Day-ahead Energy Resource Management (ERM) in smart grids faces significant challenges due to high-dimensional uncertainties and risk environments. This research proposes a Grouped Importance-based Ring Cellular Encode-Decode UMDA algorithm (GIRCEDUMDA) for solving this problem. Due to the limited evaluations on expensive optimization, centralized and efficient allocation of search space is required. The algorithm grounded in the optimization potential of each group, proposes “grouped importance weights” to refine the probability estimates of population. Additionally, utilizing a cellular structure for decentralization and discretization helps reduce the search space. Experimental results demonstrate that GIRCEDUMDA outperforms the state-of-the-art algorithms, and achieve a 3.5% improvement in fitness value and a 32% reduction in standard deviation under the IEEE SSCI 2025 competition framework. This superior performance secured the algorithm first place in the IEEE SSCI 2025 competition.