Electric Load Transient Recognition With a Cluster Weighted Modeling Method

This paper considers the use of sequential cluster weighted modeling (SCWM) for electric load transient recognition and energy consumption prediction that are promising for isolating the deleterious load transients from delicate renewable sources. Two computational processes co-exist in the SCWM scheme. In the training process, we propose a cluster weighted normalized least mean squares modification of the expectation maximization method to address the singular matrix inversion problem in updating the local model parameters. For the prediction process, we propose a sequential version of the CWM prediction that not only improves the real time performance of load transient recognition, but also resolves online overlapping transients. Other real time transient processing issues are also addressed. The methods are demonstrated using benchmark electric load transients.