Traditional database cost models work well with small datasets but struggle as table sizes and joins increase, often leading to suboptimal query plans. Neural network-based optimization methods face challenges like high training costs, uncontrollable tail latency, and lack of interpretability, limiting their practicality. To overcome these issues, an adaptive query optimization method based on the elite annealing algorithm (EA-AQO) is proposed. This method uses the elite annealing algorithm to select elite solutions, expanding search scope and efficiency in each cooling step. A customized random forest regression model with uncertainty measurement is used as the fitness evaluator, reducing query execution overhead while improving accuracy and robustness. EA-AQO enables adaptive query optimization in traditional databases, allowing automatic learning and selection of better query plans, significantly improving database efficiency, ease of deployment, and reducing execution costs.