Introducing recursive learning algorithm for system identification of nonlinear time varying processes

Several methods have been introduced for identification of nonlinear processes via locally or partially linear models. Unfortunately, most of these methods have a training phase which should be done offline. There are phenomena that possess time varying behavior. Furthermore, the amount, distribution and/or quality of measurement data that is available before the model is put to operation may be insufficient to build a model that would meet the specification. One of the most popular learning methods in nonlinear system identification is locally linear model tree (LoLiMoT) algorithm as an incremental learning method which needs to be carried out by an offline data set. This paper introduces a recursive version of this algorithm called recursive locally linear model tree algorithm (RLoLiMoT) for time varying and online applications. The proposed method also eliminates some of the LoLiMoT restrictions in tuning premise parameters of the locally linear models (LLMs). Two case studies are considered to test the performance of the proposed method. The results depict the power of the proposed method in online system identification of nonlinear time varying systems.