A Machine-Learning Based Nonintrusive Smart Home Appliance Status Recognition

Pseudocode 3

Training and disaggregation process of the RNN (LSTM) (mdl2)

(i)	Acquire the aggregate and ground truth signals for the selected parameter (p)
(ii)	Define: no. of LSTM memory cells, parallel MLP networks (MLPn) connected in series with LSTM layer, hidden layers in dense parallel layers (dsp), dense layers (ds) connected to series combination of LSTM + MLPp, epochs, and early stopping condition
(iii)	Initialize all weights
	While parameter (p) is True: do(1)Epochs = 0 (train)(i)Iterate through LSTM(ii)Iterate through LSTM + each MLPn arm(iii)Concatenate all MLPn layers(iv)Iterate through each ds(v)mdl2 converged ⟶ stop(2)Disaggregation(i)Pseudocode 1(3)mdl2 not converged(i)epochs = epochs + 1(ii)Return [(1) (i)]
	Otherwise: do(4) (next parameter): do (2) to (3).(5)Train and disaggregate all parameters:(i): return [(4)](ii): Exit