A method for predicting the maneuvering motion of hydrogen fuel cell-powered ships based on multi-source data fusion

Abstract

This invention discloses a method for predicting the maneuvering motions of hydrogen fuel cell-powered ships based on multi-source data fusion, relating to the field of intelligent ship navigation. The method collects geodetic coordinate system position and heading data from the BeiDou Navigation Satellite System and ship hull coordinate system acceleration data from the Inertial Measurement Unit (IMU). After differential and integral preprocessing, a neural network model is built. A fusion-type loss function is constructed based on the transformation relationship between the ship hull coordinate system and the geodetic coordinate system. The model is trained using a backpropagation algorithm and can also be extended to fuse data from multiple sensors such as GPS. This invention fully leverages the complementary advantages of multiple sensor sources, effectively compensating for the inherent deficiencies of single-sensor data, significantly improving the accuracy and robustness of ship maneuvering motion prediction, and providing reliable technical support for intelligent navigation decision-making and safety assurance of hydrogen fuel cell-powered ships.

Description

Technical Field

[0001] This invention relates to the field of intelligent navigation technology for ships, specifically to a method for predicting the maneuvering motion of hydrogen fuel cell-powered ships based on multi-source data fusion. Background Technology

[0002] With the global energy structure transformation and increasingly stringent environmental requirements, hydrogen fuel cell ships, as a core development direction for green shipping, have an increasingly urgent need for intelligent navigation and energy efficiency management. Ship maneuvering motion prediction methods based on artificial intelligence technology can provide crucial decision support for intelligent ship navigation, thus becoming a current research hotspot and demonstrating a certain level of prediction accuracy in practical applications.

[0003] However, existing neural network-based ship maneuvering motion prediction methods generally suffer from the problem of relying on a single data source, typically depending on data from only one sensor, either the BeiDou Navigation Satellite System or an Inertial Measurement Unit (IMU). While the BeiDou Navigation Satellite System can provide real-time ship position and heading information, it is susceptible to factors such as satellite signal blockage and multipath effects, resulting in significant real-time data errors that are insufficient for high-precision prediction. The IMU can collect high-frequency acceleration and angular acceleration data from the ship, but it suffers from integration and accumulation errors, leading to a continuous decline in prediction accuracy over long-term use. The limitations of single-sensor data restrict the improvement of ship maneuvering motion prediction accuracy, thereby affecting the intelligent development and navigation safety of hydrogen fuel cell-powered ships.

[0004] In recent years, multi-sensor data fusion technology has gradually attracted attention. By integrating the global positioning capability of the BeiDou Navigation Satellite System with the high-frequency motion sensing capability of the IMU, the shortcomings of a single sensor can be effectively compensated, providing a new technical path for improving the accuracy of ship maneuvering motion prediction. Based on this, this invention proposes a method for predicting the maneuvering motion of hydrogen fuel cell-powered ships based on multi-source data fusion. By optimizing the neural network algorithm and loss function design, high-precision ship maneuvering motion prediction is achieved. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies where single sensor data leads to insufficient accuracy in ship maneuvering motion prediction. It provides a method for predicting the maneuvering motion of hydrogen fuel cell-powered ships based on multi-source data fusion. By fusing BeiDou positioning system and IMU data and constructing a loss function based on ship kinematics model, the prediction accuracy is significantly improved, providing reliable technical support for the intelligent navigation of hydrogen fuel cell-powered ships.

