Zihao Wang

Zihao Wang

Associate Professor at SHU

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

Future Blog Post

less than 1 minute read

Published:

This post will show up by default. To disable scheduling of future posts, edit config.yml and set future: false.

Blog Post number 4

less than 1 minute read

Published:

This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.

Blog Post number 3

less than 1 minute read

Published:

This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.

Blog Post number 2

less than 1 minute read

Published:

This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.

Blog Post number 1

less than 1 minute read

Published:

This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.

portfolio

publications

Identification modeling and prediction of ship maneuvering motion based on LSTM deep neural network

Published in Journal of Marine Science and Technology, 2021

This paper proposes a novel system identification scheme to obtain a MIMO model of ship maneuvering motion, which can leverage the temporal correlation from the constructed training data to learn the underlying feasible model robust to extraneous noise. The scheme is based on long–short-term-memory (LSTM) deep neural network, which is more easily trained than traditional feedforward neural network with more complicated network structure. First, multiple datasets of simulated standard maneuvers (10°/10° and 20°/20° zigzag, 35° turning circle) of a KVLCC2 model are artificially polluted with white noise of various levels and used simultaneously to train the deep neural network model. Meanwhile, the data of 15°/15° zigzag maneuver are used to facilitate the training process to alleviate overfitting problem. Second, different datasets of modified zigzag tests are used to validate the generalization performance and robustness to noise of the trained neural network model. The training and validation results demonstrate that a mapping between the dynamics of ship motion and the computation in LSTM deep neural network is correctly identified. This mapping indicates that the complex nonlinear features of ship maneuvering motion can be learned from the measured temporal data, using standard training techniques for deep neural networks. An equivalent LSTM deep neural network model with better generalization performance and robustness is established, and its accuracy in predicting ship maneuvering motion is validated.

Recommended citation: Jiang, Y., Hou, X. R., Wang, X. G., Wang, Z. H., Yang, Z. L., & Zou, Z. J. (2022). Identification modeling and prediction of ship maneuvering motion based on LSTM deep neural network. Journal of Marine Science and Technology, 27(1), 125-137.
Download Paper

Nonparametric modeling of ship maneuvering motion based on self-designed fully connected neural network

Published in Ocean Engineering, 2022

A self-designed fully connected neural network is proposed for nonparametric modeling of ship maneuvering motion by extracting the dynamic characteristics of ship motion from observed input-output data. The automatic design is conducted by Bayesian optimization, which aims at minimizing the validation error with respect to different hyperparameter settings of the neural network, including learning rate, activation function, as well as network depth and widths. Gaussian process is adopted as the surrogate model to estimate the unobserved validation errors, and lower confidence bound is adopted for trading off exploration and exploitation in the process of optimization. Taking the KVLCC2 ship as the study object, the experimental data provided by Hamburg Ship Model Basin is utilized for modeling and validation of prediction, and the detailed procedures to prepare and preprocess the input-output data are discussed. After well designed and trained, the selected neural network model is evaluated by the prediction results of ship maneuvers in comparison with the experimental data. It shows the high prediction accuracy and strong generalization ability of the established neural network model, indicating the effectiveness of the proposed modeling method.

Recommended citation: He, H. W., Wang, Z. H., Zou, Z. J., & Liu, Y. (2022). Nonparametric modeling of ship maneuvering motion based on self-designed fully connected neural network. Ocean Engineering, 251, 111113.
Download Paper

Black-box modeling of ship maneuvering motion based on multi-output nu-support vector regression with random excitation signal

Published in Ocean Engineering, 2022

This paper proposes a novel method for offline black-box modeling of ship maneuvering by utilizing the training data from random maneuvers under medium rudder angle with random amplitude and duration. The identification algorithm adopted is a multi-output ν(‘nu’)-Support Vector Regression, MO-ν-SVR, that has higher computational efficiency and better operability than a conventional ν-SVR. The ONRT vessel is taken as the study object, and numerical simulations are conducted to provide the training, validation and testing datasets. The superiority of the proposed random maneuver over the standard zig-zag maneuver is demonstrated by a contrastive study where the excitation signals from the random maneuver and the 20°/20° zig-zag maneuver are used for training the model separately. To examine the robustness of the proposed modeling method and the identified model, three levels of white noise are added into the raw simulation data for training the model. To explore the effectiveness and generalization ability of the identified model on different motion patterns of ship maneuvering, course-keeping, course-changing, and turning motions are examined separately. The results demonstrate that the model trained by the excitation signals of the random maneuver has better generalization ability and robustness, verifying the feasibility and practicality of the proposed modeling method.

