Scribd
Upload
2 views34 pages

Load Balancing Optimization Presentation-2

The document discusses load balancing optimization in LTE cellular networks using deep learning techniques for time series forecasting. It outlines the methodology, including the use of Long Short-Term Memory Networks (LSTMs) for predicting traffic and the implementation of a load balancing simulation based on forecasted data. The results indicate improved accuracy over baseline models and highlight the importance of automated load balancing for future network management.

Uploaded by

ahmedessam890915
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
2 views34 pages

Load Balancing Optimization Presentation-2

The document discusses load balancing optimization in LTE cellular networks using deep learning techniques for time series forecasting. It outlines the methodology, including the use of Long Short-Term Memory Networks (LSTMs) for predicting traffic and the implementation of a load balancing simulation based on forecasted data. The results indicate improved accuracy over baseline models and highlight the importance of automated load balancing for future network management.

Uploaded by

ahmedessam890915
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/ 34

.

dd
Load Balancing Optimization in LTE Cellular Networks
Using Deep Learning Techniques for Time Series
Forecasting

Prepared By:

Ahmed Alaa Mohammed


Ahmed Essam Salah
Amir Elmomneen Saber Elsid
Mohammed Ayman Mohammed
Mohammed Fawky Mohammed
Mohammed Tarek Hosny
Tahany Mohammed Khodier

Under the Supervision of: Sponsored by:


Dr. Mohamed Yasin Ibrahim Afifi Vodafone
H.O Definition
occurs when a device
moves from the cell
coverage serving it
towards another:
a process where the user
established session must not
be interrupted due to this cell
change
H.O
PROCESS
Artificial Intelligence
MACHINE LEARNING
AND DEEP LEARINING

Artificial
Intelligence

MACHINE
LEARNING

DEEP
LEARINING
MACHINE LEARNING Machine learning is a subset, an application
of Artificial Intelligence (AI) that includes algorithms that parse data, learn from that data,
and then apply what they have learned to make informed decisions without the need for
explicit programming

Based
AI Types
on
supervised classification

MACHINE Semi
LEARNING regression
supervised

clustering
Unsupervised

dimensionality
Time series
A sequence of data points
organized in time.
What is Long Short-Term Memory Networks
(LSTMs)?
Chapter 4
Phase One Forecasting the Traffic of an LTE Network
and Comparison

● Actual data
● Prediction train
data
● Prediction test
data
1- What is important libraries?
HOW TO
IMPLEMENT
THE MODEL IN
AI ?
the train data vs the predicted model data

Shows the test data vs the predicted model data


The Results

Base Line Model accuracy = 96.63 %


Model accuracy = 98.03 %
❏ Roadmap for the task

❏ Cleaning the data.


❏ Plot elbow method to detect optimal number of k ( k = 4 ).

❏ Divide the data into 4 clusters.


❏ Preprocessing.
❏ Train the Model.
❏ Test the Model and Compare Results.
❏
Anomaly Detection.
❏
Summary of our results.
Phase Two Forecasting the Downlink Traffic of Multi- Cells and
Comparing the Results with the Persistence Model
CLUSTERING

Divide the data into 4 clusters.


Test the Model and Compare Results.

Cluster number 0

Model_accuracy = 96.4%
❏ Test the Model and Compare Results.

Cluster number 1

Model_accuracy = 94.07%
❏ Test the Model and Compare Results

Cluster number 2

Model_accuracy = 89.2%
❏ Test the Model and Compare Results.

Cluster number 3

Model_accuracy = 91.01%
Defination

Anomaly
Detection

Importance
Algorithms

Anomaly
Types
➢ Anomaly detection refers to the process of
identifying patterns or data points that deviate
significantly from the expected behavior within a
given dataset.

❖ Importance
● Cybersecurity
● In the financial sector
● Manufacturing and Industrial
Processes
❖ Types of Anomalies:

● Point anomalies
Single data points that
are significantly different
from the rest

● Contextual anomalies
Data points that
are anomalous within a specific
context or subset of the data.

● Collective anomalies
Groups of data points that
together exhibit anomalous behavior,
even if individual points are not anomalous.

es of Anomalies:
❖ Techniques for Anomaly Detection:
● Statistical Methods : These methods assume that the data follows a specific
statistical distribution and identify anomalies based on deviations from that
distribution.

● Machine Learning Approaches : Utilize algorithms like clustering,


classification, and dimensionality reduction to learn patterns from the data and identify
anomalies based on deviations from those patterns.

● Time Series Analysis: Focuses on detecting anomalies in sequential data by


analyzing temporal patterns and deviations from expected behavior.
Anomaly types:
Detect Anomaly Cells
Load Balancing Simulation
•Using the Forecasted DL Traffic Volume in Load Balancing
One approach that we took was to build an algorithm that its input is a group of neighbor
cells and their forecasted traffic of a future day. The output of this code is a list for each
cell informing it with the candidate cells where it can offload some UE in case any
overload occurrence in this specific future day and this list is updated each day based on
the forecasted traffic volume.
Example
•Let’s take 5 neighboring cells as an example. The cells are called 4G_BINTUMANI-1, 4G_BINTUMANI-
2, 4G_BINTUMANI-3, 4G_BINTUMANI-6 and 4G_BINTUMANI-7 for short let’s name them A, B, C, D
and E respectively. It was forecasted that on the 10th of March 2018 the DL Traffic Volume, in GBytes,
was 47.729093, 23.962354, 24.432021, 71.296806 and 153.880125 so clearly load is not balanced
here.
• Each cell will receive its own list of candidate cells in case an unexpected load occurred on cells A, B
and C or incase the load forecasted is already high. So, for cell E, the mean traffic volume of its
neighbor cells and their median is calculated and added together then halved, this is the threshold.
So, if any cell has traffic lower than the threshold, then it is added to the list.
Simulation
• Using MATLAB we implemented 3 cells, cell A, B and C with radius of 500 meters and
centers (0,0), (600,0) and (0,600) respectively
.

• These cells were chosen from our database of phase two cell level and were assumed to be neighbors as displayed
and according to their DL traffic volume (GBytes). The maximum DL traffic volume was mapped to the maximum
number of active UE that can be served at the same time in LTE which is 100 users. Cell A has the maximum traffic
volume which is 107.3908 GBytes while cell B has 21.8813 GBytes which is mapped to approximately 20 active users
and cell C has 34.233 GBytes which is mapped to 32 active users.
• The users are randomly distributed around the coverage area. Some users are near the eNB while others are on the
edge of the cell in the overlapped area.
● The UEs served by cell A in the overlapped area are our target users that will handover as they are
receiving less power and since power is inversely proportional with the distance squared in the free
space path loss model and the power transmitted is assumed to be the same for all cells with no
power control then we can model the handover threshold as a function of distance.

.
The new coverage area of cell A

.
Conclusion
The operating conditions in the network vary with demand of the subscribers resulting in
unexpected loading on resources of the network with sudden increase in usage of mobile data.
Implementation of load balancing helps to offload excess traffic. A preliminary step to achieve load
balancing is time series forecasting which is done using a seq2seq LSTM model. The forecast is
divided into two phases namely network level and cell level. The forecast results are compared to
a baseline model (persistence model) and do surpass it. Then, based on those forecasts, a load
balancing simulation is conducted using MATLAB. Nowadays, the process of load balancing
optimization is handled manually by radio engineers. Load balancing is needed for improving
efficiency. Accordingly, the development of the methods of load balancing will cause the planning
and management of the future generations much easier.

You might also like