基于先验知识的混凝沉淀过程神经网络建模

Neural-Network Modeling of Coagulation Sedimentation Process Based on Prior Knowledge

结合最优化方法中的惩罚函数,把先验知识通过惩罚函数加入到神经网络的性能函数当中,从而达到在少数据样本的神经网络训练下最终所得模型更加符合先验知识的要求.文中通过对具体的混凝沉淀大时滞过程进行神经网络建模仿真,发现该方法训练所得模型可靠程度更高.文中还对约束条件的强弱和惩罚因子的关系进行了论述.

In this paper,first,by using the penalty function of the optimization method,the prior knowledge of an object is added to the performance function of a neural network,which makes the eventual trained model accord well with the demands of the prior knowledge even in the condition of less data samples.Then,a simulation with the neural network model of a real coagulation sedimentation process with large time delay is performed,the results verifying the reliability of the model trained by the above-mentioned method.Finally,the relationship between the constraint weight and the penalty factor is discussed.