Recommended citation: Zhang, Y. Y., Wang, Z. H., & Zou, Z. J. (2022). Black-box modeling of ship maneuvering motion based on multi-output nu-support vector regression with random excitation signal. Ocean Engineering, 257, 111279.
Download Paper

A data driven method for multi-step prediction of ship roll motion in high sea states

Published in Ocean Engineering, 2023

Ship roll motion in high sea states has large amplitudes and nonlinear dynamics, and its prediction is significant for operability, safety, and survivability. This paper presents a novel data-driven methodology to provide a multi-step prediction of ship roll motions in high sea states. A hybrid neural network is proposed that combines long short-term memory (LSTM) and convolutional neural network (CNN) in parallel. The motivation is to extract the nonlinear dynamic characteristics and the hydrodynamic memory information through the advantage of CNN and LSTM, respectively. For the feature selection, the time histories of motion states and wave heights are selected to involve sufficient information. Taken a scaled KCS as the study object, the ship motions in sea state 7 irregular long-crested waves are simulated and used for the validation. The results show that at least one period of roll motion can be accurately predicted. Compared with the single LSTM and CNN methods, the proposed method has better performance in predicting the amplitude of roll angles. Besides, the comparison results also demonstrate that selecting motion states and wave heights as feature space improves the prediction accuracy, verifying the effectiveness of the proposed method.

Recommended citation: Zhang, D., Zhou, X., Wang, Z. H., Peng, Y., & Xie, S. R. (2023). A data driven method for multi-step prediction of ship roll motion in high sea states. Ocean Engineering, 276, 114230.
Download Paper

Online modeling of ship maneuvering motion with varying loading conditions

Published in ASME 2023 42nd International Conference on Offshore Mechanics and Arctic Engineering (OMAE 2023), 2023

Recommended citation: Yu, Y. H., Wang, Z. H., Qu, D., Song, R., & Peng, Y. (2023, June). Online modeling of ship maneuvering motion with varying loading conditions. In International Conference on Offshore Mechanics and Arctic Engineering (Vol. 86878, p. V005T06A047). American Society of Mechanical Engineers.

Evaluation of Parametric Modeling Method for Ship Maneuvering Motion With Experimental Data

Published in ASME 2023 42nd International Conference on Offshore Mechanics and Arctic Engineering (OMAE 2023), 2023

Recommended citation: Zhong, Y. X., Wang, Z. H., & Zou, Z. J. (2023, June). Evaluation of Parametric Modeling Method for Ship Maneuvering Motion With Experimental Data. In International Conference on Offshore Mechanics and Arctic Engineering (Vol. 86878, p. V005T06A046). American Society of Mechanical Engineers.

Data-Driven Modeling of Unmanned Surface Vehicle’s Maneuvering Motion Based on Real Navigational Data - 基于实航数据驱动的无人艇操纵运动辨识建模

Published in Ship Building of China - 中国造船, 2024

This study conducts a data-driven modeling research of the “Jinghai” unmanned surface vehicle based on lake trial data. The objective is to develop a kinetic model that captures the actual characteristics of the vehicle. The framework incorporates the nonlinear hydrodynamic model structure and the model of the propulsion system as prior knowledge constraints. Support vector regression is employed to complete the modeling process using a training dataset excited by a sequence of random rudder angles. The study places a particular emphasis on assessing the feasibility of modeling with environmental disturbances. The results demonstrate that the proposed method enables rapid construction of a nonlinear model for unmanned surface vehicles. The prior model structure effectively mitigates the impact of waves and currents in navigational data. The model accurately predicts the motion states during turning and zigzag trials, and exhibits good generalization ability.

