halong bay tour
0 votes
in Education by
Abstract 
This paper proposes a Matlab object oriented application based on Kohonen Self- Organizing Maps (SOM) able to classify 
consumers’  daily load profile. Firstly, the characteristics of Kohonen self-  organizing maps are briefly described in order to 
underline the advantages and disadvantages of these types of neural networks in classifications approaches. In the second part, 
data used for classification of load daily profiles is processed using statistical  methods and Matlab. The result of these 
computations is a data base composed of daily load profiles used for SOM training. In the third part, the proposed software is 
tested on several scenarios in order to classify different consumers’ load profiles. 
© 2014 The Authors. Published by Elsevier B.V. 
Selection and peer-review under responsibility of the Organizing Committee of ITQM 2014. 
Keywords: Kohonen, Self- Organizing Map, neural networks, load, classification; decision support system 
1. Introduction 
In  a smart grid with power generation from renewable energy sources, a lot of the human interactions, that 
presently try to manage system operation, will be able to be replaced by machines that have a faster response time 
and can process larger quantities of data because during real-time operations, the generation and the load must be 
matched. Typically, the amount of the load determines the amount of energy supply from power plants and depends 
on a number of factors, including time of the day, day type (weekday or weekend), temperature, humidity, season 
and location.  
© 2014 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license 
 
Selection and peer-review under responsibility of the Organizing Committee of ITQM 2014.
The conventional generation can be controlled to a large extent, but the load and the renewable energy generation 
must be estimated. Furthermore, a  load forecasting is still more critical where the dependence from weather-
depending renewable generation is particularly pronounced (micro-grids). For this kind of problem, various factors 
should be considered, such as weather data or, more in general, all the factors influencing the load/consumption 
pattern. This means that for an accurate load forecasting, exogenous variables may be considered and they differ 
according to customer type: residential, commercial and industrial. 
The ability of a smart grid to process this kind of information could result in significant improvements in the 
operation of renewable energy resources
1
. A solution that could help to make  the difference between users, in this 
context, is represented by integration  computational intelligence techniques in smart grids. Thus, these ones will 
provide intelligence to the smart grid and will help power system to optimally schedule or dispatch its resources. In 
addition, because of the inherent complexity and uncertainty between the historical data and forecasts, such 
techniques became promising solutions to deliver the expectations of a smart grid, must be fast, scalable and 
dynamic.  
Over the past several decades, different conventional techniques, such as time series analysis, regression analysis 
and other statistical methods, have been attempted to tackle the problem of load forecasting. However, deregulated 
energy markets have presented new challenges, requiring  more information which is forecasting dependent. 
Therefore, the corresponding development and maintenance efforts for dealing with hundreds of irregular data series, 
which need to be simultaneously forecasted for security and economic analyses, means that the parametric models 
are beyond practical consideration.  The relationship between the electricity load and its exogenous factors is 
complex and nonlinear, making it quite difficult to represent using linear models, or even parametric nonlinear ones. 
Thus, besides the limited accuracy, most of these traditional models could not be easily adapted to different utilities 
and nonlinear modeling of the time varying dynamics of an electric power system is still a challenge using classical 
techniques
2
.  
To overcome these problems, researchers tried other methods based on different techniques such as: time series 
analyses, fuzzy logic, neuro-fuzzy method, artificial neural network, and support vector regression have been 
proposed.  
Between these ones, neural networks (NNs) have been  shown to be promising tools for load predictions
3
. The 
NNs models are complex and difficult to understand, and are often over-fitted. Indeed, their structure is sufficiently 
opaque that it is not clear why they should forecast as well as they do. In addition, in practice it isn't easy to apply 
NNs due to the lack of efficient procedures to obtain training data and specific knowledge. So far, most of the 
applications in the literature just use experimental data for model training. Thus, data-driven approaches are highly-
dependent on the quantity and quality of system operational data. 
Considering all these aspects, in this paper we propose a Matlab object oriented application, based on NNs, 
precisely Kohonen Self- Organizing Maps (SOM), able to estimate consumer’s daily load behavior. The application 
is focus on NNs because their efficiency in the area of load approximation was demonstrated in our previous works
 

 

Please log in or register to answer this question.

You are using Adblock

Our website is made possible by displaying online advertisements to our visitors.

Please consider supporting us by disabling your ad blocker.

I turned off Adblock
...