Paper Information

Journal:   JOURNAL OF FACULTY OF ENGINEERING (UNIVERSITY OF TEHRAN)   FEBRUARY 2008 , Volume 41 , Number 7 (109); Page(s) 911 To 926.
 
Paper: 

APPLYING TWO CONSECUTIVE CLUSTERING OF SOMS TO IMPROVE DAILY PEAK LOAD FORECASTING BASED ON FEED FORWARD NEURAL NETWORKS

 
 
Author(s):  SOROUSH A.R., AMIN NASERI MOHAMMAD REZA
 
* 
 
Abstract: 

Electrical peak load forecasting has long been of attention to researchers and academics due to its significant role in effective and economic operation of power utilities. Peak load forecasting (PLF) for the subsequent day has a fundamental role in electrical power system operation, unit commitment and energy scheduling. Developing an accurate and robust peak load forecasting methodology can lead to a more accurate forecasting of electricity consumption. Furthermore, an accurate peak load forecast can significantly reduce the cost of operating power systems. Therefore, researchers have used various techniques in the past for peak load forecasting. Some have used time-series and linear regression models in PLF. In such methodologies, the relationship between independent variables and the dependent variable (the forecast) is determined through a mathematical equation which is usually linear. However, modeling the complex correlation between the load and input variables like weather conditions and differences between days of the week makes such methods quite difficult to use. Throughout the last two decades, a great deal of research has been devoted to using artificial neural networks (ANN) for PLF. ANN techniques, due to their high capability in non-linear modeling, have found widespread use in general forecasting and particularly in PLF. This paper proposes a hybrid neural network model which combines a self organizing map (SOM) and a feed forward neural network for daily electrical peak load forecasting. Since electrical peak load is strongly influenced by weather conditions, similar consumption patterns can be observed during the year. Thus, classifying data into somewhat similar clusters can lead to noise reduction and therefore higher accuracy. Different types of classifications have been proposed in the literature of PLF. However, most of the proposed methods seem to be intuitive with no justifiable reasoning. In this paper, we propose a new approach for clustering data by using a self-organizing map. Two SOM have been used successively; the first decreases the noise resulted by the temperature differences, dividing the days of the year into five categories. Then the second categorizes the data in each cluster to reduce the noise produced by the peak differences and to distinguish weekdays. The outcome of these two SOM is the categorization of the days of a year into 12 clusters. A feed forward neural network (FFNN) has been developed for each cluster to forecast the PLF. The application of the principal component analysis (PCA) reduced the dimensions of the network’s inputs and led to simpler architecture. Some influencing factors such as temperature, relative humidity, wind speed and cloud cover were introduced into the model. The daily peak load data in this research were extracted from Tehran Regional Electric Utility Company and the proposed hybrid model (PHM) was applied to these data. To evaluate the effectiveness of the PHM, forecasting has been performed by developing an FFNN that uses the un-clustered data, hybrid recurrent neural networks which consider only temperature, feed forward neural networks which use cluster data, and linear regression models. The results proved the superiority and effectiveness of the PHM.

 
Keyword(s): CLUSTERING, FEED FORWARD NEURAL NETWORKS, LINEAR REGRESSION, LOAD FORECASTING, ORGANIZING MAPS, PRINCIPAL COMPONENT ANALYSIS, SELF
 
References: 
  • ندارد
 
  Persian Abstract Yearly Visit 28
 
Latest on Blog
Enter SID Blog