Development of a Novel Soft Sensor with Long Short-Term Memory Network and Normalized Mutual Information Feature Selection

Algorithm 1

Pseudocode of NMIFS-LSTM.

	Input: dataset imprent MT shadow
	Output: predicted value
	Begin algorithm
	Initialize
	LSTM is trained to determine network hyperparameters and network structure;
	Set F = n; S = empty set (n = number of input variables);
	Computation of NMI with LSTM;
	For i = 1:j (j is frequency of the stop criterion)
	∀ F compute I (L; );
	Find a first variable that maximizes I(L) and obtain RMSE;
	Set , set , set i = 1;
	Choose the next variable  = argmax, F−S(minS(I(, ; L))) and obtain new RMSE;
	set ; set , j = j + 1;
	if new RMSE > RMSE
	Break
	Else
	RMSE = newRMSE, return and select the next variable;
	End if
	Repeat until |S| = j;
	End for
	Retrain with selected subset
	Calculate predicted value
	End algorithm