Discrete data stream has the characteristics of high uncertainty, changeable evolution trend and multiextremum,which makes it difficult to achieve accurate and realtime forecasting.Therefore, a new forecasting method based on the online segmentation algorithm of discrete data stream is proposed.The proposed method combines the shortterm trend extraction and online adaptive detection of segmentation points.Before forecasting, the shortterm trend is obtained based on the nonparametric regression model and the improved trend extraction algorithm, then the shortterm trend of the forecasted day is mined and analyzed for later online forecasting.In the online forecasting stage, the adaptive detection of segmentation points based on hypothesis testing is applied to the online data stream, which can solve the problem of determining segmentation points.The forecasting model at the segmentation point is then modified based on the shortterm trends, which reduces the dependence of the forecasting model on the buffered data.Finally, numerical results illustrate the validity and feasibility of the proposed method.