The invention relates to a wind power probability forecasting method and a forecasting device. The method comprises the steps of according to a historical output power and a historical forecasting power, acquiring a statistics characteristic of a wind power station forecasting error; according to the historical forecasting power and a wind speed fluctuation amount of a NWP forecasting result of the wind power station, acquiring a wind power probability forecasting condition total set; dividing the condition total set to a plurality of condition subsets through a K-means clustering algorithm; forming condition experience distribution of an error set in each condition subset, and checking the digital characteristic of the condition experience distribution is superposed with the digital characteristic in the statistics characteristic of the wind power station forecasting error; and if yes, performing clustering through the K-means clustering algorithm; and acquiring a wind power probability forecasting result according to the wind power forecasting result at each time point and the condition experience distribution. The invention further relates to a forecasting device. The wind power probability forecasting method provided by the invention can differently supply error distribution functions and has higher forecasting accuracy.