Keras data normalization - how to use with different data?

I have some question about data normalization and best practices with normalization when input data contains different data with different min-max values.

So for example, I have dataset contains 3 vectors:

[
    [0 ... 1],
    [-100, 100],
    [0, 3333]
]

The vecotors contains min and max values met in train dataset.

In this case I would use some simple MinMaxScaler to standardize the impact of each input.

BUT:

What if I know that in test dataset some of above values can be bigger/smaller? So e.g. in train dataset we have above dataset and in test/validation dataset we will have dataset like below:

[
    [0 ... 1],
    [-100, 100],
    [0, 6666]
]

In this case using MinMaxScaler will work really bad and will be the cause of the malfunctioning network.

What in this case? Is Keras offer any solution for this purpose? Or maybe to use some adaptive method?

Or maybe the simples solution:

Set up some max value for the third vector as for example 10 000? ((while scaler fitting) assume that the value will never exceed the limit) But yeah, in this case network will never see the "max" value of input during training...

Have a nice day!