TensorFlow convert predicted value

Question

I recently started learning TensorFlow and followed the tutorial on https://www.tensorflow.org/tutorials/structured_data/time_series using the RNN Model. Everything seems to work fine and now I want to use the Network to actually predict. For testing I thought about using it on the test dataset, which contains my own data (same input format as used in the tutorial). I used lstm_model.predict(wide_window.test, verbose=0) and it outputs following Tensor:

[[[-0.496009111]
  [-0.493492275]
  [-0.512232]
  ...
  [-1.01290524]
  [-0.913285673]
  [-0.867954075]]

 [[-1.26170385]
  [-1.33223891]
  [-1.26496506]
  ...
  [-0.424162567]
  [-0.400129378]
  [-0.373054951]]

 [[-0.0474848524]
  [-0.0308368355]
  [-0.0199640915]
  ...
  [-0.803811669]
  [-0.763414]
  [-0.717713535]]

 ...

 [[1.41174877]
  [1.31383634]
  [1.28738451]
  ...
  [0.0934850946]
  [0.0987800434]
  [0.155635804]]

 [[-0.662865698]
  [-0.695823371]
  [-0.629942596]
  ...
  [-0.0201108307]
  [0.0546201393]
  [-0.0362461507]]

 [[1.82424331]
  [1.93187594]
  [1.79579926]
  ...
  [0.503997624]
  [0.546503782]
  [0.514201224]]

]

How do I get the predicted value and "remove" the normalization, used on the train dataset as seen in the tutorial, so I can predict a single temperature value?