reference：https://github.com/LiangjunFeng/Broad-Learning-System
for Chinese:https://blog.csdn.net/weixin_40802676/article/details/100161494
I used MNIST dataset to train and predict the bls, and got 0.93471 accuracy(withought any optimization).
link is here：https://github.com/realWang-Wei/MNIST/blob/master/DL/BLS.py