Thesis - Open Access
Master of Science (MS)
Department / School
Electrical Engineering and Computer Science
Short-term load forecasting (STLF) is a key issue for the operation and dispatch of day ahead energy market. It is a prerequisite for the economic operation of power systems and the basis of dispatching and making startup-shutdown plans, which plays a key role in the automatic control of power systems. Accurate power load forecasting not only help users choose a more appropriate electricity consumption scheme and reduces a lot of electric cost expenditure but also is conducive to optimizing the resources of power systems. This advantage helps while improving equipment utilization for reducing the production cost and improving the economic benefit, and improving power supply capability. Therefore, ultimately achieving the aim of efficient demand response program. This thesis outlines some machine learning based data driven models for STLF in smart grid. It also presents different policies and current statuses as well as future research direction for developing new STLF models. This thesis outlines three projects for load profile data analytics and machine learning based STLF models. First project is, load profile classification and determining load demand variability with the aim to estimate the load demand of a customer. In this project load profile data collected from smart meter are classified using recently developed extended nearest neighbor (ENN) algorithm. Here we have calculated generalized class wise statistics which will give the idea of load demand variability of a customer. Finally the load demand of a particular customer is estimated based on generalized class wise statistics, maximum load demand and minimum load demand. In the second project, a composite ENN model is proposed for STLF. The ENN model is proposed to improve the performance of k-nearest neighbor (kNN) algorithm based STLF models. In this project we have developed three individual models to process weather data i.e., temperature, social variables, and load demand data. The load demand is predicted separately for different input variables. Finally the load demand is forecasted from the weighted average of three models. The weights are determined based on the change in generalized class wise statistics. This projects provides a significant improvement in the performance of load forecasting accuracy compared to kNN based models. In the third project, an advanced data driven model is developed. Here, we have proposed a novel hybrid load forecasting model based on novel signal decomposition and correlation analysis. The hybrid model consists of improved empirical mode decomposition, T-Copula based correlation analysis. Finally we have employed deep belief network for making load demand forecasting. The results are compared with previous studies and it is evident that there is a significant improvement in mean absolute percentage error (MAPE) and root mean square error (RMSE).
Library of Congress Subject Headings
Smart power grids.
Electric power systems -- Load dispatching.
Electric power-plants -- Load.
Electric power-plants -- Load -- Forecasting -- Mathematics.
Number of Pages
South Dakota State University
Haq, Md. Rashedul, "Machine Learning for Load Profile Data Analytics and Short-term Load Forecasting" (2019). Electronic Theses and Dissertations. 3414.