Development of a deep learning and discretization based adaptive hybrid forecast model for wind speed forecasting

Estimating the wind speed correctly and reliably plays a key role in managing and operating wind energy power systems. Therefore an novelty adaptive estimation model (NAEM) combined with deep learning-based mode discretization has been developed for use in wind speed estimation in this study. This developed model consists of the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), the continuous wavelet transforms (CWT), the contrast limited adaptive histogram equalization (CLAHE), the particle swarm optimization (PSO), and convolutional neural network (CNN). Adaptive estimation model using decomposition method was presented as an alternative to the traditional data matrix transformation used in data preprocessing stage. Thus, the usage of this model for the first time in the data preprocessing stage and the creation of a hybrid structure by combining the methods included in this method for the first time constitute the most important innovative aspect of the study. Proposed model (NAEM) was tested in different case studies and RMSE, MAPE, and R2 were used as performance metrics. In the comparison with commonly used deep learning models (CNN-RNN, GRU, LSTM) the root means square error (RMSE) values decrease by 25.80%, 61.17% and 63.60% respectively. In addition, the power density value of the actual wind speeds was approached by 95.1% with the proposed NAEM.