Modeling of power generated in wind turbine using artificial neural networks

Energy is one of the most important resources used to sustain life throughout human history, and the demand for energy has increased considerably in recent years. Humanity first benefited from carbon-based energy sources to meet its energy needs. With the depletion of carbon-based fossil fuels and the emergence of their damage to the environment, the orientation to clean energy sources has increased day by day. The fact that these resources are renewable and cause the least damage to the environment is considered as the main preferable advantage. Wind energy is one of the most widely used renewable energy sources. Before a wind-based power plant is installed at any location, feasibility studies for that location should be done in advance so that planning can be done. Basically, before installing a power plant in any location, the wind potential of the target location should be determined first. The most widely used method in determining the wind potential is measurement methods, and new estimation methods have been developed due to the disadvantages of costly and long duration. One of the most frequently used new methods today is artificial intelligence-based methods. Artificial Neural Networks (ANN) is one of the most preferred methods in this field. In this study, the wind power potential of the target turbine and the region was estimated with the help of hourly turbine data obtained from Gökçedağ Wind Power Plant belonging to Zorlu Energy group operating in Bahçe district of Osmaniye province using ANN. First, with the help of data-based processes, 8 different feature-based data groups were created to be used in ANN models, including and excluding wind speed data. Then, the hyperparameter selection process was applied in order to optimize the ANN architecture to be used in the prediction models according to the target data set. Prediction models based on the specified input groups have been developed and it has been seen that rate-based models have better estimation precision. In all four of the velocity-based models, it was observed that the power generated with R2 = 0.97 was estimated at a fairly close to reality ratio. Among these models, the MDL-6 model, in which Wind Speed and Temperature data are used together, has achieved a superior success rate compared to other models.