DeepWake: Deep learning for wind farm wake modelling
Date:
In this meeting, I have introduced our lastest study in wind energy, DeepWake. DeepWake means that we use physics-informed deep learning to train a wind farm wake model. Additionally, we also expect to deeply reveal the wake generated by wind turbines using this model.
We can input the force information into the neural network and the output is the wind velocity of the field. The next question is how we collect training samples to train it. The input is force. We can use the inflow speed, turbine location, control actions to calculate the force matrix. For the output, we can use the Gaussian wake model to calculate the wind velocity. But you may notice that when we use the wind velocity generated by Gaussian model to train DeepWake, the final learned Deep Wake model is of low-fidelity. Then, we can collect a few expensive high-fidelity data from the Sowafa simulator. Using these high-fidelity data we can refine the trained DeepWake model. Now, DeepWake can learn more knowledge about the wake from the high-fidelity data.