[0006] To solve the above problems, the technical solution of the present invention is as follows: A method for predicting the maneuvering motions of a hydrogen fuel cell-powered ship based on multi-source data fusion includes the following steps: S1: Acquire multi-source sensor data of the hydrogen fuel cell-powered ship. The multi-source sensor data includes the ship's longitudinal position x, lateral position y, and heading ψ in the geodetic coordinate system collected by the Beidou positioning system, as well as acceleration and angular acceleration data in the ship's coordinate system collected by the inertial measurement unit. S2: Process the multi-source sensor data collected in step S1, obtain the ship speed in the geodetic coordinate system through first-order difference operation, and obtain the ship motion state vector V in the hull coordinate system through integration operation. S3: Construct a neural network model for ship maneuvering motion prediction. The inputs of the neural network model include the initial motion state of the ship and control commands, and the output is the predicted value of the ship's motion state at future time steps. ; S4: Based on the transformation relationship between the ship's coordinate system and the geodetic coordinate system in the ship's kinematics model, a loss function that integrates BeiDou positioning data and IMU data is constructed. S5: The backpropagation algorithm is used to train the neural network model, and the model parameters are updated by minimizing the loss function value to obtain the trained ship maneuvering motion prediction model; S6: Input the real-time collected initial motion state of the ship and control commands into the trained prediction model, and output the prediction results of the ship's maneuvering motion state at future time steps. The prediction results include longitudinal velocity u, lateral velocity v and bow roll rate r.

[0007] Furthermore, in step S2, the ship speed in the geodetic coordinate system is calculated as follows: based on the position and heading data of adjacent time steps i and i+1 collected by the BeiDou positioning system. , The first-order difference formula is used. The calculation yields the result, where Δt is the sampling interval. .

[0008] Furthermore, in step S2, the ship motion state vector V in the hull coordinate system is calculated in the following way: based on acceleration and angular acceleration data collected by the IMU. Assuming the ship starts from a standstill, the integral formula is used. The calculation yielded, where u is the longitudinal velocity, v is the lateral velocity, and r is the yaw rate.

[0009] Furthermore, in step S3, the structure of the neural network model is determined by hyperparameters, which include the number of hidden layers, network width, activation function, and connection method. The activation function is selected from one or more combinations of ReLU function, Sigmoid function, or Tanh function.

[0010] Furthermore, in step S4, the expression for the loss function is:

[0011] ,in This represents the predicted motion state at time step p, output by the neural network model. Let p be the ship's velocity in the geodetic coordinate system at time step p. Let p be the ship's heading at time step p.

[0012] Furthermore, in step S5, the backpropagation algorithm calculates the gradient of the loss function with respect to the neural network model parameters and iteratively updates the model parameters using the gradient descent method until the loss function value converges to a preset threshold.

[0013] Furthermore, in step S3, the control commands include propeller speed n and rudder angle δ, and the initial motion state of the ship includes longitudinal speed u, lateral speed v and yaw rate r.

[0014] Furthermore, the transformation relationship between the ship's coordinate system and the geodetic coordinate system is achieved through a rotation matrix, and the transformation formula is as follows: , where η is the ship's state vector in the geodetic coordinate system, V is the ship's motion state vector in the hull coordinate system, and ψ is the ship's heading.

[0015] Furthermore, the sampling interval Based on the sampling frequencies of the BeiDou positioning system and the IMU, the sampling frequency of the BeiDou positioning system is 1-10Hz, and the sampling frequency of the IMU is 100-1000Hz.

[0016] Furthermore, the multi-source sensor data may also include one or more of GPS positioning data and ship radar data, and the fusion of multi-sensor data can be achieved by expanding the input dimension of the loss function.

[0017] The beneficial effects of this invention are as follows: 1. This invention, by integrating BeiDou positioning system and IMU multi-source data, fully leverages the global positioning advantage of BeiDou positioning and the high-frequency sensing advantage of IMU, effectively compensating for the limitations of single sensor data and significantly improving the accuracy of ship maneuvering motion prediction. 2. A fusion loss function is constructed based on the ship kinematics model, and the organic fusion of multi-source data is achieved through coordinate system transformation, ensuring the rationality and effectiveness of model training; 3. The hyperparameters of the neural network model can be flexibly adjusted to adapt to the motion characteristics of different types of hydrogen fuel cell-powered ships, and it has strong versatility and scalability. 4. The forecasting method has a simple process, high computational efficiency, and can output the forecast results of ship motion status in real time, meeting the real-time requirements of intelligent navigation of hydrogen fuel cell-powered ships. Attached Figure Description

