To deeply mine the information of features in a data-driven model with high-dimensional heterogeneous data inputs, a Light Gradient Boosting Machine (LGBM) prediction model is proposed which considers the dynamic characteristics of the unit and the uncertainty quantitatively based on interval estimation. First, taking the generator rotor speed and pitch angle as the indicators of the dynamic status of the wind turbine generator, the LGBM algorithm is explored to build the prediction model. Secondly, the historical predicted power and error are divided into different groups using fuzzy C-means clustering. Then considering the conditional dependence of prediction output and error, the error probability distribution is established by nonparametric estimation, and the confidence interval is used to discretize the prediction interval of wind power. Last, a case study from an actual wind farm is conducted to test the proposed study framework. The results show that the prediction accuracy and calculation efficiency of the LGBM model considering the dynamic characteristics of unit are significantly improved, and the uncertainty quantification method based on interval estimation decouples the fitting process and prediction method. This has high reliability and flexibility. This work is supported by the National Natural Science Foundation of China (No. 51779206) and Key Scientific and Technological Project in Henan Province (No. 162102210236).