Interval forecasting on renewable power generation
Date of Issue2019-05-05
School of Electrical and Electronic Engineering
This report aims to predict the solar irradiance with the combination of point value and the interval value. Algorithms in machine learning were used. Solar irradiance dataset was collected from the solar radius situated at NUS Geography Weather Station and it was further divided into training dataset and testing dataset. Gradient Descent and Long Short-Term Memory were used to predict point value then generating the prediction intervals based on the probabilistic analysis of training error. Later, evaluations were made to measure the point values and prediction intervals. By comparing the results in Gradient Descent and Long Short-Term Memory, the importance of tuning parameters was revealed. Furthermore, while Gradient Descent had clearer relationships between parameters and final results, Long Short-Term Memory had more complicated layers to process sequence data. While Gradient Descent and Long Short-Term Memory in this report both provided reasonable results for prediction intervals, there is a trade-off between PICP and Interval Score. Hence, in the future work, coordinated evaluation of PICP and Scores should be worked out to find out an optimal balance between the two metrics.
DRNTU::Engineering::Electrical and electronic engineering::Electric power::Production, transmission and distribution
Final Year Project (FYP)
Nanyang Technological University