Numerical simulation of waves has always been a very important research topic in the field of marine engineering. However, the limitations of wave simulation by commercial simulation software are found to be very obvious in practical experiments in the field of computational fluid dynamics. In a real marine environment, it is relatively difficult to simulate irregular waves. Many researchers prefer to write their desired irregular waves in open-source numerical simulation software, but writing them by themselves requires complex derivation of formulas. In addition, the poor accuracy of simulated waves can lead to a variety of problems. In this paper, the traditional numerical wave-making method is improved and optimized. Combining with the current stable deep learning framework in the computer field, we propose a model of superposition of solitary waves based on deep reinforcement learning to simulate complex waves in the real world. Solitary waves are used as the basis for deformation and superposition to simulate various complex waveforms. A combination of deep learning and the open-source software OpenFOAM is used to train a model that can be used to approximate the irregular waves needed in practice. By using deep learning to fit a function instead of a complex computational formula and by automatically exploring the wave stacking strategy, the intelligent body learns to fit the target wave automatically during interaction with the environment in order to obtain the simulation of irregular waves. It provides new ideas for the future wave simulation direction and constructs more realistic waves. In the future, we will investigate the direction of a further combination of deep learning and computational fluid dynamics.
In the field of coastal engineering, there are three ways to study waves: firstly, on-site observation study by manual or equipment, secondly, simulation study in a laboratory, and thirdly, theoretical analysis study. Due to the complexity and variability of waves, it is difficult to obtain data for theoretical observations, and theoretical studies deal with irregular waves (Zhao, 2020). In the field of numerical simulation, the OpenFOAM open-source software has a large consumer base. The software has the advantage of being editable source code and having fine control granularity. However, as far as is known, OpenFOAM is slightly less powerful than other numerical simulation software in the field of wave creation. However, wave calculation is very important in computational fluid dynamics. Therefore, a large number of scholars at home and abroad have conducted a lot of research in the field of wave generation in the early days. Some research about the principle was published, but the source code was not made public and was only available for internal use in the numerical simulation software developed independently. Higuera (2012a) developed olaFlow based on ihFoam. At present, the most effective open-source wave generation software is the wave2foam toolkit and the olaFlow toolkit, (Higuera, 2014) which are based on the secondary development of OpenFOAM (Hu, 2016) In this paper, we use the olaFlow toolkit to modify it based on the case of "wavemakerFlume" (Higuera, 2018), and the solitary waves created are manually superimposed. Using DRL (Deep reinforcement learning) technology coupled with traditional numerical simulation models, a free wave-making model is trained. The original case can only generate regular waves, but the observed waves are usually irregular waves to solve practical problems. Qu (2016) studied the comparison of solitary waves and tsunami waves, where the effects of two different waveforms have been analyzed. It was concluded that the main component of a tsunami waves acts at completely different spatial and temporal scales than solitary waves. The linear overlay theory of waves also guarantees the stability of the solver in the original case. Besides, there are also precedents for the combination of deep reinforcement learning and OpenFOAM dynamic grid technology (Xie, 2021). Based on the above ideas, this paper decides to use DRL technology to optimize the OpenFOAM wave-making method, and learn a model, which can automatically modify the wave-making function, calculate the displacement formula of the pusher, and use the pusher wave-making to create complex waves to fit the irregular wave of the target. At this stage, the target wave is an irregular wave superimposed by a more flexible solitary wave overlay model. Subsequent experiments will use observers of real-time real-world waves as target waves for further learning.
