Name: VICTOR PEREIRA FIRMES
Type: MSc dissertation
Publication date: 11/03/2020
Advisor:
Name |
Role |
|---|---|
| WANDERLEY CARDOSO CELESTE | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| DANIEL JOSÉ CUSTODIO COURA | Internal Examiner * |
| FELIPE NASCIMENTO MARTINS | External Examiner * |
| WANDERLEY CARDOSO CELESTE | Advisor * |
Summary: This work deals with the problem of identifying technically identical equipment, defined here
as highly similar equipment, monitored from a point of common coupling. An experimental
approach is given here, WHERE four fluorescent lamps and four computers highly similar
are used, and none or even all loads can be in simultaneous operation, resulting in two
datasets called dataset A, with about 8 million voltage and current lamps required by
each of the 16 possible configurations, and dataset B, with 999600 voltage and current
required by each of the 16 possible configurations of computers. Such samples are acquired
at a sampling rate of 99960 samples per second, quantized at 16 bits. The objective is to
use part of these randomly selected samples and, by means of manually and empirically
configured convolutional neural networks, to extract characteristics that allow to train,
validate and test such networks to obtain accuracy compatible with those observed in the
literature for non-similar loads. Therefore, 999600 samples from dataset A are selected,
giving rise to dataset A1_0, which is used to configure 4 distinct neural networks. Thus,
it is proposed an index to evaluate the performance of the networks that considers the
number of network parameters and the training time so that the neural network can reach
93% accuracy. The index justifies the choice of Network 2 for the training, validation and
testing stages from dataset B and 14 new banks generated from dataset A, WHERE the type
of load (dataset B) and number of samples vary (datasets A1_i, with i = 0, 1, 2 and 3),
the sampling rate (datasets A1_i, with i = 0, 4, 5 and 6) and sample resolution (datasets
A1_i, with i = 0, 7, 8, 9, 10, 11, 12, 13 and 14). The objective is to verify the robustness
of the methodology due to variations in the nature of the behavior of electrical equipment
under identification, the reduction in the number of samples and the limitations in the
acquisition hardware to sample the data in lower frequencies and resolutions. The results
show that the network maintained its performance even with loads of variable nature over
time. They also show that reducing the number of samples negatively impacts accuracy.
However, it becomes significant from a 40 % reduction in the total number used in the
network setup process. Regarding the reduction of the sampling rate, it is possible to
verify the non-compromise of the system. Finally, decreasing sample resolution causes
significant degradation when the resolution is less than 10 bits. Therefore, this work proves
that the non-intrusive method is also efficient to identify highly similar loads and shows
that the presented methodology is a viable alternative when dealing with the high cost
of identification involved, that is, the ability to obtain, store and process large masses of
data in a non-prohibitive time.
