Scribd
Upload
44 views19 pages

Rope C 20 Neural Stick

This document proposes improving forecasting and classification models in smart metering systems using neural compute sticks for edge computing. It presents two models - one for energy consumption and production prediction using an RNN, and one for energy fraud classification using decision trees. Both models saw training and implementation time reductions when using one or two neural compute sticks for edge computing. The conclusions discuss how edge computing is becoming more important for IoT and smart grid applications due to increasing data volumes.

Uploaded by

Juan Carlos Olivares Rojas
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
44 views19 pages

Rope C 20 Neural Stick

This document proposes improving forecasting and classification models in smart metering systems using neural compute sticks for edge computing. It presents two models - one for energy consumption and production prediction using an RNN, and one for energy fraud classification using decision trees. Both models saw training and implementation time reductions when using one or two neural compute sticks for edge computing. The conclusions discuss how edge computing is becoming more important for IoT and smart grid applications due to increasing data volumes.

Uploaded by

Juan Carlos Olivares Rojas
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 19

Improvement of forecasting and

classification in
smart metering systems using a neural
compute stick

Juan Carlos Olivares Rojas, Enrique Reyes Archundia, José


Antonio Gutiérrez Gnecchi, Ismael Molina Moreno, Juan
Gabriel González Serna

November, 2020
Introduction

The Data Revolution

Source: IEEE
Technologies 4.0
Big Data ¿Mejora
Robotics o empeora
Digital
la seguridad?Twins
Integration
AR/VR DX: Digital Systems
Transformati
on
IoT
3D Printing

Cloud Cibersecurity
Computing Source: WebNMS
Smart Grid

Source: New Energy News


Smart Metering System

Source: PGE
SMS Applications
Solution

Edge Computing
Literature Review
Improving a SMS Architecture
Edge Computing SMS Architecture
C
A
O
N
S
U
M NCS2
P SM
T
I
O NCS2
N
P VPU
SA
R
O
D PV
U
C
T
I
O
N
WT
Consumption/Production
Prediction Model
Input Process Output
Seven variables: RNN one-to- Consumption and
date-time, active many using the Production
power, reactive ReLU activation Prediction in the
power, voltage, function next time.
current,
consumption,
and production of
two years
horizon.
Energy Fraud Classification Model
Input Process Output
Seven variables: Decision A binary tag
date-time, active Tree with the
power, reactive existence of
power, voltage, energy fraud
current,
consumption, and
production of two
years horizon.
Training Times in seconds
Model Without With one With
NCS NCS two NCS
Prediction 28,552 21,770.9 18,042

Classification 4,862.44 4,528.88 4,467.12


Implementation Times in Seconds

Model Without With one With


NCS NCS two NCS
Prediction 474.19 441.57 413.21

Classification 178.78 175 169.65


Reduction Times in Training and
Model
Implementation
With one With two Average
NCS NCS
Prediction Times 23.75 36.81 30.28
Training
Classification 6.86 8.13 7.49
Times Training
Prediction Times 6.88 12.86 9.87
Implementation
Classification 2.11 3.06 2.59
Times
Implementation
Conclusions
The use of edge computing in the coming years will
increase significantly, as IoT devices have more
significant hardware capabilities.

In particular, edge computing applications in SG


and SMS are a necessity today, mainly due to the
increasing increase in data and the fact that data is
needed closer.
Conclusions
The models that are not based on ANN are not
improved when using NCS VPU.

In contrast, models based on some ANN can be


enhanced exponentially incrementally according to
the number of NCS VPUs with which they are
available. Particularly, the training times can be
reduced significantly.
Conclusions
For some models in edge computing to work
adequately, it will be necessary to work in
conjunction with other computing paradigms such
as cloud computing and recently fog computing.

The proposed models must be tuned. In general,


classification algorithms must be
combined/assembled to achieve better
performance. The prediction model must be
improved with the use of other ANN architectures.
Questions?
This work is partially supported by the
Tecnológico Nacional de México under the grant
8000.20-P and 9002.20-P.

juan.or@morelia.tecnm.mx, enrique.ra@morelia.tecnm.mx,
jose.gg3@morelia.tecnm.mx, ismael.mm@morelia.tecnm.mx,
gabriel.gs@cenidet.edu.mx

Thanks you so much!

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy