Electricity Theft Detection in Smart Grids Using a Hybrid BiGRU–BiLSTM Model with Feature Engineering-Based Preprocessing

. 2022 Oct 14;22(20):7818. doi: 10.3390/s22207818

Algorithm 1: Bi-GRU- and Bi-LSTM-based Detection Scheme.

1
Step 1:
2
Input: Benign Consumers $B_{C}$ , Output: Fraudulent Consumers $F_{C}$
3
Step 2: Generating Theft Samples
4
$T 1$ = $B_{C}$ $* r a n d o m (0.1, 0.9);$
5
$T 2$ = $B_{C}$ $* x_{t}$ where $(x_{t} = r a n d o m (0.1, 0.9));$
6
$T 3$ = $B_{C}$ $* r a n d o m [0, 1];$
7
$T 4$ = $m e a n$ ( $B_{C}$ ) $* r a n d o m (0.1, 1.0);$
8
$T 5 = M e a n (S)$ for each column;
9
$T 6 = S (T) - t$ revesing a time sequence;
10
Step 3: concatenation
11
Concat ( $B_{C}$ + $F_{C}$ );
12
Step 4: Balancing Data
13
$B_{C}$ = $F_{C}$ ;
14
Step 5:
15
$S_{i t h}$ of majority class having smaller EU Distance with decision boundary is removed;
16
Step 6: Data Leakage
17
$p (s)$ = $C_{i}$ + $C_{j}$ ;
18
$C_{i}$ ⊆ $p (s)$ ;
19
$C_{j}$ ⊆ $p (s)$ ;
20
$S_{j 1}, S_{j 2}, S_{j 3}, \dots, S_{j n}$ $ε$ $C_{j}$ ;
21
$S_{i 1}, S_{i 2}, S_{i 3}, \dots, S_{i n}$ $ε$ $C_{i}$ ;
22
$S_{i}$ ∉ $S_{j}$ ;
23
$C_{i} (S_{i 1, \dots, n})$ ∉ $C_{j} (S_{j 1, \dots, n})$ ;
24
Step 7: Feature Engineering
25
$F 1$ = $M e a n$ of $P_{s}$ against each row;
26
$F 2$ = $S t d$ of $P_{s}$ against each row;
27
$F 3$ = $M i n$ $\in$ $C_{i}$ against each row;
28
$F 4$ = $M a x$ $\in$ $C_{j}$ against each row;
29
Output: Honest Consumers $ε$ $B_{C}$ , Fraudulent Consumers $ε$ $F_{C}$ .