本文提出一种运用人工神经网络结合正交变换的方法，即通过正交变换滤除噪声，通过交叉验证确定网络最佳构型，以充分发挥正交变换和神经网络各自的长处，避免出现过拟合，实现更准确的预报。作为一个应用实例，对初轧钢板坯样本集进行了处理并预报了不同工艺参数下的钢坯废品率。结果表明，用神经网络结合正交变换新方法可达到很好的预报效果。 This paper presents a method using artificial neural network combined with orthogonal transformation, that is, through orthogonal transformation to filter out noise and determine the best configuration of network through cross-validation to give full play to the respective advantages of orthogonal transformation and neural network to avoid Fitting, to achieve more accurate forecast. As an application example, the sample slab of the slab is processed and the billet rejection rate under different process parameters is predicted. The result shows that the neural network combined with the new orthogonal transform method can achieve good prediction results.