<p>This study proposes a hybrid improved butterfly optimization algorithm-support vector machine (SVM) to address the nonlinear and nonstationary characteristics of short-term wind power signals caused by uncertain wind speed factors. (1) This study proposes the Levy flight strategy after each iteration to improve the optimization performance of the butterfly algorithm because the dynamic switching probability strategy and adaptive weight are considered in the traditional butterfly algorithm; (2) the influence of different meteorological factors on wind power output is analyzed, and the input features of the short-term wind power prediction model are determined; and (3) the short-term wind power prediction model is applied to predict the wind power in different seasons and compared with existing prediction methods. High-precision wind power prediction is the solution for promoting the exploitation and utilization of wind. Large-scale wind power grid connections lead to challenges for the safe operation of power grids. The testing results reveal that the proposed improved butterfly optimization algorithm-SVM model improves the short-term wind power prediction accuracy, with a mean value of less than 0.21 compared with those of the butterfly optimization algorithm-SVM, particle swarm optimization-SVM, genetic algorithm-SVM, and back propagation neural network models. High-precision short-term wind power prediction can be used to establish a reasonable economic dispatch plan for power systems and improve the economic benefits of wind farms.</p> Graphical abstract <p></p>