The forecast of solar flares is an

The forecast of solar flares is an important task in the space science research. Both the shape and evolution of magnetic fields in active regions are closely related to whether or not to produce flares, but many current studies have not fully taken into account the two information. In this paper, we use the 3d convolutional neural networks (3D CNN) to utilize the information, and builds two forecast models to predict flare outbreaks with ≥ C- and ≥ M-level, respectively.We collected the data from May 2010 to December 2019 obtained by the Space-weather Helioseismic and Magnetic Imager Active Region Patches line-of-sight magnetogram of the Joint Science Operations Center for training and testing the model. The index called true skill scores (TSS) is used to evaluate our model performance.The TSS values of the forecast models, with ≥ C-- and ≥ M--level, reach 0.732 ± 0.028 and 0.762 ± 0.064, respectively. This proves that our proposed forecast model is effective.