现有拉漏预报方法经常会发生误报，本文尝试了一种基于神经网络的漏钢预报方法，它采用神经网络进行模糊模式识别和预测拉漏事故．实验表明该方法能比原有方法更快速准确地检测出铸坯粘结和裂缝等缺陷，可有效预防连铸中的漏钢事故． The existing methods for the prediction of false positives and false positives often generate false positives. In this paper, we try to find out the method of breakout forecasting based on neural network, which uses neural network to recognize the fuzzy pattern and predict the leakage accident. Experiments show that this method can detect defects such as bond and crack of slab faster and more accurately than the original method, which can effectively prevent the steel from breaking accident in continuous casting.