ground station control command, and the sensor data of the real unmanned boat is obtained and fused in real time.

[0069] 1.2 Scenario 1: Real-time state prediction of a 20° right turn test in windy and wavey conditions The 20° starboard turn maneuvering test of the unmanned vehicle was conducted under sea conditions of Category 2, with a wave height of approximately 0.5m. The wind speed was approximately 4.5m / s from the southeast. During the test, the environment changed continuously over time. The unmanned vehicle maintained a 20° starboard rudder angle at rated speed to test its maneuverability and dynamic response. To ensure the reliability of the results, this test was repeated four times. Throughout the test, the sensor sampling frequency was set to 5Hz to collect sufficient dynamic data to evaluate the unmanned vehicle's steering performance and stability.

[0070] The identification process of each parameter of the twin model of the unmanned boat in this experiment is as follows: Figures 10 to 12 As shown. Among them, Figure 10 The identification process of longitudinal dynamic parameters is shown. Figure 11 It shows the identification process of lateral dynamic parameters. Figure 12 The figure shows the identification process of yaw dynamic parameters. The dynamic sliding window based unscented Kalman filter (DSW-UKF) algorithm shows fast identification and stable convergence capabilities during the parameter identification process. In the process of longitudinal dynamic parameter identification, each parameter basically reaches the convergence state after the 90th identification, such as 、 ,and After the initial fluctuations, it quickly stabilizes. The convergence effect is significant, and the thruster related parameters ( 、 ) remains constant, showing strong resistance to wind and wave interference. The lateral dynamic parameters affect the lateral motion of the USV, which is greatly affected by wind and waves. and The initial variation is large, but it gradually converges after multiple identifications, showing the adaptability of the algorithm. The overall performance of the yaw dynamics parameters is stable, and the nonlinear parameters High precision, linear parameters Although there are initial fluctuations, it eventually stabilizes. Based on the above analysis, the adaptive ability of the twin model gradually increases during the recognition process, and the fluctuations during the recognition process eventually converge, indicating that the model has good adaptability to dynamic changes.

[0071] After 210 identifications, all parameters basically reached a convergence state. At this time, the twin model was in a stable state. The experiment selected the results of the 210th parameter identification (the parameter identification values ​​are shown in Table 2) and substituted them into the twin prediction model. The navigation attitude of the USV was predicted by combining the real data and the online fusion data of the twin data.

[0072] Table 2 210th identification values ​​of various parameters in the 20° right turn test in wind and wave environment

[0073] Factors such as wind and waves in the ocean environment will change over time, and long-term prediction accuracy is difficult to guarantee. Therefore, 10 seconds is selected as a reasonable prediction window to reduce the impact of environmental uncertainty. The prediction results for the next 10 seconds of a 20° right turn maneuver in a windy and wavey environment are as follows: Figure 13 As shown in the figure, different prediction methods exhibit different performance in this prediction result. DT-Model (DSW-UKF), the twin model iteratively optimized by the DSW-UKF algorithm, performs best in all prediction parameters and can accurately reflect the motion state of the unmanned boat. In particular, it demonstrates high real-time performance and accuracy in handling time-varying and uncertainties.

[0074] In the posture prediction of a 20° right turn, the proposed method (DT-Model (DSW-UKF)) performed superiorly in most cases, accurately capturing the unmanned vehicle's motion characteristics, including longitudinal and lateral velocities, pitch velocity, heading angle, and position changes. Compared with the DT-Model (UKF) and LSTM black-box models, the proposed method was closer to the true values ​​in all predictions, especially in the eastward position and pitch velocity predictions, where it achieved higher accuracy. However, the DT-Model (UKF) and LSTM black-box models exhibited significant prediction bias in certain situations, with the LSTM model exhibiting significant errors in the predictions of lateral velocity, heading angle, and position. Overall, the DT-Model (DSW-UKF) demonstrated strong adaptability and high prediction accuracy in handling complex motions.

[0075] 1.3 Scenario 2: Real-time state prediction of 20° / -20° Z-type maneuvering test in windy and wave environment The UAV's 20° / -20° Z-shaped maneuvering test was conducted under the same sea conditions. At rated speed, the vehicle was driven at a 20° rudder angle for a period of time, then again at a -20° rudder angle for the same period of time. To avoid haphazard behavior, the test was repeated four times. The sensor sampling frequency was 5 Hz. Data from the 0th to 80th second of the test was selected for parameter identification of the UAV's twin model.

