search

OdysseasKr / online-nilm

Code for the experiments in "Sliding Window Approach for Online Energy Disaggregation Using Artificial Neural Networks"
60 stars 16 forks source link

Online nilm _ Window size #1

Closed ikrambennasrbennasr closed 5 years ago

ikrambennasrbennasr commented 5 years ago

Hello @OdysseasKr , -Does the choice of the window_size allow us to talk about real time disaggregation? -Can we say, more the size of the window is small, more we are close to the real time disaggregation phenomenon? -And if you have an idea, what's the shortest window size that we can use it in ShortSeq2Point model, to have a significant predictions? Thanks

OdysseasKr commented 5 years ago

Hi @ikrambennasrbennasr ,

  1. I would say that almost any reasonable window_size is suitable for real-time disaggregation. The main concern is that a larger window_size leads to bigger inputs to the neural network. Thus the NN requires more memory and computation time to produce results. Other than that, since we use overlapping windows, I don't see any reason why it wouldn't be used for online disaggregation.

  2. I would say that the shorter the window, the lower the predictive power. A smaller window size definitely makes computation faster

  3. "Significant predictions" depends a lot on the type of the predicted device, the complexity of the circuit etc. For the scenario that we assumed in this paper, the suggested window sizes were the ones that performed "sufficiently" well.

ikrambennasrbennasr commented 5 years ago

@OdysseasKr yes i see, thanks for your fast and helpful answers.