The nearest points in phase space are determined by Euclid distance in chaotic local prediction. The prediction accuracy depends on quality of the nearest points. But the shortest distance does not imply better forecasting effect. While false nearest neighboring point or high embedding dimensions appear evolvement track of some nearest neighboring point should be apart from prediction point. Because it is difficult for Euclid distance to reflect the correlation degree between the nearest points and prediction point. So the idea of combining Euclid distance with correlation degree is put forward. The method is applied to short-term electrical load forecasting. The result of load series forecasting by the presented method is more effective to improve prediction accuracy.