Recommended citation: Wang, Z., Cheng, J., Xie, W., Song, R., Peng, Y.. Data-Driven Modeling of Unmanned Surface Vehicle's Maneuvering Motion Based on Real Navigational Data[J]. Ship Building of China, 2024, 65(1). doi: 10.3969/j.issn.1000-4882.2024.01.013

Online prediction of ship maneuvering motions based on adaptive weighted ensemble learning under dynamic changes

Published in Engineering Applications of Computational Fluid Mechanics, 2024

Dynamic changes in ship maneuverability challenge the accuracy and effectiveness of ship maneuvering models. This paper proposes an online prediction method based on the adaptive weighted ensemble learning framework, which can adaptively update the model according to changes in maneuverability, especially for reoccurring changes. The method contains two main mechanisms: the change monitoring mechanism and the adaptive weighting mechanism. The former identifies the change in ship dynamics and decides when to incorporate a new base model; the latter adjusts the weights of the base models to align with current scenarios, thus ensuring the predictive accuracy. To assess the method’s effectiveness under varying ship dynamics, the online prediction of ship maneuvering motions under speed-induced dynamic changes is investigated. Compared with the offline model, the result demonstrates the superiority of the adaptive weighted ensemble model. The proposed method can consistently provide accurate predictions in the scenarios with reoccurring changes, and can also enhance the model capability by adjusting weights to cope with some unencountered changes.

Recommended citation: Yu, Y., Hao, H., Wang, Z., Peng, Y., & Xie, S. (2024). Online prediction of ship maneuvering motions based on adaptive weighted ensemble learning under dynamic changes. Engineering Applications of Computational Fluid Mechanics, 18(1), 2341922.
Download Paper

Optimizing Roll Motion Forecasting for Unmanned Surface Vessels Using PSO-Enhanced LSTM Networks

Published in IEEE 4th International Conference on Information Technology, Big Data and Artificial Intelligence, 2024

https://ieeexplore.ieee.org/abstract/document/10867884/

Recommended citation: Qu, T., Wang, Z., & Xie, W. (2024, December). Optimizing Roll Motion Forecasting for Unmanned Surface Vessels Using PSO-Enhanced LSTM Networks. In 2024 IEEE 4th International Conference on Information Technology, Big Data and Artificial Intelligence (ICIBA) (Vol. 4, pp. 1539-1543). IEEE.

Construction of Maritime Route Networks for USVs in Complex Waterways Based on Geometric Element Analysis

Published in IEEE 4th International Conference on Information Technology, Big Data and Artificial Intelligence, 2024

The construction of maritime route networks is crucial for the autonomous navigation of unmanned surface vehicles (USV), as it supports global coordination among multi-USV systems in narrow and interconnected waterways, thereby avoiding path conflicts and blockages. This paper proposes a method for constructing route networks based on geometric element analysis (RN-GEA). https://ieeexplore.ieee.org/abstract/document/10867893/

Recommended citation: Xu, Z., Wang, Z., & Xie, W. (2024, December). Construction of Maritime Route Networks for USVs in Complex Waterways Based on Geometric Element Analysis. In 2024 IEEE 4th International Conference on Information Technology, Big Data and Artificial Intelligence (ICIBA) (Vol. 4, pp. 1624-1628). IEEE.

Hybrid Physics-ML Modeling for Marine Vehicle Maneuvering Motions in the Presence of Environmental Disturbances

Published in ArXiv Preprints, 2024

A hybrid physics-machine learning modeling framework is proposed for the surface vehicles' maneuvering motions to address the modeling capability and stability in the presence of environmental disturbances. From a deep learning perspective, the framework is based on a variant version of residual networks with additional feature extraction. Initially, an imperfect physical model is derived and identified to capture the fundamental hydrodynamic characteristics of marine vehicles. This model is then integrated with a feedforward network through a residual block. Additionally, feature extraction from trigonometric transformations is employed in the machine learning component to account for the periodic influence of currents and waves. The proposed method is evaluated using real navigational data from the 'JH7500' unmanned surface vehicle. The results demonstrate the robust generalizability and accurate long-term prediction capabilities of the nonlinear dynamic model in specific environmental conditions. This approach has the potential to be extended and applied to develop a comprehensive high-fidelity simulator.

