Abstract: Aiming at the uncertainty of new energy and load， the quantile gradient boosting regression tree （QGBRT） combined with chance constrained programming method based on quantile-confidence interval prediction is proposed， and a two-stage stochastic scheduling optimization model based on energy hub is established. In the form of confidence interval， the method can reflect the adjustable margin of the unit commitment from the statistical level， and provide reliable reference for intraday and real-time dispatching. In the QGBRT prediction method， pinball function describing uncertainty is used as the loss function， and prediction results are used as the constraint information of chance constrained programming. At the same time， in order to make full use of the constantly updated information and relieve the pressure of intraday scheduling plan， in the first stage of the day， according to the published information， the start and stop states of all devices and trading energy state in the microgrid are determined. In the second stage of the day， according to the updated information， the equipment outputs on the basis of the first stage are adjusted and optimized to achieve the accurate and optimal scheduling， which is applicable to actual scheduling. Finally， a typical residential area with multi-energy system is taken as an example to verify the effectiveness and economy of the two-stage stochastic programming method.

Key words: quantile gradient boosting regression tree（QGBRT）, chance constrained programming, uncertainty, energy hub, integrated energy system, microgrid