Page 53 - C:\Users\am_se\OneDrive - Higher Education Commission\Desktop\FlipBook\

International Conference on

Recent Trends in Environmental Sustainability

noisy data of Pakistan’s power systems. Hence, we plan to review and evaluate, an integrated
and customized model for day ahead forecasting. The model under study is “Machine Learning
based switching model.” This model utilizes two major machine learning techniques Bayesian
Clustering by dynamics (BCD) and Support Vector Regression (SVR). The BCD classifier has
been utilized to recognize the switching of the load series and segregation has been performed
utilizing SVR.

To overcome the short comings of historical data we included various components to the
system, such as: pre-treatment of historical data, analysis, transformation, cross validation, and
over-ratting modules. The pre-processing module deals with outliers, anomalies, resolving the
missing data problems in load series, curve smoothing, and data normalization. The data was
tested defined models on electric demand and load affecting parameters data collected from
national and local distribution companies of Pakistan.

Keywords: Clustering by dynamics (BCD), Support Vector Regression (SVR), vector
machines (SVM), Load forecasting

Department of Environmental Sciences, COMSATS University Islamabad, Vehari Campus

20