You are on page 1of 2

Discussion

The percent error House given the 1/3Hz disaggregation as the accepted value is as
follows:
For the 1/6Hz house3 down sampling:
Electric Heat
1.88%
lighting1
0.57%
microwave
6.52%
lighting2
7.69%
washer_dryer
1.35%
lighting
6.43%
oven
15.56%
dishwasher
0.92%
kitchen_outlets1
5.62%
bathroom_gfi
4.34%
kitchen_outlets
0.58%
stove
169.38%
refrigerator
0.05%
Then house3 down sampled to 1/60Hz yield these percent error:
Electric Heat
14.31%
lighting1
4.65%
microwave
7.98%
lighting2
15.38%
washer_dryer
25.24%
lighting
9.09%
oven
12.78%
dishwasher
6.56%
kitchen_outlets1
3.81%
bathroom_gfi
8.60%
kitchen_outlets
0% to 4 decimal places
stove
86.12%
refrigerator
20%

There was a loss of precision in 9 out of 12 appliances when down sampling from
1/6Hz to 1/60Hz. The average percent error however in 1/60Hz sampling of 16.5%
compared to 17% in the 1/6Hz sampling. This was largely due to the outlier, the
stove. Removal of the stove brings the average percent error down 4.29% for the
1/6Hz sampling and 6.42% for the 1/60Hz. This means that if a better training model
could be acquired for the stove, there would only be about a 6% error in the
accuracy of data for a 20 times smaller file size and thus 20 times less traffic
sent by the Smart Meter. I think this is a very acceptable trade off.

You might also like