Caracterización del régimen de viento y predicción de velocidad de viento y energía eólica a corto plazo, utilizando redes multivariable lstm y narx en la cordillera de los Andes, Ecuador

Resumen

In recent years, research has revealed that hydropower designs in Ecuador have not adequately considered sensitivity to climate change. Furthermore, climatic conditions determine the variations in electricity generation from this renewable energy source. Moreover, variations in rainfall patterns cause a drought from July to October due to the lack of rain. Therefore, it would decrease the flow of the rivers that feed the dams for hydroelectric generation, which would result in a significant reduction of its generating capacity. In previous years, the drought caused an increase in thermal generation with diesel engines, generating millions of tons of greenhouse gases that were emitted into the atmosphere. To prevent this inconvenience from occurring, it is necessary to promote the use of wind energy to diversify the Ecuadorian energy matrix. This matrix is made up of 70% hydropower and 0.26% wind power. However, the integration of wind energy into the electrical grid produced by high-power wind farms located in mountainous areas is a difficult task due to the variability of the wind. In addition, the Ecuadorian Andes have significant untapped wind potential due to their complex orography. Currently, there are no detailed studies on wind potential or wind energy prediction in the Andes. As a result, wind resource characterization is necessary. As well as, precise methods are required to predict the amount of wind energy generated in the short term every day, which is done to integrate it into the electrical grid. The grid operator needs the hourly power forecast to schedule and manages renewable generators and other stable generators to keep the grid balanced and power distribution efficient. In order to address the challenges mentioned above, promote wind energy development, assess wind potential, and diminish the need for thermal generators. This thesis describes a methodology to implement a hybrid model based on linear regression models as a baseline for WS forecasting and Dynamic Neural Networks and Recurrent Neural Networks (DNN-RNN) to optimize prediction and Wind Resource Characterization in the Ecuadorian Andes to install a potential wind farm, which represent a relevant contribution in this research scope. The proposed methodology allows for wind resource characterization, wind speed forecasting, wind energy prediction, and estimating energy costs six hours in advance, both for winter and summer. It also favors the modeling of wind characteristics through the Ansys Computational Fluid Dynamics (CFD software for the positioning of 11 Goldwind 70/1500 KW wind turbines to optimize the Annual Energy Production (AEP) of a hypothetical wind farm. The proposed WS forecasting model was trained and validated using data measured by two meteorological towers installed in the mountainous study area. The studys main findings indicate that the wind passing between two volcanoes has a high wind potential. This potential is dependent on meteorological variables, orography, and the accelerating effect of wind speed. These favorable conditions make it possible to install a wind farm in this area with 11 high-power wind turbines. Fur- thermore, the wind farm design using Ansys CFD showed that the K-epsilon model can model the profile of wind speeds and Turbulence Intensity (TI) over a simulated mountain with great precision. Moreover, the Long-Short Term Memory (LSTM) network, due to its embedded memory cell that allows remembering previous states to predict future values, reached the lowest values of Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) compared with the Nonlinear Autoregressive network with Exogenous Input (NARX), to predict wind speed and wind energy in summer at the height of 80 m Above Ground Level (AGL), working with nominal wind speeds with low Turbulence Intensity (TI), which allow continuous operation of wind turbines at nominal power.