With the enhancement of computing capacity and the development of power icing prediction model

the model can provide more rich prediction information. However

traditional evaluation methods and indicators are not objective and reasonable when it comes to evaluate the existing prediction products. To make full use of existing icing forecast products

we evaluate icing prediction products of Sichuan power grid by comparing with traditional grading methods in winter

2016—2017. The results show that the selected icing forecast products can accurately reflect the icing spatial and temporal distribution though the prediction thickness is 2～3 mm thicker than the observations. Meanwhile

compared with the traditional method

the fraction skill score（FSS） is more flexible and more instructive for the electricity production and operation due to reducing of tendency to miss alarm rates and false alarm rates. In addition

forecast products are credible and reliable in the range of 1～3 km below 20 mm thickness. Few observed samples of 20～28 mm icing data lead to low availability of forecast products. Applying FSS to evaluate the existing icing prediction products is conducive to maximizing the utilization of icing products under the existed prediction accuracy. And it is significant for the use of existing prediction products during power network operation. Moreover

the evaluation results are also feedback to prediction model to optimize the numerical model.