This paper aims to contribute to the improvement of the energy extracted from Wave Energy Converters, arrayed in a wave farm, through the implementation of several actions, namely wave extrapolation using a dynamic neural network, wave type identification resorting to a machine learning system, and coordination of the interoperability, communication, and control of all the devices belonging to the wave farm via a Multi-Agent System. The Multi-Agent System was initially developed and subsequently tested through simulations. Simulations used the Inertial Sea Wave Energy Converter, which extracts power from sea waves, using gyroscopic motion, to generate electricity. This device can be originally controlled by proportional-derivative controllers with parameters that change with the sea state. With this Multi-Agent System, ocean waves and corresponding sea states are identified using a machine learning wave classifier method. Information is passed on so that all devices can use optimal control parameters. Therefore, energy extraction by the wave farm is enhanced due to the combination of the Multi-Agent System with automatic learning methods, while its architecture is resilient to communication problems between devices. Simulation results for a case study showed an increase in the average power absorbed, proving the effectiveness of combining a Multi-Agent System with automatic learning methods. Conclusions demonstrated that the proposed system architecture is a viable alternative to the original wave forecasting model. Additionally, this Multi-Agent System can be adapted to wave farms with other Wave Energy Converters and other types of control.

本文旨在通过执行以下几种动作来改善从排列在波浪场中的波浪能量转换器中提取的能量，即使用动态神经网络进行波浪外推，借助机器学习系统进行波浪类型识别以及通过多智能体系统协调所有属于该波浪场的设备的互操作性，通信和控制。最初开发了Multi-Agent系统，随后通过仿真对其进行了测试。仿真中使用了惯性海浪能量转换器，该惯性海浪能量转换器通过陀螺仪运动从海浪中提取能量来发电。该设备最初可以由比例微分控制器控制，其参数会随着海况的变化而变化。有了这个多智能体系统，使用机器学习波分类器方法识别海浪和相应的海况。信息被传递，以便所有设备都可以使用最佳控制参数。因此，由于多智能体系统与自动学习方法的结合，波场的能量提取得到了增强，而其体系结构则可以抵抗设备之间的通信问题。案例研究的仿真结果表明，平均吸收的功率有所增加，证明了将Multi-Agent系统与自动学习方法结合使用的有效性。结论表明，所提出的系统架构是原始波浪预测模型的可行替代方案。此外，该多智能体系统可适用于带有其他波浪能转换器和其他类型控制装置的波浪场。