[0018] The invention will be further described below with reference to the accompanying drawings: Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] A method for predicting the maneuvering motions of a hydrogen fuel cell-powered ship based on multi-source data fusion includes the following steps: S1: Multi-source data acquisition, real-time acquisition of the longitudinal position of the hydrogen fuel cell-powered ship in the geodetic coordinate system via the BeiDou positioning system. Horizontal position and heading The inertial measurement unit (IMU) collects real-time acceleration and angular acceleration data of the ship in the ship's coordinate system. Simultaneously, it collects the ship's control commands, including propeller speed. With rudder angle By simultaneously collecting data from two types of sensors, the BeiDou positioning system's global positioning capability and the IMU's high-frequency motion sensing capability are fully utilized, thus compensating for the inherent defects of single sensor data from the source. This addresses the issues of large real-time errors caused by satellite signal blockage and multipath effects in BeiDou positioning data, as well as the long-term accuracy degradation of IMU data due to integrated cumulative errors. S2: Data preprocessing, performing targeted processing on the multi-source sensor data collected in step S1: calculating the ship speed in the geodetic coordinate system based on adjacent time step data collected by the BeiDou positioning system. Acceleration and angular acceleration data acquired by IMU Calculate the ship's motion state vector in the hull coordinate system This preprocessing process enables the adaptation of data in different coordinate systems, laying the foundation for subsequent multi-source data fusion and ensuring the consistency of data in both time and space dimensions. S3: Building a neural network architecture, constructing a neural network model for ship maneuvering motion prediction, the input of which is the ship motion state vector at the current time step. propeller speed and rudder angle The output is the predicted value of the ship's motion state for the next time step. The structure of the neural network model is determined by hyperparameters such as the number of hidden layers, network width, activation function and connection method. With the help of the nonlinear fitting ability of the neural network, it can accurately learn the complex mapping relationship between the initial motion state of the ship, control commands and future motion state, and adapt to the motion characteristics of different types of hydrogen fuel cell powered ships. S4: Construction of a fusion loss function. Based on the transformation relationship between the ship's coordinate system and the geodetic coordinate system in the ship's kinematics model, a loss function that can fuse BeiDou positioning data and IMU data is constructed. This loss function realizes the organic unification of multi-source data, so that the model training process is constrained by two types of data sources at the same time, avoiding model bias caused by a single data source, and ensuring the rationality and effectiveness of model training. S5: Model training optimization. The backpropagation algorithm is used to train the neural network model. By calculating the gradient of the loss function with respect to the model parameters, the model parameters are iteratively updated to minimize the loss function value until the loss function converges to the preset threshold, and the trained ship maneuvering motion prediction model is obtained. This training process allows the model to fully learn the ship motion characteristics contained in multi-source data and continuously optimize the parameter combination to reduce prediction error. S6: Real-time motion prediction. This involves inputting the real-time collected initial motion state of the ship, control commands, and preprocessed multi-source sensor data into the trained prediction model, and outputting a prediction result of the ship's maneuvering motion state within a preset time step. The prediction result includes longitudinal velocity. lateral velocity and bow roll rate This forecasting process can quickly output high-precision results, providing reliable data support for intelligent navigation decisions of ships and effectively improving the navigation safety and energy efficiency management level of hydrogen fuel cell-powered ships.

[0021] Furthermore, in step S2, the ship speed in the geodetic coordinate system Calculate as follows: Let and For adjacent time steps, the data collected based on the BeiDou positioning system is the first... Time step geodetic coordinate system state vector With the Time step geodetic coordinate system state vector The first-order difference formula is used. Calculated; where, , This represents the longitudinal position of the ship in the geodetic coordinate system. This represents the ship's lateral position in the geodetic coordinate system. For the ship's heading in the geodetic coordinate system, The sampling interval is... For the first The ship velocity vector in the time-step geodetic coordinate system; this first-order difference operation can transform discrete position data collected by the BeiDou positioning system into continuous velocity data, solving the problem that the BeiDou positioning system cannot directly output high-frequency velocity data. Moreover, the obtained velocity data has a global positioning benchmark and no error accumulation characteristics, which can provide a reliable velocity reference benchmark for subsequent multi-source data fusion and further improve the accuracy of data fusion.

