Wind energy forecasting based on integration of CNN and Bidirectional RNN

The variability and intermittency of integrating large-scale wind generation pose a significant risk to the reliability and integrity of the electricity grid. Precise wind power forecasting is crucial for ensuring the reliability of wind power grid integration. This research presents a Wind Energy Forecasting (WEF) method that combines a Convolutional Neural Network with a Bidirectional Recurrent Neural Network (CNN-BiRNN). The Convolutional Neural Network (CNN) analyzes the spatial characteristics present in weather observations, extracting significant features. Meanwhile, the Bidirectional Recurrent Neural Network (BiRNN) manages the temporal relationships within the serial time-series information. By using this holistic strategy, the model can successfully represent the intrinsic temporal and spatial trends in wind energy production, leading to more precise and reliable forecasts. A refined ResNet150-based CNN architecture has been utilized to obtain profound latent characteristics that are highly representative and selective, resulting in enhanced accuracy. An assessment of the CNN-BiRNN model and the individual CNN and BiRNN models was conducted using weather data and a wind energy dataset from a wind farm. The study aimed to assess the performance of these models in multi-step WEF. The findings demonstrate that the suggested CNN-BiRNN has superior capability in extracting spatial and temporal features compared to the conventional structural model. The comparison indicates that the CNN-BiRNN model surpasses the performance of separate CNN and BiRNN models with reduced error at different time intervals, highlighting its potential to improve the accuracy of WPF.

Перевод пока недоступен