研究了一种基于特征变量的复杂生产过程预测模型．与传统的建模方法相比，该方法不需要经过机理分析，而从信息科学的角度出发，在对反映生产过程工况原始动态数据进行特征选择的基础上，运用时间序列分析法建立其预测模型．同时讨论了它的神经网络实现方法．仿真结果表明了该方法的可行性． A complex production process prediction model based on eigenvalue is studied. Compared with the traditional modeling methods, this method does not need to go through the mechanism analysis, but from the perspective of information science, based on the feature selection of the raw dynamic data reflecting the production process conditions, the method uses time series analysis to establish its prediction model. At the same time it discusses its neural network implementation method. Simulation results show the feasibility of the method.