[0022] Furthermore, in step S2, the ship motion state vector in the hull coordinate system Calculated as follows: Assuming the ship starts from a standstill (initial speed 0), the [number]th [time] data collected by the IMU is [data / data / mechanical data]. Time step acceleration and angular acceleration data Using the integral formula Calculation yields the first Motion state vector at time step ;in, , Let be the longitudinal velocity of the ship in the hull coordinate system. Let be the ship's lateral velocity in the hull coordinate system. Let be the ship's bow roll angular velocity in the ship's coordinate system. The acceleration and angular acceleration data are directly measured by the IMU in the ship's coordinate system. Index for summation of integrals (values ​​range from 1 to...) ), The sampling interval is defined as follows: This integral operation fully utilizes the characteristics of high-frequency data acquisition from the IMU to convert acceleration signals into velocity signals, enabling precise capture of instantaneous changes in ship motion. It complements the velocity data derived from the BeiDou positioning system, further enriching the perception dimensions of ship motion status and providing a solid data foundation for the deep fusion of multi-source data.

[0023] Furthermore, in step S3, the range and selection method of the hyperparameters are as follows: the number of hidden layers is 2-8, the network width (number of neurons per layer) is 64-512, the activation function is selected from one or more combinations of ReLU, Sigmoid, or Tanh functions, and the connection method is fully connected or residual connected; wherein, the core mapping relationship of the neural network model satisfies In the formula The predicted motion state output by the neural network. To predict longitudinal velocity, To predict lateral velocity, To predict the bow roll rate, For the first The actual motion state vector at each time step. For the first Time step propeller speed, For the first Time rudder angle, The neural network constructs a maneuvering motion prediction function; by flexibly adjusting hyperparameters, the neural network model can be adapted to hydrogen fuel-powered ships of different tonnages and navigation characteristics, enhancing the model's generalization ability and versatility, reducing the model's dependence on specific ships, and ensuring the consistency of model input and output through explicit mapping relationships, providing stable model architecture support for high-precision forecasting.

[0024] Furthermore, in step S4, the expression for the loss function is: ; in, This represents the cumulative loss value during model training. For time step index (value range is 1 to 1) ), For the first Ship velocity in the time step geodetic coordinate system For the first The motion state prediction value output by the time-step neural network. ( For the first Predicting longitudinal velocity using time steps For the first Time step prediction of lateral velocity, For the first (Time step prediction of bow roll rate) For the first The ship's heading, matrix, is collected by the BeiDou positioning system in time step. For the first The coordinate system rotation matrix for each time step; this loss function uses the coordinate system rotation matrix to represent the motion state in the ship's coordinate system predicted by the neural network. Converting the equivalent velocity to the geodetic coordinate system and comparing it with the geodetic coordinate system velocity derived from the BeiDou positioning system. Error calculation is performed, and the loss value is obtained by summing the squares. This achieves deep integration of BeiDou positioning data and IMU data, ensuring that the model can absorb the advantages of both types of data during training, and significantly improving the model's prediction accuracy and robustness.

[0025] Furthermore, in step S5, the execution process of the backpropagation algorithm is as follows: first, the loss function is calculated. The partial derivatives (gradients) of the weights, biases, and other parameters of each layer of the neural network are used to iterate and update the model parameters in the negative direction of the gradient using gradient descent or its improved algorithms (such as Adam's algorithm and RMSProp algorithm). During the iteration, the change in the loss function value is monitored in real time. When the loss function value remains within a preset threshold (10) for a preset number of consecutive preset number of iterations (50-200), the algorithm is considered successful. -4~10 -6 When the error is within a certain range, the model training is considered converged, and parameter updates are stopped. This algorithm uses the chain rule to propagate the error of the loss function from the output layer to the input layer, accurately calculates the contribution of each parameter to the error, and gradually adjusts the parameters through gradient descent algorithms to reduce the model's prediction error. This ensures that the neural network model converges quickly to the optimal parameter combination, improves model training efficiency, and effectively avoids overfitting or underfitting, ensuring that the trained model has stable prediction accuracy and generalization ability in practical applications.

