I have question about dropout, predict commands, LSTM

hi. I am working with LSTM Regression sample in matlab

I got some wonder about dropout, predict commands(predictAndUpdateState , predict) , LSTM

1. How can I set dropoutrate to LSTM layer?

I don't want to use dropoutlayer. I may think dropoutlayer turns some input data to zero by it's rate. so it might print out "NaN"considering what I learned about dropout, it is about turing hidden unit(or neuron) off. then, how can I set dropoutrate to lstm layer?

2. I don't know what the difference is between "predictAndUpdateState" and "predict"

When I was working on a fine dust prediction project, I used the command "predictAndUpdateState" as I learned from the matlab example. However, when predicting the amount of indoor fine dust inflow due to the concentration of fine dust in the outside air, it was found that the forecast performance varies according to the order of the that commands. For example, when predicting indoor fine dust ingress with a pm 10 concentration of 103,58,73,22 micrometers, the predictive performance varied in the order of the case (and at that time lstm net was used,) of course when using a "resetState". So, he used the "predict" command. But the situation was similar. So I want to know some of the differences between "predictAndUpdateState" and "predict" and I want to solve the problem where thepredicted performance of the lstm network varies according to case order.

3. When performing the LSTM regression prediction, it was confirmed that NaN was output. How can this problem be solved?

4. Can lstm network be linked to simulink, as is the ANN-Control example of pathwork?

thank you.