regressor = Sequential()
regressor.add(LSTM(units=32, return_sequences=True, input_shape=(X_train.shape[1],1)))
regressor.add(LSTM(units=32,return_sequences=False))
regressor.add(Dense(units=1))


Layer (type)                 Output Shape              Param #
=================================================================
lstm_3 (LSTM)                (None, 60, 32)            4352
_________________________________________________________________
lstm_4 (LSTM)                (None, 32)                8320
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 33
=================================================================
Total params: 12,705
Trainable params: 12,705
Non-trainable params: 0

for i in range(240)
  1. Select the last 60 values (my timestep)
  2. Use sequence generated in step 1 to predict the new value.
  3. Remove [0] item on my sequence generated in step 1, and push the value that I generated in step 2 to my sequence.

sequence_timestep = 60
last_sequence_train = X_train[-1]

predictions = []

def sequence_constructor():
    if len(predictions) >= sequence_timestep:
        new_sequence = predictions[-dimension_seq:]
    else:
        splitter = sequence_timestep - len(predictions)
        part_1 = last_sequence_train[-splitter:]
        new_sequence = np.append(part_1,predictions) #Concatenate 2 list
    new_sequence = np.array(new_sequence)
    return new_sequence

for i in range(240):
    new_sequence = sequence_constructor()
    new_prediction = model.predict(new_sequence)
    predictions.append(new_prediction)