Algorithm 1. Network Partition Algorithm based on Spatial-Temporal Features (NPA)

Input: The set of sensor nodes $X=\left\{x_i\right|i=1,\dots ,n\}$; Output: The probability distribution of model prediction $P=\left\{p_{ij}\right|i=1,\dots ,n;j=1,\dots ,m\}$;

1: Using the embedding function $\phi :X\to Z$, and the original samples are mapped into the latent feature space

2: K-Means cluster algorithm is used to initialize the objective distribution $Q$

3: While not convergence

4: fix the objective distribution $Q$ to compute $p_{ij}=f(z_i,\mathsf{\Theta })$

5: update the prediction distribution $P$

6: fix the parameters $\mathsf{\Theta }$ to compute $q_{ij}=\psi (p_{ij},\mathsf{\Theta })$

7: update the objective distribution $Q$

8: minimize $\underset{\mathsf{\Theta }}{min}-\frac{1}{n}{\displaystyle {\sum }_{i=1}^n{\displaystyle {\sum }_{j=1}^m{q}_{ij}\log p_{ij}}}$ through fixing the objective distribution $Q$

9: update the parameters $\mathsf{\Theta }$

10: End while

11: Return $P=\left\{p_{ij}\right|i=1,\dots ,n;j=1,\dots ,m\}$