Solar energy forecasting for grid-connected PV systems

Khosim, Muhammad Aiman (2021) Solar energy forecasting for grid-connected PV systems. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Solar energy forecasting for grid-connected PV systems.pdf - Submitted Version Download (453kB)
	Text (Full Text) Solar energy forecasting for grid-connected PV systems.pdf - Submitted Version Restricted to Repository staff only Download (3MB)

Abstract

Renewable energy sources, such as solar energy, are an excellent investment for a future source of alternative electricity. However, the inherent characteristics of solar energy created several obstacles for renewable energy management systems in terms of power output management, monitoring, and scheduling. The challenge of sustaining the photovoltaic (PV) system's output for a grid-connected solar power plant must be resolved. Therefore, this paper will provides a novel way of forecasting modelling approaches for predict the power output of a photovoltaic systems based on the daily weather of historical data that generated by UTeM solar panel. This forecasting method is required to enhance the controllability and stability of photovoltaic power system performance and to ensure the power grid operates in a reliable and cost-effective manner. A comparative performance of several forecasting systems utilizing Linear Regression Toolbox approaches have also been documented. Generally, the raw data from the photovoltaic (PV) panel are acquired and organised into historical data with the assistance of Microsoft Excel Software. Subsequently, any disinformation or null data are filter out from the historical data before incorporate into the simulation. These historical data then provided to the MATLAB Toolbox that is used to simulate the data and create the appropriate forecasting model. In the simulation, several machine learning techniques are used to develop a PV power output prediction model that can be used to predict hourly, daily and weekly PV power output using machine learning algorithms such as linear regression, Support Vector Machines (SVM), and Gaussian Process Regression (GPR). The regression learner tool in MATLAB software version R2020b is used to create, train, and evaluate the prediction models. The findings show that complex regression models such as the Exponential GPR are more accurate in the long run when compared to linear regression methods such as the Interactions Linear Regression model. Additionally, by removing multiple criteria from being trained in the regression learner have an influence on the prediction models and the accuracy of these regression models may be further enhanced.

Actions (login required)

View Item