基于时延挖掘模糊时间认知图的化工过程多变量时序预测方法

模糊认知图(fuzzy cognitive maps, FCM）作为一种复杂系统的建模工具，能够对系统的非线性和不确定性进行处理。由于工业过程变量间往往存在着时间延迟，传统的FCM模型难以处理这类多变量的时间序列数据，建立的预测模型往往不能反映系统内各变量真实的因果关系，从而导致预测结果的解释性差、准确度低等问题。为此，提出了一种时延挖掘模糊时间认知图(time-delay-mining fuzzy time cognitive maps, TM-FTCM)，它使用互相关函数(cross-correlation function，CCF)从数据中挖掘时延信息，并通过在推理机制中添加自我影响因子和偏置及优化转换函数等参数，有效地解决了由于工业过程变量间的时延导致的预测模型不准确等问题。通过数值仿真实例及实际化工过程数据，验证了所提方法的有效性。

Chemical process multivariate time series predictions based on time-delay-mining fuzzy time cognitive maps

Fuzzy cognitive maps (FCM), as a modeling tool for complex systems, can handle the nonlinearity and uncertainty of the system. However, time-delay among industrial process variables is always ignored in traditional FCM models. The causal relationship between variables can t be calculated accurately. Therefore, the prediction result is inconvincible and unpredictable. The time-delay-mining fuzzy time cognitive maps (TM-FTCM) method is proposed to enhance the accuracy of the time-delay prediction model. The cross-correlation functions (CCF) helps to find the time-delay factors hiding in the big data, thus revealing the actual structure of the model. Furthermore, the optimization of self-impact factors, bias and transfer functions enhances the efficiency of the prediction process. The TM-FTCM method has been verified by numerical simulations and actual chemical plant process data to be efficient and practical.