Algorithm 1: DNN_ multi-view learning method (MVL).

Input: Original Data Matrix $M_{S\times T},\omega ,\alpha ,\beta ,t$; Structure Graph $G=\left(V,E,D\right)$; Output: Final filling data matrix

1. O ← Get-All-Missing-Values ($M_{S\times T}$);

2. If there are successive missing readings ($M_{S\times T}$)

3. $M_{S\times T}$ ← Initialization ($M_{S\times T},\alpha ,\beta$);

4. For each target sensor $s_u$ at the time slot t in O

5.     $m_{ls}$ ← $UCF(M_{S\times T},t,\omega )$

6.     $m_{lt}$ ← $MD-CF(M_{S\times T},t,\omega )$

7.     $m_{gs}$ ← $IDW(M_{S\times T},t,\alpha )$

8.     $m_{gt}$ ← $SES(M_{S\times T},t,\beta )$

9.     $m_{se}$ ← $f(W_{fe}M{M}^i+b_{fe})$

10.     $mm$ ← $DNN-MVL(m_{ls},m_{lt},m_{gs},m_{gt},m_{se})$

11.     Add $mm$ into $M_{S\times T}$

12. Return $M_{S\times T}$