[0026] Furthermore, in step S3, the initial motion state vector of the ship... ,in The physical meaning of is the longitudinal velocity of a ship along the x-axis of the ship's coordinate system (unit: m / s). The lateral velocity of the ship along the y-axis of the ship's coordinate system (unit: m / s). The bow roll rate (in rad / s) is the angle of rotation of the ship about the z-axis of the ship's coordinate system; in the control command, the propeller speed is... The unit is r / min (range: 0~2000 r / min), rudder angle The unit is rad (range: -π / 5 rad to π / 5 rad, corresponding to -36° to 36°); by clarifying the physical definition, unit and value range of ship motion state parameters and control commands, standardized input data is provided for neural network models, avoiding model training failure or forecast distortion caused by ambiguous parameter definitions or chaotic value ranges. At the same time, the operability and reliability of the method are improved, laying the foundation for engineering applications and ensuring that the model can stably receive effective input and output accurate forecast results in different navigation scenarios.

[0027] Furthermore, in step S4, the transformation relationship between the ship's coordinate system and the geodetic coordinate system is achieved through a rotation matrix, and the transformation formula is: ; in, The ship's state vector in the geodetic coordinate system ( Vertical position, unit: m; Horizontal position, unit: m; (Heading, unit: rad) This is the vector representing the ship's motion state in the ship's coordinate system. For heading A 2D rotation matrix is ​​used to achieve attitude and position transformation between the ship's coordinate system and the geodetic coordinate system; this transformation relationship is achieved through the heading... By constructing a rotation matrix and using matrix multiplication to solve the compatibility problem of data under different coordinate systems, a coordinate system unification of multi-source data is achieved, providing a key mathematical foundation for multi-source data fusion. This ensures that BeiDou positioning data and IMU data can be used for error calculation and model training in the same coordinate system, further improving the accuracy and effectiveness of data fusion.

[0028] Furthermore, the sampling interval The sampling frequency is determined as follows: It is set collaboratively based on the sampling frequencies of the BeiDou Navigation Satellite System (BDS) and the IMU (Integrated Measurement Unit). The sampling frequency of the BDS is 1-10Hz (corresponding to a sampling interval of 0.1-1s); the sampling frequency of the IMU is 100-1000Hz (corresponding to a sampling interval of 0.001-0.01s). During data preprocessing, the IMU data is downsampled to ensure that the sampling interval of the preprocessed IMU data matches the sampling interval of the BDS. Maintaining consistency; by collaboratively setting the sampling interval and downsampling the high-frequency IMU data, the time-dimensional data asynchrony error can be eliminated, ensuring the synchronization of the two types of sensor data in the time dimension, avoiding data misalignment caused by sampling frequency differences, ensuring the reliability of multi-source data fusion, reducing data processing volume and model training complexity, and achieving a balance between forecast accuracy and computational efficiency.

[0029] Furthermore, the multi-source sensor data in step S1 can be expanded to include one or more of GPS positioning data, ship radar data, and wind speed and direction sensor data. When expanding sensor data, the feature quantities corresponding to the expanded data are incorporated into the loss function calculation by increasing the input dimension of the loss function, thereby achieving the fusion of more source data. By expanding the sources of sensor data, the perception dimension of ship motion status can be increased, making up for the insufficient perception of the original dual-sensor data in complex sea conditions (such as strong winds, giant waves, and severe signal blockage scenarios). By expanding the dimension of the loss function, the collaborative constraint between the new data and the original data can be achieved, further improving the prediction accuracy and robustness of the method in complex navigation scenarios, enhancing the scalability and adaptability of the method, and meeting the intelligent navigation needs of hydrogen fuel cell-powered ships in different navigation environments.

