Bayesian tree and differential radial network with Bi-LSTM for wind power prediction in coastal and non-coastal regions in Andhra Pradesh

• Effectively predicts Wind Power in Coastal and Non-Coastal regions. • The Tree based approximation aids in determining the Wind Speed and Direction. • Nested Ensemble Framework robustly analyses the Sea breeze dynamics. • Captures the temporal dependencies and patterns in data, for accurate predictions. • Implemented in Python Platform and performance analyzation is carried out. Wind energy is an essential renewable resource that is utilized to generate clean electricity in isolated regions, but it is difficult to harvest efficiently in both coastal and non-coastal places due to its unpredictable nature. Hence, a novel Bayesian Tree and Differential Radial Network with Bi-LSTM algorithm is proposed for addressing the wind power prediction challenges in coastal and non-coastal regions of Andhra Pradesh. Existing algorithms struggle to capture the complex dynamics of sea breezes caused by solar radiation, leading to low forecast accuracy in coastal areas due to sensitivity to initial conditions. Hence, novel Bayesian Nested Tree Approximations is introduced to predict the wind speed and direction in the coastal region. Here, Nested Integrated Laplace Approximations (NILA) and Bayesian Tree Ensemble Regression (BTER) are fused within the Bi-LSTM, which effectively captures uncertainties in initial conditions and nonlinear relationships between predictors thereby, allows robust predictions of sea breeze dynamics influenced by solar radiation. Existing algorithms for non-coastal wind power prediction fail to handle local factors such as height gradients, soil moisture content, and temperature variations, resulting in inaccurate forecasts. Therefore, to predict the wind speed and direction in non-coastal regions, a Differential Radial Activation Network (DRANet) is implemented, in which Neural Ordinary Differential Equations (NODEs) and Radial Activation Networks (RAN) are incorporated to Bi-LSTM, for capturing the spatial and temporal dynamics of environmental variables. The Bi-LSTM enhances both models by capturing temporal dependencies and patterns in data, ensuring accurate predictions of wind speed and direction over time. The results show that when compared to other existing models, the proposed model has a low Mean Average Error of about 0.521, Mean Squared Error of about 0.604, Root Mean Squared Error of about 0.743 and high accuracy of about 98%.