Advanced ensemble machine-learning and explainable ai with hybridized clustering for solar irradiation prediction in Bangladesh

The solar revolution in Bangladesh stands as a symbol of hope and self-reliance, illuminating communities and steering the nation towards a more sustainable future. Highlighting the crucial importance of solar irradiance forecasting in attaining sustainable energy objectives, this study utilizes advanced machine learning techniques to enlighten solar irradiance prediction in Bangladesh. Ensemble machine learning has become a vital tool that significantly enhances the precision and reliability of solar radiation predictions by amalgamating diverse model outputs, providing a more robust and accurate forecasting frame- work. Previous research efforts have overlooked the exploration of this diverse range in solar radiation prediction using ensemble machine learning. This study addresses this gap by conducting a detailed experiment on ensemble artificial intelligence techniques. Furthermore, the research introduces Explain- able AI (XAI) with ensemble machine learning, shedding light on factors influencing predictions and providing valuable insights for decision-makers in the solar energy sector. Notably, the study pioneers an XAI-based analysis specific to Bangladesh, marking a significant stride in solar radiation predic- tion. Additionally, a novel hybridized approach incorporating various clustering techniques and the LightGBM algorithm is introduced, offering an efficient framework for solar radiation prediction. As a result, this study contributes to our understanding and optimization of solar irradiance prediction by offer- ing a comprehensive method that integrates XAI, ensemble approaches, and machine learning. We have developed an autoML tool based on XAI and Ensem- ble as a further contribution. We have validated our result with the low-code PyCaret machine learning package to see that, among all the methods, lightGBM has shown promising results in terms of solar irradiance prediction. Ensem- ble machine learning boosting techniques, notably LightGBM and CatBoost, exhibited superior performance, achieving high R² scores of 0.91 and demonstrat- ing remarkable accuracy. Through feature importance analysis, it was discerned that”Sunshine (Hours)” emerged as the most impactful factor influencing the predictive models. Employing a hybridized clustering approach, incorporating K- means-LightGBM, MiniBatchKMeans-LightGBM, Fuzzy C-Means-LightGBM, and GaussianMixture-LightGBM models, excelled in forecasting solar radiation into distinct clusters, particularly excelling in the”Very Cloudy” category.