Multi-State Energy Classifier to Evaluate the Performance of the NILM Algorithm

. 2019 Nov 28;19(23):5236. doi: 10.3390/s19235236

Algorithm 4 Energy estimation penalty (EEPenalty)

Input:

G_{T} = {g_{1}, g_{2} \dots g_{t}}

is the ground truth of appliance m

P_{T} = {p_{1}, p_{2} \dots p_{t}}

is the predicted values of appliance m

C_{S t a t e}

is the clustered operational state data of appliance m

G_{S l i c e}

is a vector of all

I R

P_{S l i c e}

is a vector of all

I R

Output:

\sum_{i = 1}^{T} E E_{i}^{(m)}

is the total Energy estimation Penalty for appliance m

1:
Init $k, l = 1$
2:
for $t = 1$ toTdo
3:
Obtain data point $g_{t}$ and $p_{t}$
4:
if $g_{t} > 0$ and $p_{t} > 0$ then
5:
Compute $C_{g} =$ closestCluster $(C_{S t a t e}, g_{t})$
6:
Compute $C_{p} =$ closestCluster $(C_{S t a t e}, p_{t})$
7:
Obtain $C_{g_C_{R a t e}}$ and $C_{p_C_{R a t e}}$
8:
Set $k = t$ and $l = t$
9:
while ( $(\frac{| g_{t + 1} - g_{t} |}{g_{t}} < C_{g_C_{R a t e}})$ and $(\frac{| p_{t + 1} - p_{t} |}{p_{t}} < C_{p_C_{R a t e}})$ ) do
10:
Add data point $g_{t}$ to $G_{S l i c e}$
11:
Add data point $p_{t}$ to $P_{S l i c e}$
12:
Increment l
13:
end while
14:
Set $t = l + 1$
15:
$P_{j w} =$ Call ComputePenalty( $G_{S l i c e}, P_{S l i c e}$ )
16:
Call AssignPenalty( $G_{T}, P_{T}, P_{j w}$ )
17:
end if
18:
end for
19:
return $\sum_{i = 1}^{T} E E_{i}^{(M)}$
20:
21:
Procedure ComputePenalty $(G_{S l i c e}, P_{S l i c e})$
22:
Compute $J_{w} (G_{S l i c e}, P_{S l i c e}) = \frac{\sum_{i} m i n (g_{i}, p_{i})}{\sum_{i} m a x (g_{i}, p_{i})}$
23:
Compute $P_{j w} = (1 - J_{w} (G_{S l i c e}, P_{S l i c e}))$
24:
EndProcedure
25:
26:
Procedure AssignPenalty $(G_{T}, P_{T}, P_{j w})$
27:
for $i = k$ toldo
28:
if $((g_{i} > 0) and (p_{i} > 0) and (\frac{| p_{i} - g_{i} |}{g_{i}} > C_{g_C_{T h r e s}}))$ then
29:
Assign penalty $E E_{i}^{(m)} = P_{j w}$
30:
else
31:
Assign penalty $E E_{i}^{(m)} = 0$
32:
end if
33:
end for
34:
EndProcedure