Algorithm 1: Co-integrated time series selection

1: TS ← sensor series

2: for each time series i do

3:   # fit a linear model with each other node j

4:   lm[i][j]← linear model TS(i, j)

5:   resd[i][j] ← residual(lm[i][j])

6: end for

7: for each residual i,j do

8:   # run Dicky − Fuller test

9:   DF [i][j] ← ADFtest(resd(i, j))

10: end for

11: for each time series i do

12:   ts ← maximum(abs(DF(i, j)))

13:   Cointegrated[i] ← ts

14: end for