Online System Identification and Prediction
暂无分享,去创建一个
This chapter is the first one in the sequence of four chapters demonstrating real-world applications of evolving fuzzy systems. Starting with a generic on-line system identification strategy in multi-channel measurement systems (Section 7.1), it deals with concrete on-line identification and prediction scenarios in different industrial processes, namely:
System identification at engine test benches (Section 7.1.2)
Prediction of NOx emissions for engines (Section 7.2)
Prediction of resistance values at rolling mills (Section 7.3)
As such, it should clearly underline the usefulness of on-line modelling with evolving fuzzy systems in practical usage (industrial systems). This is completed by two classical non-linear system identification problems, for which several EFS approaches are compared. Also, there will be a comparison of evolving fuzzy systems 1.) with off-line trained fuzzy models kept static during the whole on-line process and 2.) with an alternative re-training of fuzzy systems by batch modelling approaches during on-line mode. One major result of this comparison will be that re-training is not fast enough in order to cope with real-time demands. Another one is that off-line trained models from some pre-collected data can be significantly improved in terms of accuracy when adapting and evolving them further during on-line mode. In case of predicting NOx emissions, EFS could outperform physical-oriented models in terms of predictive accuracy in model outputs.