[0076] The identification process of each parameter of the twin model of the unmanned boat in this experiment is as follows: Figure 14 、 15 and 16, where Figure 14 The identification process of longitudinal dynamic parameters is shown. Figure 15 It shows the identification process of lateral dynamic parameters. Figure 16 The identification process of yaw dynamic parameters is shown. In the 20° / -20° Z-type maneuvering test of the unmanned boat in wind and wave environment, the longitudinal dynamic parameters (such as and )) exhibited significant fluctuations due to frequent acceleration and deceleration and wave disturbances, but ultimately converged quickly, demonstrating that the model effectively captures the dynamic balance between thrust and drag. However, the lateral parameters, affected by sideslip and wind and wave thrust, remained dynamic and converged slowly. The yaw parameters also converged slowly due to sharp turns and nonlinear steering effects. Overall, the process twin model dynamically adjusted to environmental changes, demonstrating its adaptability to complex motions.

[0077] Compared to the right turn experiment, the Z-turn maneuver requires more lateral and yaw response, but the algorithm still demonstrates strong robustness and accuracy. The results of both experiments demonstrate that the algorithm is applicable to different maneuvering scenarios.

[0078] In the experiment, the results of the 210th parameter identification (the parameter identification values ​​are shown in Table 3) were substituted into the twin prediction model to predict the navigation posture of the USV.

[0079] Table 3 210th identification values ​​of various parameters in the Z-type maneuvering test in wind and wave environment at 20° / -20°

[0080] The prediction results for the next 10 seconds of 20° / -20° Z-type maneuvering in wind and wave environment are as follows: Figure 17 As shown, the DT-Model (DSW-UKF) demonstrated high prediction accuracy during this motion process, accurately tracking trends in longitudinal velocity, lateral velocity, heading angle, yaw angle, and position. The DT-Model (UKF) also performed relatively close to the real data in most cases, but its accuracy was slightly lower than that of the improved DSW-UKF model. While the LSTM model was able to capture some key trends, it exhibited some deviations in accuracy and detail, particularly in late-stage predictions.

[0081] In general, DT-Model (DSW-UKF) outperforms the other two methods in terms of accuracy and response speed, and can provide more accurate prediction results in complex wind and wave environments.

[0082] In order to evaluate the performance difference between the prediction method proposed in this application and the other two prediction methods, the application introduces the mean absolute error (MAE) and mean square error (MSE) to judge the accuracy of the three methods in predicting the navigation state of the unmanned boat. The smaller the value of MAE and MSE, the higher the prediction accuracy. The expressions of MAE and MSE are: (1) (2) In the formula, m represents the total number of test samples, Indicates the The predicted value of the test sample, Indicates the The actual value of the test samples.

[0083] According to formulas (1) and (2), the mean absolute error and root mean square error of the future navigation state of the unmanned boat predicted by the three prediction methods are calculated, as shown in Table 4.

[0084] Table 4 Performance analysis of the prediction results of the unmanned boat's navigation posture

[0085] According to the analysis results in Table 4, the DT-Model (DSW-UKF) achieved the best performance in the "20° right turn in a windy and wavey environment" prediction, achieving the lowest MAE (0.1592) and MSE (0.2292). This demonstrates that the model, through dynamic parameter calibration and noise optimization, effectively reproduces the navigational state of both maneuvers in a windy and wavey environment. In the "20° / -20° Z-type maneuvering test in a windy and wavey environment," the MSE of the DT-Model (DSW-UKF) dropped to 0.1837. Although the MAE slightly increased to 0.1733, it still outperformed the DT-Model (UKF) and LSTM. The LSTM had the largest error across all scenarios, with a MAE of 0.2792, indicating poor prediction performance for both maneuvers in a windy and wavey environment. In contrast, the digital twin model balances accuracy and robustness by combining real-time sensor data with a mechanistic model. Through recursive updates and a dynamic sliding window mechanism, DSW-UKF can track uncertain parameters in real time, while efficiently responding to sudden disturbances through adaptive noise estimation and feedback mechanisms, providing reliable support for the autonomous decision-making of the unmanned boat.

[0086] In summary, the LSTM black box model mainly uses training data and simulation results to perform motion prediction, and can handle prediction tasks in different scenarios. However, in dynamic and complex environments, such as Figure 17 The prediction of r in the 84-second interval shows a significant error when the unmanned vehicle rapidly turns. Due to its lack of dynamic sliding window support, the DT-Model (UKF) is limited in its ability to cope with rapid changes and large amounts of data, and its prediction accuracy needs to be improved. The DT-Model (DSW-UKF) can dynamically adjust parameters based on recent data, adapting to time-varying characteristics and effectively capturing the unmanned vehicle's motion state. It achieves the best prediction results, especially when dealing with complex dynamic environments, with greater stability and accuracy.

