Advanced Hybrid Deep Learning Models for Solar Energy Holding Capacity Prediction in Bangladesh
Bangladesh receives a significant amount of solar energy throughout the year, making it an ideal location to produce a substantial amount of solar energy. However, in many regions, the actual amount of solar energy produced is often lower than the amount of solar energy installed due to unforeseen weather conditions and seasonal variations. To address this issue, the proposed study introduces a hybrid ensemble deep learning approach that can effectively predict the amount of solar energy that can be stored based on key environmental and weather variables. The proposed approach combines a number of powerful base learners, including Random Forest, XGBoost, Gradient Boosting, and AdaBoost, which by themselves can model intricate relationships between variables like cloud cover, temperature, humidity, wind speed, and solar irradiance with an accuracy of up to 92%. The predictions are then combined using a neural network metalearner, which guarantees improved resilience and generalization skills while increasing accuracy to 95%. A combination of regional data gathered from different parts of Bangladesh and publicly accessible solar data sets were used to train and test the model. According to the experimental findings, the hybrid model outperformed any single model by offering lower RMSE and greater R² values. By precisely predicting the availability of solar energy, this method can help engineers, legislators, and solar energy planners optimize energy storage, minimize unused capacity, and create effective and sustainable solar energy policies in Bangladesh.