Recommended citation: Wang, Z., Cheng, J., Xu, L., Hao, L., & Peng, Y. (2024). Hybrid Physics-ML Modeling for Marine Vehicle Maneuvering Motions in the Presence of Environmental Disturbances. arXiv preprint arXiv:2411.13908.
Download Paper

Target tracking control for Unmanned Surface Vehicles: An end-to-end deep reinforcement learning approach

Published in Ocean Engineering, 2024

Target tracking serves as a fundamental motion control function for Unmanned Surface Vehicles (USVs), requiring the USV to rapidly approach a moving target without prior knowledge of its behavior. However, system time lag, underactuation, and environmental disturbances often lead to response delays, degrading tracking efficiency. To address this, we propose a deep reinforcement learning-based end-to-end control method aimed at enhancing the USV’s tracking efficiency and responsiveness. Unlike conventional approaches, this method directly learns the mapping from sensor observations to control commands, optimizing decision-making and control actions within a unified framework. A specific deep reinforcement learning algorithm for target tracking is designed based on soft actor–critic framework and integrating predictive target information into the observation space to learn an anticipatory control policy. This paradigm enables the USV to comprehensively account for target movement uncertainty and its own maneuverability under environmental disturbances. Comparative studies are conducted using a high-fidelity simulator that considers the USV’s nonlinear dynamics and external influences. The results demonstrate that the proposed method outperforms conventional pure pursuit-based strategy, exhibiting a more efficient and adaptive tracking behavior, akin to human driving habits.

Recommended citation: Wang, Z., Hu, Q., Wang, C., Liu, Y., & Xie, W. (2025). Target tracking control for Unmanned Surface Vehicles: An end-to-end deep reinforcement learning approach. Ocean Engineering, 317, 120059.
Download Paper

A Geometric Analysis-Based Safety Assessment Framework for MASS Route Decision-Making in Restricted Waters

Published in arXiv preprints [eess.SY], 2025

To enhance the safety of Maritime Autonomous Surface Ships (MASS) navigating in restricted waters, this paper aims to develop a geometric analysis-based route safety assessment (GARSA) framework, specifically designed for their route decision-making in irregularly shaped waterways. Utilizing line and point geometric elements to define waterway boundaries, the framework enables to construct a dynamic width characterization function to quantify spatial safety along intricate waterways. An iterative method is developed to calculate this function, enabling an abstracted spatial property representation of the waterways. Based on this, we introduce a navigational safety index that balances global navigational safety and local risk to determine the safest route. To accommodate ship kinematic constraints, path modifications are applied using a dynamic window approach. A case study in a simulated Port of Hamburg environment shows that GARSA effectively identifies safe routes and avoids the risk of entering narrow waterways in an autonomous manner, thereby prioritizing safety in route decision-making for MASS in confined waters.

Recommended citation: Xu, Z., Wang, Z., Li, H., Yu, D., Yang, Z., & Wang, J. (2025). A Geometric Analysis-Based Safety Assessment Framework for MASS Route Decision-Making in Restricted Waters. arXiv preprint arXiv:2501.06670.
Download Paper

Modeling of Ship Maneuvering Motion Based on Neural Ordinary Differential Equations

Published in ISOPE International Ocean and Polar Engineering Conference, 2025

The safety of ship navigation and the optimization of maneuvering paths are closely dependent on the accurate prediction of maneuvering motions, which are typically governed as nonlinear dynamic models. This study proposes a Neural Ordinary Differential Equation (Neural ODE)-based modeling approach to improve the prediction of ship maneuvering behavior. In the framework, ship motion states and control signals are fed into a multi-layer neural network, which is then embedded within an ODE solver to model the system dynamics from a continuous-time perspective. This integration enables the model to capture the underlying dynamic characteristics of ship motion. The basic effectiveness of the proposed Neural ODE method is evaluated by the simulated data of the KRISO container ship under various scenarios, including turning circle, zigzag, and random rudder tests. The results demonstrate that the model has good generalization ability across different test conditions, confirming the fundamental applicability of the proposed method.

