Abstract: To solve the problem that some important parameters in chemical process are difficult to measure on-line, a soft sensor modeling method based on time series neural network for multirate system is proposed. A dynamic interpolation neural network (DINN) is established according to the proposed new method, and the enhanced particle swarm optimization (EPSO) algorithm is applied for the network parameter optimization. Efficiency of the proposed method has then been verified through experiments, the online estimate of output variables has been realized , and the training and generalization properties of the network have been analyzed. The proposed model based on EPSO-DINN has been proved having a better performance than normal BPNN.

摘要： 针对某些化工过程关键变量难以在线测量的问题，提出了一种基于多采样率系统的时间序列神经网络的软测量建模方法，建立了动态插值神经网络模型，并利用增强粒子群算法实现了网络参数的优化。将此方法用于实验室模拟建模，实现了变量的在线预估，并对网络的训练效果和泛化性能进行了分析，表明其建模效果明显优于普通静态神经网络。