A new hybrid approach to wind speed forecasting based on two-stage decomposition and wavelet neural network

The literature is frequently encountered with studies on the relationship between energy consumption and various factors such as globalisation, urbanisation, energy policies and economic growth. This relationship emphasises the importance of policies and strategies for sustainable energy systems. Renewable energy sources offer environmental benefits, reduce carbon footprint and provide energy security and independence. However, the transition to renewable energy can be costly and geopolitical risks can also affect supply chains and distribution. Diversifying renewable energy sources and regional cooperation are crucial to overcome these challenges. Renewable energy is a sustainable energy source with many benefits for society and the environment. It reduces greenhouse gas emissions contributing to climate change, improves public health, reduces water pollution, protects ecosystems, and promotes sustainable resource utilization. It also supports sustainable development and advances innovation and research in the energy sector. Renewable energy technologies improve plant efficiency and are a trend in low-carbon technologies. Wind speed is an important factor in understanding and utilizing wind energy. Accurate wind speed prediction is important for the efficient operation of wind turbines, weather forecasting, climate modelling and offshore operations. Neural networks are widely used for wind speed prediction because they can capture complex nonlinear relationships. Hybrid approaches that combine predictions from multiple models are also promising. Decomposition techniques such as variable mode decomposition (VMD) and ensemble empirical mode decomposition (EEMD) have proven to be important in wind speed prediction. Wavelet neural networks (WNNs) utilize wavelets as the activation function, allowing neural networks to capture different input data characteristics and improve performance. Although there is no direct reference to WNNs for wind speed prediction, existing studies show the potential of combining different neural network architectures with wavelet decomposition techniques to improve wind speed prediction accuracy. Our study proposes a hybrid model with secondary decomposition and a wavelet neural network for very short-term wind speed forecasting. In line with this proposal, 256 hybrid models are created with sixteen decomposition pairs in pairs with four popularly used decomposition models and sixteen activation functions, three of which are neural network activation functions and thirteen of which are wavelet activation functions. With these models, simulations were ibrit r the middle months of the four seasons, and then the other eight months of the year were simulated with 48 models with high performance in line with the results of these four months. Our results ibr that the CEEMDAN method, which is a derivative of the Empirical Mode Decomposition, the binary decomposition model created with the Variable Mode Decomposition management (VMD), and the hybrid forecasting model created with the SLOG2 wavelet activation function neural network perform high-performance forecasts.