Recommended citation: Liu, S., & Wang, Z. (2025, June). Modeling of Ship Maneuvering Motion Based on Neural Ordinary Differential Equations. In ISOPE International Ocean and Polar Engineering Conference (pp. ISOPE-I). ISOPE.
Download Paper

A Sim-to-Real Transfer Framework for Enhancing Marine Vehicle Performance in Ocean Environments

Published in 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025

Reinforcement learning (RL) has gained attention for complex decision-making in uncertain environments. However, high costs and risks of real-world experimentation limit its direct application to marine vehicles. This motivates the use of simulation-based training and sim-to-real transfer techniques. Despite growing interest, a systematic understanding of how to design effective transfer strategies for marine contexts remains lacking. This paper presents a sim-to-real transfer framework tailored for marine vehicles, integrating high-fidelity, data-driven dynamics modeling with multi-factor domain randomization to address marine environmental uncertainties. Maneuvering data is utilized to extract nonlinear hydrodynamic characteristics of marine vehicles to enhance model realism. Additionally, domain randomization is explored across multiple environmental factors, including wind, wave, and current. To evaluate transferability, we construct a sim-to-sim platform with a pseudo-real environment that emulates the reality gap and adopt a path-following task using Soft Actor-Critic. We comprehensively assess the impacts of model fidelity and environmental randomization strategies on sim-to-real transfer performance. Results indicate that model accuracy positively impacts transfer performance, while aggressive domain randomization may reduce adaptability in calm conditions. Finally, a data-driven modeling and multi-factor randomization recipe is proposed for RL policy transfer in marine applications.

Recommended citation: Z. Zheng, Z. Wang and W. Xie, "A Sim-to-Real Transfer Framework for Enhancing Marine Vehicle Performance in Ocean Environments," 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hangzhou, China, 2025, pp. 1558-1565, doi: 10.1109/IROS60139.2025.11246159.
Download Paper

Multi-Factor Excitation-Based Data-Driven Approach for Wide-Speed-Range Modeling of Marine Vehicles

Published in IFAC-PapersOnLine, 2025

Accurate modeling of marine vehicle dynamics under varying maneuvering conditions is essential for control and trajectory planning. For data-driven modeling of marine vehicles, there is a lack of effective experimental design and modeling strategies to capture dynamic behaviors across different speed regimes. This study proposes a systematic data-driven framework that integrates multi-factor excitation signal design to sufficiently stimulate representative motion patterns. Specifically, a Latin Hypercube Sampling (LHS)-based method is employed to generate diverse control inputs that span a wide range of propeller revolutions and rudder angles, thereby enriching the training data. A Long Short-Term Memory network is then adopted to capture both transient dynamics and long-term dependencies from time-series data. The framework is evaluated on three representative marine vehicles and benchmarked against conventional excitation strategies, including Stratified Sampling, Random Rudder Sampling at fixed speed, and standard Zigzag maneuvers. Experimental results demonstrate that the LHS-based approach effectively captures the dynamic behaviors of marine vehicles across a broad speed range, ensuring consistent modeling performance under varying maneuvering conditions.

Recommended citation: Xia, A., Wang, Z., Wang, A., & Hao, L. (2025). Multi-Factor Excitation-Based Data-Driven Approach for Wide-Speed-Range Modeling of Marine Vehicles. IFAC-PapersOnLine, 59(22), 663-668.
Download Paper

A Physics-Informed Neural Network Method-based Prediction of Ship Rolling Motion

Published in IFAC-PapersOnLine, 2025

To enhance the prediction accuracy and robustness of ship rolling motion while addressing the limited generalization capability of traditional identification methods under complex sea conditions, a high-fidelity model for ship rolling motion based on Physics-Informed Neural Networks (PINN) is proposed. By constructing a time-series input-output relationship for ship rolling motion, incorporating its physical mechanisms, and employing a multi-layer neural network to model ship motion, the PINN-based rolling motion identification model is ultimately trained using a Mean Squared Error (MSE) loss function. Ship motion data collected from the South China Sea via inertial motion sensors is utilized for comparative experiments against a Support Vector Machine (SVM) algorithm across four dimensions: generalization validation, time scales, data volume, and sampling intervals. The results demonstrate that PINN-based significantly outperforms SVM, reducing the Root Mean Square Error (RMSE) by 36.98% in long-term predictions while improving the coefficient of determination (R²) by over 60%. Under limited training data conditions, PINN-based achieves an 87.1% reduction in RMSE and a 98.5% increase in R². These findings indicate that PINN-based enables high-fidelity identification of ship rolling motion models while exhibiting superior robustness and applicability.

Recommended citation: Fang, X., Zhu, M., Wang, Z. H., Guo, H. T., & Tian, K. (2025). A Physics-Informed Neural Network Method-based Prediction of Ship Rolling Motion. IFAC-PapersOnLine, 59(22), 878-883.
Download Paper

talks

teaching