[0087] The digital twin method for predicting the navigation posture of an unmanned vehicle (UV) first establishes a highly refined twin prediction model using digital twin technology, accurately reflecting the motion characteristics and stress conditions of the UV in complex marine environments. Secondly, by acquiring real-time sensor fusion data from a real UV, combined with an improved unscented Kalman filter algorithm, the constructed twin prediction model is dynamically adjusted and updated to produce a highly realistic UV model. Finally, the identified twin model is used to predict the UV's navigation status within a certain timeframe, providing support for intelligent navigation decision-making and improving its operational efficiency and safety.

[0088] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0089] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the appended claims.

Claims

1. A digital twin method for unmanned boat navigation posture prediction, characterized in that: The method includes: Build a twin prediction model based on the unmanned boat's kinematic model, dynamic model, and environmental disturbance data; Acquire GPS data and IMU data in real time, and use Kalman filter fusion technology to fuse the data to obtain real-time observation data; Use Kalman filter fusion technology to fuse real-time observation data and twin historical data to obtain fused state data; The fusion state data is dynamically extracted through a sliding window and processed segment by segment, and the state variables and model parameters of the unmanned boat are recursively calculated to optimize the twin prediction model; When the sliding window reaches the data boundary, the optimized twin prediction model is obtained, and the navigation posture of the unmanned boat is predicted using the optimized digital twin model.

2. The digital twin method for unmanned vehicle navigation posture prediction according to claim 1 is characterized in that: The steps to build a twin prediction model based on the unmanned boat's kinematic model, dynamic model, and environmental disturbance data include: Based on the kinematic model, the mathematical model of the unmanned boat is constructed by combining its kinematic model, control model and environmental disturbance data; The viscosity coefficient and propeller thrust coefficient are taken as parameter vectors to be optimized, and the mathematical model of the unmanned vehicle is differentially discretized to construct a twin prediction model.

3. The digital twin method for unmanned boat navigation posture prediction according to claim 2 is characterized in that: The steps of acquiring GPS data and IMU data in real time and fusing the data using Kalman filter fusion technology to obtain sensor fusion data include: Acquire GPS data and IMU data in real time and perform preprocessing; Use Kalman filter to perform state estimation and prediction on the preprocessed GPS data and IMU data to obtain GPS observation data and IMU observation data; The Kalman gains of GPS observation data and IMU observation data are weighted, and the state prediction generated by the fusion is calculated to obtain real-time observation data.

4. The digital twin method for unmanned boat navigation posture prediction according to claim 3 is characterized in that: The steps of dynamically extracting fused state data through a sliding window, processing it segment by segment, and recursively calculating the state variables and model parameters of the unmanned boat to optimize the twin prediction model include: Introducing a dynamic sliding window to process the fusion state data segment by segment; The parameter vector of the twin prediction model is used as the extended state vector to construct a three-degree-of-freedom state equation containing kinematic parameters and parameters to be identified; The three-degree-of-freedom unmanned boat motion twin model is used to recursively calculate the twin historical data and real-time observation data in the dynamic sliding window, estimate the state variables of the unmanned boat and update the model parameters in real time to optimize the twin prediction model.

5. The digital twin method for unmanned boat navigation posture prediction according to claim 4 is characterized in that: The total length of the sliding window consists of a basic window and an adaptive adjustment window. The basic window length is determined based on the dynamic response time constant of the unmanned boat, and the adaptive adjustment window length is dynamically adjusted according to the posture data characteristics and prediction error. Adopting a periodic adaptive adjustment strategy, combined with the wave spectrum main period and UKF information covariance feedback, the sliding window length is adjusted in real time; When the environmental disturbance parameter exceeds the preset threshold, the window length is forcibly extended to enhance the robustness of the model.

6. The digital twin method for unmanned vehicle navigation posture prediction according to claim 5 is characterized in that: When the sliding window reaches the data boundary, the optimized twin prediction model is obtained. The steps of using the optimized digital twin model to predict the navigation posture of the unmanned boat include: When the sliding window reaches the data boundary, the final identification result is output, the twin model parameters are updated, and the optimized twin prediction model is obtained; By using the optimized digital twin model, combining real-time observation data with twin historical data, the future navigation status of the unmanned boat is predicted through the rolling time window technology to predict the navigation posture of the unmanned boat.

7. The digital twin method for unmanned vehicle navigation posture prediction according to claim 6 is characterized in that: Twin model parameters include: State matrix, state transfer matrix, observation matrix, process noise covariance matrix and measurement noise covariance matrix.

Images