17 - FWS-DL: forecasting wind speed based on deep learning algorithms

Abstract

Recently, the amount of electricity produced worldwide from renewable energy sources has increased significantly, with the United States leading the way. Wind speed forecasting has become an increasingly important and expanding task in electricity generation, considering the growing demand for wind resources and its penetration of the modern-day electrical power grid. Consequently, this chapter endeavors to develop a new robust hybrid deep neural network algorithm modeled with a modified version of the machine learning algorithm. A new version of a compatible deep neural network algorithm with local search capability and enhanced exploration capability is handled by the long short-term memory neural network, gated recurrent unit, bidirectional long short-term neural network (Bi-LSTM), and convolution neural network, thereby resulting in a robust hybrid optimization algorithm. The algorithm obtains optimal parameters for the proposed variant of the machine learning neural network models. The speed of water, wind, and other natural resources has been predicted by artificial intelligence models in several previous studies. These models have been used in many new studies as well. Many researchers have used a variety of state-of-the-art developments to achieve their results in forecasting wind speed, including physical and statistical models and smart heuristic models. For this project, various authenticated sources, such as Windmills in Jaipur, the Global Wind Atlas, and Data.gov.in, were used to compile the datasets, with the most notable being Windmills in Jaipur and Global Wind Atlas. Several parameters are included, including the temporal, wind speed, dew point and humidity, air temperature (temperature), the month, and pressure. With these data, it is possible to calculate density and wind speed. In addition to wind speed, air density, and blade sweep area, output power can also be determined using these variables. The design predicts the swept area, air density, and wind speed based on the input data provided to the program. Several diverse parameters have been calculated, including the root-mean-square error (RMSE) and mean absolute error (MAE). In addition to the RMSE and MAE, the mean square error (MSE) for the given algorithms is also evaluated, and the performance of the Bi-LSTM is comparatively good, with an MSE of 0.003027, RMSE of 0.055026, and MAE of 0.031164.