Research On Short Term Electric Load Prediction Using Kernel Partial Least Square Regression

Load prediction is utilized essentially to predict the future power demand to maintain and set the power quantity for power generation scheduling in purpose of avoiding power dissipating.Load Prediction has been an influential component of providing an efficient power and it is considered as an important task in the intelligent power systems either to cover the increasing of power demand or delivering an efficient power distribution.Hence,load prediction has become one of major studies in electricity market.Nowadays researchers are derived to invent electric trains,cars,busses and other light and heavy electric vehicles those are used and touched directly by poor and rich people in some way in their daily life.different prediction models have been used due to achieve an accurate power demand prediction in purpose of saving and managing this valuable component for avoiding power dissipation and to grant an efficient power delivering to end customers.This project aimed to deal with load prediction task by applying the electric consumptions reading data of three different regions: New England,Slovakia,and Victoria Island into Kernel Partial Least Square Regression.ISO New England data are used to create an additional comparison to test the effectiveness of weather attributes(dry bulb and dew point).The data applied first without taking the weather attributes into account,the second comparative was by taken the weather attributes into considerations.Kernel Partial least square Regression is carried out by common Gaussian Kernel and Polynomial Kernel in purpose of transfiguring the original influencing features into high dimension features space.Kernel Partial Least Square Regression Prediction results have compared with other prediction models such: Linear principal Component AnalysisKernel Support Vector Regression,Kernel Principal Component Analysis-Linear Support Vector Regression,and Kernel Principal Component Regression.Models results are evaluated by three statistical: Mean Absolute Percentage Error,Root Mean Squared Error,and Normalized Mean Squared Error.The main contents of this thesis are divided as follows:(1)Electric load behavior and the most influencing attributes on short term load prediction.(2)Employing Kernel Partial Least Square Regression by applying Gaussian and Polynomial Kernel due to transfigure the original features into high dimension features space to forecast the electric load of three different regions by its technique that has extracted new uncorrelated features corresponded with robust correlated extracted object.(3)Comparing the results of proposed method(Kernel Partial Least Square Regression)with other method: Principal Component Analysis,Kernel Principal Component Analysis those used to integrate their uncorrelated extraction features into art of Support Vector Regression,and Kernel Principal Component Regression by using its own regression technique due to forecast the electric load of the three different regions.(4)Proving that the obtained prediction amounts of electric load application can be more accurate in case of not considering the weather attributes than considering them into account.