There are many changes in the way BLS links in each layer. Through structural changes, BLS can utilize its advantages of flat structure and incremental learning better.

Here, the framework of several BLS variants with their mathematical modeling is given [1].

In addition, a novel recurrent BLS (RBLS) [2] is proposed based on the typical BLSs. The nodes in the enhancement units of the BLS are recurrently connected, for the purpose of capturing the dynamic characteristics of a time series.
Related articles are listed below.

[1] Universal Approximation Capability of Broad Learning System and Its Structural Variations

[2] Recurrent Broad Learning Systems for Time Series Prediction