[0030] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion, characterized in that, Includes the following steps: S1: Acquire multi-source sensor data of the hydrogen fuel cell-powered ship. The multi-source sensor data includes the ship's longitudinal position x, lateral position y, and heading ψ in the geodetic coordinate system collected by the Beidou positioning system, as well as acceleration and angular acceleration data in the ship's coordinate system collected by the inertial measurement unit. S2: Process the multi-source sensor data collected in step S1, obtain the ship speed in the geodetic coordinate system through first-order difference operation, and obtain the ship motion state vector V in the hull coordinate system through integration operation. S3: Construct a neural network model for ship maneuvering motion prediction. The inputs of the neural network model include the initial motion state of the ship and control commands, and the output is the predicted value of the ship's motion state at future time steps. ; S4: Based on the transformation relationship between the ship's coordinate system and the geodetic coordinate system in the ship's kinematics model, a loss function that integrates BeiDou positioning data and IMU data is constructed. S5: The backpropagation algorithm is used to train the neural network model, and the model parameters are updated by minimizing the loss function value to obtain the trained ship maneuvering motion prediction model; S6: Input the real-time collected initial motion state of the ship and control commands into the trained prediction model, and output the prediction results of the ship's maneuvering motion state at future time steps. The prediction results include longitudinal velocity u, lateral velocity v and bow roll rate r.

2. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, In step S2, the ship speed in the geodetic coordinate system is calculated as follows: based on the position and heading data of adjacent time steps i and i+1 collected by the BeiDou positioning system. , The first-order difference formula is used. The calculation yielded, where The sampling interval is... .

3. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, In step S2, the ship motion state vector V in the hull coordinate system is calculated in the following way: based on acceleration and angular acceleration data collected by the IMU. Assuming the ship starts from a standstill, the integral formula is used. The calculation yielded, where u is the longitudinal velocity, v is the lateral velocity, and r is the yaw rate.

4. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, In step S3, the structure of the neural network model is determined by hyperparameters, including the number of hidden layers, network width, activation function, and connection method. The activation function is selected from one or more combinations of ReLU, Sigmoid, or Tanh functions. The core mapping relationship of the neural network model satisfies... In the formula The predicted motion state output by the neural network. To predict longitudinal velocity, To predict lateral velocity, To predict the bow roll rate.

5. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, In step S4, the expression for the loss function is: ;in This represents the predicted motion state at time step p, output by the neural network model. Let p be the ship's velocity in the geodetic coordinate system at time step p. For the ship's heading at time step p, the matrix... Let be the coordinate system rotation matrix at time step p.

6. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, In step S5, the backpropagation algorithm calculates the gradient of the loss function with respect to the neural network model parameters and iteratively updates the model parameters using the gradient descent method until the loss function value converges to a preset threshold.

7. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, In step S3, the control commands include propeller speed n and rudder angle. The initial motion state of the ship includes longitudinal velocity u, lateral velocity v, and bow roll rate r.

8. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, The transformation relationship between the ship's coordinate system and the geodetic coordinate system is achieved through a rotation matrix, and the transformation formula is as follows: , where η is the ship's state vector in the geodetic coordinate system, V is the ship's motion state vector in the hull coordinate system, and ψ is the ship's heading.

9. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 2, characterized in that, The sampling interval Δt is determined based on the sampling frequency of the BeiDou positioning system and the IMU. The sampling frequency of the BeiDou positioning system is 1-10Hz, and the sampling frequency of the IMU is 100-1000Hz.

10. The method for predicting the maneuvering motion of a hydrogen fuel cell-powered ship based on multi-source data fusion according to claim 1, characterized in that, The multi-source sensor data may also include one or more of GPS positioning data and ship radar data, and the fusion of multi-sensor data is achieved by expanding the input dimension of the loss function.

Images